<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Maria Dunham on Medium]]></title>
        <description><![CDATA[Stories by Maria Dunham on Medium]]></description>
        <link>https://medium.com/@mariadunhum?source=rss-d83cbad5dc9e------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*NprylAegSWLL53vX6xkiEg.jpeg</url>
            <title>Stories by Maria Dunham on Medium</title>
            <link>https://medium.com/@mariadunhum?source=rss-d83cbad5dc9e------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Fri, 20 Nov 2026 04:17:41 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@mariadunhum/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[Seeing my interacted posts history on X, I read replies before likes]]></title>
            <link>https://medium.com/@mariadunhum/seeing-my-interacted-posts-history-on-x-i-read-replies-before-likes-fb6c33dffaf4?source=rss-d83cbad5dc9e------2</link>
            <guid isPermaLink="false">https://medium.com/p/fb6c33dffaf4</guid>
            <category><![CDATA[engagement]]></category>
            <category><![CDATA[replies]]></category>
            <category><![CDATA[twitter-analytics]]></category>
            <category><![CDATA[content-strategy]]></category>
            <dc:creator><![CDATA[Maria Dunham]]></dc:creator>
            <pubDate>Sat, 31 Oct 2026 16:11:01 GMT</pubDate>
            <atom:updated>2026-10-31T16:11:01.584Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*vunQT9ZB0ZaMWDcw5RB0mA.jpeg" /></figure><p>The most useful column in my post record is replies, and it is the one I left alone the longest. Everything else in the record counts a reaction. The reply column counts an answer, and an answer is a thing you can go read.</p><p>That difference took me an embarrassing amount of time to act on. I had spent two years judging an account on a column that cannot tell me a single thing about what anybody thought.</p><blockquote><strong><em>Likes:</em></strong><em> how many people approved, as a number and nothing else. </em><strong><em>Replies:</em></strong><em> how many people wrote something back, each answer still sitting there for you to read. Circleboom ranks every X post inside your reachable history by its reply count, next to the rest of the response columns, so the conversation record stops being buried under the applause record.</em></blockquote><blockquote><strong><em>→ </em></strong><a href="https://circleboom.com/twitter-management-tool/twitter-post-analytics/engagement-analytics"><strong><em>rank your interacted posts history on X</em></strong></a><em> </em>(this link takes you to Circleboom, outside Medium)</blockquote><blockquote><em>One column in that table leads somewhere you can actually read. The rest stop at a total.</em></blockquote><h3>Replies are the only column that leaves something behind</h3><p>Every other response in your record is a count with nothing attached.</p><p>A like is a number. A bookmark is a number. A repost at least puts your words somewhere else, but you still end up with a total rather than a sentence. Sort your history by any of those and you learn how much, never what.</p><p>Replies break that pattern. The count points at a post, and under that post sits a thread with people’s actual words in it. It is the only ranking in the record that ends in reading rather than in arithmetic.</p><p><strong>Sorting by replies is how you find the conversations you missed, not just the posts that did well.</strong></p><p>That is what I use it for now. Once a month I rank the accessible history by replies, take the top ten, and go read the threads under them rather than the numbers beside them.</p><p>Some of those threads are worth answering months later, which is its own small argument for the habit. If reply etiquette on X is the bit you feel unsure about, <a href="https://circleboom.com/blog/do-my-followers-see-when-i-reply-to-a-tweet/">do my followers see when I reply to a tweet</a> settles the mechanics.</p><h3>Why scrolling your own profile will not give you this</h3><p>The obvious method is to scroll back and look, and it fails for a reason most people never notice.</p><p>X’s own <a href="https://help.x.com/en/using-x/x-conversations">About conversations on X</a> page says some replies in a conversation are not shown in chronological order, because X groups and ranks them instead. <a href="https://help.x.com/en/using-x/mentions-and-replies">X replies and mentions</a> adds that a reply from somebody you do not follow never reaches your Home timeline, only your Notifications.</p><p>Both behaviors are sensible for reading X. Both are useless for assembling a record.</p><p>So the thing you would reconstruct by scrolling is a ranked, filtered, partially routed view of what happened, in an order the platform chose. It is not your reply history. It is a feed of it.</p><p>A table sorted by a reply count has no such opinion. Every post is a row, and the ordering is the one you asked for.</p><p>The scroll method has a second problem. <a href="https://circleboom.com/blog/why-twitter-doesnt-show-all-replies-and-how-to-see-them/">Why Twitter does not show all replies</a> covers the visibility side of it in detail.</p><p>The ordered version takes less time than arguing with the feed. <a href="https://circleboom.com/twitter-management-tool/twitter-post-analytics/engagement-analytics">Run the reply ranking on your X history</a> <em>(this link takes you to Circleboom, outside Medium)</em> for the last thirty days, then compare it against what you remember.</p><h3>Do more replies always mean a better post?</h3><p>No, and learning that properly cost me most of a quarter.</p><p>A reply count measures volume of response, not quality or sentiment of it. Ten thoughtful answers and ten people telling you that you are wrong produce the same cell in the table.</p><p>That still leaves the column more informative than likes, because the argument is legible and a like never is. A post with forty replies gives you forty sentences to read. A post with forty likes gives you the number forty.</p><p>Where the count genuinely misleads is comparison. A post that provoked an argument will outrank a post that started a good discussion, every time, because outrage replies faster than agreement.</p><p>So I read the count as a pointer and never as a verdict. It says which threads deserve twenty minutes of reading. What those threads contain is a separate question and the table does not pretend to answer it.</p><h3>The ratio I track instead of the raw count</h3><p>Raw reply counts move whenever your posting volume moves, which makes them useless for comparing one month against another.</p><p>So the number I keep is a ratio: replies divided by likes, across a bounded period. Both columns sit on the same row, so the arithmetic is a spreadsheet away once you have the table in front of you.</p><p>The reason it works is that both halves scale with audience size, so the ratio strips out the thing you were not asking about. A month with 4,000 likes and 200 replies and a month with 1,000 likes and 50 replies are the same month in terms of how much your posts asked of anybody.</p><p>Run it once and you have a baseline. Run it every month and you can see a content style drifting toward safety a full quarter before the follower graph notices anything.</p><p>One caveat worth repeating: this is my own arithmetic on two columns, not a metric X reports. Nothing stops you defining it differently, and the value is in tracking one definition consistently rather than in the definition itself.</p><h3>The month I mistook a reply spike for a good month</h3><p>My replies more than doubled one month and I reported it as a win.</p><p>It was a complaint thread. A pricing change had landed badly, one post about it collected most of the month’s replies, and the column moved exactly the way a genuinely popular month would have moved it.</p><p>Nothing in the record flagged the difference, because the record counts responses and does not read them.</p><p><strong>A metric that cannot tell approval from anger is still worth having, as long as you never quote it alone.</strong></p><p>What I do differently now takes about ten minutes. I rank by replies, then open the top three threads and skim them before writing a single line of the summary. If the top thread is an argument, that fact goes in the report next to the number.</p><p>The habit of pairing a count with the thing it counts is boring and it has saved me from two bad recommendations since.</p><h3>What I do when the reply column is flat</h3><p>A flat reply column with a healthy like column is one of the more precise diagnoses available in the whole record.</p><p>It usually means the posts are complete. A statement that answers itself gives a reader nothing to add, so the reader agrees, taps a like and moves on. That is not a reach problem or an audience problem, and no amount of posting more will fix it.</p><p>Three changes move the column, in rough order of how quickly they work:</p><ul><li>Leave the last step of the argument to the reader instead of closing it yourself.</li><li>Ask something you genuinely do not know the answer to, not a question you have already answered in the post.</li><li>Publish a format that requires a response, since a poll converts a passive reader into a tap.</li></ul><h3>Why the poll fix works fastest</h3><p>The poll route gives the fastest feedback. When the goal is participation rather than reach, an interactive format is the shortest path to it.</p><p>Circleboom’s <a href="https://circleboom.com/social-media-scheduler/twitter-scheduler/schedule-twitter-polls">schedule Twitter polls</a> side handles the planning if you would rather queue them than post by hand. <a href="https://circleboom.com/blog/how-to-use-twitter-polls-to-increase-twitter-engagement/">How to use Twitter polls to increase engagement</a> makes the case for why they work at all.</p><p>Circleboom is a <a href="https://docs.x.com/enterprise/customer-directory#enterprise-customers-listing">verified Enterprise partner of X</a>. That matters on this particular number. A reply count is exactly the kind of figure that drifts between tools when the data is scraped from what happens to be publicly visible, and these come from X’s own reporting on your account.</p><p>Give the changes a month. Then <a href="https://circleboom.com/twitter-management-tool/twitter-post-analytics/engagement-analytics">pull that month’s interacted posts history</a> <em>(this link takes you to Circleboom, outside Medium)</em> and check whether anything moved.</p><h3>How to see your interacted posts history on X ranked by replies</h3><p>The pass takes about five minutes.</p><h3>Open the record for the account you post from</h3><ul><li><a href="https://circleboom.com/twitter-management-tool"><strong>Log in to Circleboom Twitter</strong></a> <em>(this link takes you to Circleboom, outside Medium)</em>, authorizing whichever profile holds the threads you want to read.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*jBr_GdB5iDWeA7AM.jpg" /></figure><ul><li><strong>Pick X Post Planner, and inside it post analytics.</strong> Every post you have published is laid out as one row.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*m9K3vAvEhmnm5Kk7.png" /></figure><ul><li><strong>Set a date range</strong> covering the stretch you want to review, so the ranking describes a period rather than everything at once.</li></ul><h3>Rank by replies, then go read</h3><ul><li><strong>Sort by the replies column, highest first.</strong> The ten rows at the top become your reading list.</li><li><strong>Open each of those posts</strong> and read the thread underneath rather than the number beside it.</li><li><strong>Note which format each one was</strong> before you close the tab, because the pattern across ten posts is the actual finding.</li></ul><p>The sequence matters because the sort is only a way of choosing what to read. Treat the ranking as the answer and you end up with a leaderboard nobody asked for; treat it as a reading list and you end up knowing what your audience argues about.</p><p>Quick recap: connect, bound the period, sort by replies, read the top ten threads, write down the format each one used.</p><h3>What to do with all this next</h3><p>The most common pushback: the account does not get enough replies for any of this to be worth doing. That is usually the strongest reason to do it.</p><p>An account with six replies across a month has six sentences from real readers, and six is a small enough number that you can read all of them properly. Ranking is only necessary when there are too many to read, and a sparse reply column is not a reason to skip the record.</p><p>It is a reason to read every row of it.</p><h3>The version of the objection I have more sympathy for</h3><p>The other version is that likes are easier. They are, and they will tell you the same pleasant thing every month until you stop asking.</p><p>Start with the last sixty days and read the top five threads. If none of them contains anything you did not already know, the reply column has earned its keep by telling you that too.</p><p>Reply-guy culture is a real part of that read, and <a href="https://circleboom.com/blog/are-twitter-reply-guys-dead-now/">are Twitter reply guys dead now</a> is a fair sketch of who actually answers posts on X these days.</p><p>If drafting the responses is the bottleneck rather than finding them, the <a href="https://circleboom.com/twitter-management-tool/x-post-planner/ai-reply-generator">AI reply generator</a> drafts a starting point you edit before anything goes out.</p><p><strong>Watch it run: the practical moves that lift both impressions and replies on an X account.</strong></p><p><a href="https://www.youtube.com/watch?v=GzGW70aSZqE">https://www.youtube.com/watch?v=GzGW70aSZqE</a></p><p><strong>→ Begin with one month: </strong><a href="https://circleboom.com/twitter-management-tool/twitter-post-analytics/engagement-analytics"><strong>open your interacted posts history on X</strong></a> <em>(this link takes you to Circleboom, outside Medium)</em></p><h3>What to know once the reply column is in front of you</h3><h3>Do my own replies to other people show up in this record?</h3><p>The table covers the posts published from the connected account, and replies you write are posts, so they sit in the same dataset and can be narrowed with the post type filter. What the reply column on any given row counts is the responses that post received, not the responses you sent.</p><h3>How much of my reply history can I actually reach?</h3><p>As far back as the accessible window goes, which is the 3,200 most recent posts on standard account access. Filters narrow what has already loaded rather than fetching more, so a date range inside that window works and one outside it returns nothing.</p><p>Busy accounts meet that boundary sooner than they expect, which is the argument for reading the conversation record while the conversations are still recent. For how X stitches those threads together in the first place, <a href="https://circleboom.com/blog/everything-you-need-to-know-about-improved-twitter-conversations/">improved Twitter conversations</a> covers the mechanics.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=fb6c33dffaf4" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[How I filtered adult content out of my Twitter feed and kept it out]]></title>
            <link>https://medium.com/@mariadunhum/how-i-filtered-adult-content-out-of-my-twitter-feed-and-kept-it-out-4ea12dfc11ed?source=rss-d83cbad5dc9e------2</link>
            <guid isPermaLink="false">https://medium.com/p/4ea12dfc11ed</guid>
            <category><![CDATA[feed-cleanup]]></category>
            <category><![CDATA[digital-wellbeing]]></category>
            <category><![CDATA[twitter]]></category>
            <category><![CDATA[content-filtering]]></category>
            <dc:creator><![CDATA[Maria Dunham]]></dc:creator>
            <pubDate>Fri, 30 Oct 2026 13:06:01 GMT</pubDate>
            <atom:updated>2026-10-30T13:06:01.555Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*ZRRG_uf0fnI9z2qbV8yt8Q.png" /></figure><p>I fixed my feed three times before it stayed fixed.</p><p>The first two fixes lasted about a week each.</p><p>The reason is boring and it took me embarrassingly long to see: I kept changing what X displayed to me, and never once changed what I had asked it to display.</p><blockquote><strong><em>What I did the first two times:</em></strong><em> toggled sensitive media settings, muted six words, felt better for a week. </em><strong><em>What worked the third time:</em></strong><em> audited 1,100 accounts I follow, unfollowed 260 of them in paced batches, and the feed changed permanently. Circleboom reads the following list through official X API access and scores each account on current behaviour rather than on why I followed it.</em></blockquote><blockquote><strong><em>→ </em></strong><a href="https://circleboom.com/twitter-management-tool/twitter-search-tool/twitter-follower-audit/twitter-unfollow-tool/mass-unfollow"><strong><em>audit the accounts feeding your Twitter feed</em></strong></a><em> </em>(this link takes you to Circleboom, outside Medium)</blockquote><h3>The two fixes that did not hold</h3><p>The first thing I did was the obvious thing: I went into settings and turned on every control X offers for sensitive media.</p><p>That took four minutes and it genuinely helped, because it put warnings in front of images that had been appearing uncovered.</p><p>It did not reduce how much of that content was in my feed. It changed how it was displayed.</p><p>Those are not the same problem and I had solved the wrong one.</p><p>So the second time I muted words.</p><p>Six terms that seemed to recur, applied to timeline and notifications both.</p><p>Better again, and then worse again within about ten days, for two reasons I should have anticipated.</p><p>Muted words match text, and the content here is mostly visual.</p><p>An image with a three-word caption does not match anything.</p><p>And muting filters what reaches me without touching what is generating it, so the accounts kept posting, kept counting in my following total, and kept telling the recommendation system what I wanted.</p><p><strong>I spent two rounds filtering the output of a machine I could have just turned down.</strong></p><h3>Where it was actually coming from</h3><p>The third time I stopped guessing and looked at my following list, which I had not properly examined in six years.</p><p>Eleven hundred accounts. I could account for maybe four hundred of them.</p><p>The rest were follow-backs from a growth phase I ran years ago, accounts I followed for one post and forgot, and a category I had not thought about at all.</p><p>Accounts that were completely fine when I followed them and are not fine now.</p><p>That last group turned out to be the largest single source.</p><p>Handles I had followed years earlier because they posted good work, since abandoned, recovered by somebody else, and now posting something entirely different.</p><p>Nothing notifies you when that happens.</p><p>The follow relationship carries none of the history.</p><p>Around sixty of them were posting adult promotion under handles I had once followed for photography, music and football.</p><p><strong>A following list is a record of decisions I made once. My feed reflects what those accounts are doing today.</strong></p><p>That gap is where the whole problem lived.</p><p><a href="https://circleboom.com/blog/how-to-remove-people-from-your-twitter-timeline/">Removing people from your Twitter timeline</a> covers the mechanics of closing it.</p><h3>Why the accounts matter more than the settings</h3><p>Volume is the part people underestimate.</p><p>An account posting forty times a day occupies far more of a timeline than its single line in a following count suggests.</p><p>Ten such accounts can dominate a feed built from four hundred follows.</p><p>Which means a timeline can become unrecognisable without the following list changing much at all, purely because a handful of accounts changed their posting rate.</p><p>There is also a feedback effect.</p><p>Recommendation systems read what you follow and what you linger on as signal, so a following list carrying a chunk of adult promotion is teaching the For You tab what to bring you from accounts you do not follow.</p><p>Filtering the output does not correct the input.</p><p>And there is the public dimension.</p><p>My following list is visible to anyone who opens it.</p><p>And a list full of adult spam accounts reads as a judgment about me regardless of how those accounts got there.</p><p>That was not why I cleaned it up, and it was a reasonable second reason.</p><p><strong>Short demo: cutting a timeline down by unfollowing the accounts posting at the highest volume.</strong></p><p><a href="https://www.youtube.com/watch?v=-YWvGUWQg3A">https://www.youtube.com/watch?v=-YWvGUWQg3A</a></p><h3>The audit that worked</h3><p>Three phases. The middle one took about forty minutes and it is the whole job.</p><h3>Pull the following list</h3><ul><li><strong>Open </strong><a href="https://circleboom.com/twitter-management-tool"><strong>Circleboom Twitter</strong></a> <em>(this link takes you to Circleboom, outside Medium)</em> and connect the account whose feed you are fixing.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*LjEHVvTrZP0mQoli.png" /></figure><ul><li><strong>Go to the Follower and Following menu</strong> and open Fake/Bot Following, which applies the same multi-signal scoring to the accounts you follow rather than to the ones following you.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*-akMkAz9CwbB2-zK.png" /></figure><ul><li><strong>Let the full retrieval finish.</strong> On eleven hundred follows it took a few minutes and the list kept shifting while it loaded, which had me drawing conclusions from a partial view the first time I tried this.</li></ul><h3>Separate the categories before acting on any of them</h3><ul><li><strong>Whitelist the accounts you follow deliberately first</strong>, including parody, brand and support accounts that look odd but are followed on purpose. The label persists across every feature and every future session.</li><li><strong>Stack Fake/Spam with Eggheads, a Tweet Count range and a Follow Ratio ceiling</strong>, which pulls out accounts that were always synthetic.</li><li><strong>Then run the Overactive filter separately.</strong> This is the one that catches drift, because it finds real accounts posting at a volume that has taken over your timeline, and they score fine on every other signal.</li><li><strong>Read a full page of each segment by hand.</strong> Twenty minutes, and it is the only step that finds a filter set too wide.</li></ul><h3>Unfollow in paced batches</h3><ul><li><strong>Select and click Unfollow.</strong> Circleboom processes 50 unfollow actions every 15 minutes with a daily ceiling of 800, and pauses automatically when a quota is reached.</li><li><strong>Keep each session to 100 or 150 accounts.</strong> That is the documented recommendation, and unfollowing aggressively is what X’s automated systems treat as suspicious behaviour.</li></ul><p>I ran mine across four sessions over three days rather than one large batch, partly for safety and partly because splitting it let me judge the feed between runs.</p><p><a href="https://circleboom.com/blog/how-to-unfollow-fake-twitter-accounts-in-3-steps/">Unfollowing fake Twitter accounts in three steps</a> is the compressed version if your list is smaller.</p><h3>What the pacing is actually protecting</h3><p>The 800-a-day ceiling is not the number to work to, and I want to be clear about why.</p><p>Unfollow limits vary between accounts on X.</p><p>If the platform returns a rate-limit response, the request is simply not fulfilled, and repeated fast requests are read as aggressive behaviour rather than as enthusiasm.</p><p>Accounts get temporarily locked for exactly this.</p><p>So Circleboom pauses instead of pushing, and the recommended working pace sits well under the ceiling at 100 to 150 per session.</p><p>On a 260-account cleanup that means several sittings rather than one, which felt slow until I considered what the alternative failure looks like.</p><p>Circleboom runs the whole thing as an <a href="https://docs.x.com/enterprise/customer-directory#enterprise-customers-listing">official X Enterprise Developer company</a>, inside X’s documented limits rather than around them.</p><p>On an operation touching hundreds of relationships in a row, that is the difference between a clean cleanup and a locked account.</p><p>If you would rather not unfollow at all, <a href="https://circleboom.com/twitter-management-tool/twitter-circle-tool/low-quality-followers-and-friends-on-twitter">finding your low quality followers and friends</a> gives you the same read without committing to any action.</p><p>Muting individual accounts is the other soft option, leaving the follow in place while removing them from your timeline.</p><p><a href="https://circleboom.com/blog/how-to-mute-someone-on-twitter-x/">How to mute someone on X</a> covers that route.</p><h3>The three categories I found, and what each one needed</h3><p>Sorting the 1,100 into groups made the decisions obvious in a way that scanning a flat list never did.</p><p><strong>Always synthetic.</strong> Accounts that were automated from the day I followed them, mostly picked up during the follow-back phase.</p><p>Every behavioural signal fires at once here: no photo, near-zero tweets, absurd follow ratio.</p><p>Around 90 accounts, and the filter did all the work.</p><p><strong>Degraded.</strong> Accounts that were real and are not any more.</p><p>Abandoned, recovered, repurposed, now posting promotion under a handle I trusted.</p><p>This group scores oddly, because the account age is old and the tweet count is high, which are both signals that usually mean legitimate.</p><p>Around 60 accounts, and only a manual read found them.</p><p><strong>Overactive but genuine.</strong> Real people and real publications posting at forty or fifty times a day.</p><p>Nothing wrong with any of them, and collectively they were eating my timeline.</p><p>Around 110 accounts, and this is the group where unfollowing felt least comfortable and helped the most.</p><p>The third group is the one most cleanups miss entirely, because no fake-account filter will ever flag a real account behaving normally at high volume.</p><p>The Overactive view exists precisely for it, and running it separately from the spam filters is what made the categories legible.</p><p><strong>Two of those three groups are invisible to a bot filter. Both of them were shaping my feed.</strong></p><h3>What changed, and what did not</h3><p>The feed changed within a day.</p><p>Not gradually, which surprised me, though it should not have: I removed the source of roughly a quarter of what was being posted into it.</p><p>The For You tab took about two weeks to follow.</p><p>That one is a recommendation system relearning what I want from a following list that now says something different, and it moved slower than the chronological timeline because it had more inertia to work against.</p><p>What did not change was content reaching me from accounts I do not follow.</p><p>Recommendation feeds pull from the whole platform, so trending and reply threads still bring things through, and no amount of unfollowing closes that door.</p><p>Settings and muted words are the right tools there, which is why I keep all three running rather than treating the audit as a replacement.</p><p>The thing I did not expect was how much quieter the notification tab got.</p><p>A lot of what I had been unfollowing was also interacting, and removing the follow reduced the interactions as a side effect.</p><p><strong>The audit fixes what you asked for. Settings still handle what you did not.</strong></p><p>If you want the same result, the practical starting point is to <a href="https://circleboom.com/twitter-management-tool/twitter-search-tool/twitter-follower-audit/twitter-unfollow-tool/mass-unfollow">pull your following list and sort it</a> before deciding anything.</p><h3>Ninety days later</h3><p>I re-ran the audit at the one-month mark and found eleven accounts worth removing.</p><p>At three months it was around thirty.</p><p>That rate is the honest steady state on my account.</p><p>Composition drifts continuously rather than in events, and about ten accounts a month either degrade or turn out to have been wrong choices.</p><p>Thirty minutes a quarter keeps it flat, and <a href="https://circleboom.com/twitter-management-tool/twitter-search-tool/twitter-follower-audit/twitter-unfollow-tool/mass-unfollow">running the unfollow pass in paced batches</a> is most of that thirty minutes.</p><p>What made subsequent runs cheap is the whitelist.</p><p>Everything I protected in run one stayed protected, so the review set in run four contained only accounts I had never ruled on, which on a stable following list is a small number.</p><p>If your feed has been drifting for years rather than months, expect the first pass to be the expensive one and every pass after it to be maintenance.</p><p><a href="https://circleboom.com/blog/is-there-a-way-to-reset-your-twitter-algorithm-or-do-you-need-a-new-account/">Resetting your Twitter algorithm</a> is worth reading if you are tempted to start over instead.</p><p>The short version is that a new account inherits the same problem within months.</p><p><strong>→ </strong><a href="https://circleboom.com/twitter-management-tool/twitter-search-tool/twitter-follower-audit/twitter-unfollow-tool/mass-unfollow"><strong>start the following audit</strong></a> <em>(link goes to Circleboom, outside Medium)</em></p><h3>Questions I get about feed cleanup</h3><h3>Does unfollowing tell the account?</h3><p>No.</p><p>Unfollowing is silent and generates no notification, which is one reason it is a lower-friction action than blocking.</p><h3>How long did the whole thing take?</h3><p>About forty minutes of filtering and reviewing, then four unfollow sessions across three days.</p><p>The reviewing was the slow part and it stayed the slow part on every run since.</p><h3>Did your following count matter to you?</h3><p>Not really, and it dropped by about a quarter.</p><p>The <a href="https://circleboom.com/twitter-management-tool/twitter-search-tool/twitter-follower-audit/twitter-unfollow-tool">Twitter unfollow tool</a> shows you exactly what you are removing before you commit, so the number was never a surprise.</p><h3>Can you do this without unfollowing?</h3><p>Partly.</p><p>Muting individual accounts removes them from your timeline while leaving the follow in place, which is useful when you want to keep the connection for other reasons.</p><h3>Why not just make a new account?</h3><p>Because the problem follows you.</p><p>Recommendation is driven by what you follow and engage with, and a fresh account reaches the same state in a few months unless the underlying habit changes.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=4ea12dfc11ed" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The Twitter follower language stats chart everyone skips is the reliable one]]></title>
            <link>https://medium.com/@mariadunhum/the-twitter-follower-language-stats-chart-everyone-skips-is-the-reliable-one-2786b0c88619?source=rss-d83cbad5dc9e------2</link>
            <guid isPermaLink="false">https://medium.com/p/2786b0c88619</guid>
            <category><![CDATA[audience-insights]]></category>
            <category><![CDATA[data-analysis]]></category>
            <category><![CDATA[language]]></category>
            <category><![CDATA[twitter-analytics]]></category>
            <dc:creator><![CDATA[Maria Dunham]]></dc:creator>
            <pubDate>Thu, 29 Oct 2026 13:06:01 GMT</pubDate>
            <atom:updated>2026-10-29T13:06:01.531Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*11HzuTC1FDyOBPgn6dARBQ.jpeg" /></figure><p>Most people open Language Stats, glance past the plain two-slice chart on the left, and go straight to the colorful breakdown on the right with the named languages on it. I did the same thing for months. The chart with more detail felt like the chart with more information.</p><p>It is the other way around, and understanding why changed how I read every inferred audience metric I work with.</p><blockquote><em>Twitter follower language stats are an estimate, not a lookup. Circleboom classifies each follower from public profile signals and shows two views of the result: a simple English against non-English split, and a full per-language breakdown. The simple one is asking less of the evidence, so it holds up better.</em></blockquote><blockquote><strong><em>→ </em></strong><a href="https://circleboom.com/twitter-management-tool/twitter-user-analytics/twitter-language-stats"><strong><em>Twitter language stats for your followers</em></strong></a><em> </em>(this link takes you to Circleboom, outside Medium)</blockquote><blockquote><em>The detailed chart is still worth reading. It just deserves a different level of trust.</em></blockquote><h3>The chart I used to skip</h3><p>The left donut answers one question: is this account English-speaking or not.</p><p>The right donut answers a much harder one: which of a dozen or more specific languages is this account’s dominant one. Same follower, same evidence, two very different demands placed on it.</p><p>Think about what that evidence is. An account with a four-word bio, a Latin-script display name, and eleven posts is thin material. There may be enough there to say “this person is probably not posting in English.” There is often not enough to choose confidently between Portuguese and Spanish, or between Indonesian and Malay.</p><p>Both charts will still print a number. Only one of them was asked a question the evidence could answer.</p><p><strong>Precision in a chart is a claim about the chart, not about the data underneath it.</strong></p><p>That is why the left donut earns more of my confidence, and why I now read the two in a fixed order.</p><p>The split tells me whether there is a story here at all. The breakdown tells me where to go looking for it.</p><p>The stakes are practical. Deciding to publish in a second language is a standing commitment, not a one-off post. You are signing up to write, edit, and moderate in a language on an ongoing basis.</p><p>Getting talked into that by a 9% slice that was really a classifier hedging between two similar languages is an expensive way to learn this lesson.</p><p>Start with <a href="https://circleboom.com/twitter-management-tool/twitter-user-analytics/twitter-language-stats">the language split of your X followers</a> <em>(this link takes you to Circleboom, outside Medium)</em>, and read the left chart before the right one.</p><h3>What I got wrong about the two donuts</h3><p>I assumed the two charts were the same data at two zoom levels, and that the left one was just the right one rolled up.</p><p>Roughly, that holds. Both charts read the same accounts off the same evidence, and the left one simply asks less of it. Where I went wrong was inferring that the same evidence means the same reliability, and it does not.</p><p>A coarse reading can be right while every fine-grained component of it is shaky. If the classifier is confident an account is not English but is splitting hairs about which language it is, that uncertainty disappears the moment the only question is English or not.</p><p>The coarse split survives errors that cancel each other out. The individual slices carry every one of them.</p><p>So I had been treating the long tail of the detailed chart, the 4% and 3% slices, with the same confidence as the headline number. Those small slices are exactly where a classifier’s near-misses accumulate.</p><p>I now read anything under about 5% as “something is here” rather than as a measurement.</p><h3>How to pull Twitter follower language stats and cross-check them</h3><p>Circleboom reads your follower list and your following list through official X API access, classifies each account’s dominant language from its public profile, and renders both distributions as donut chart pairs. The workflow below takes about five minutes. The cross-check in the last phase is what turns it from a screenshot into a reading you can act on.</p><h3>Open the report and read the follower side</h3><ul><li><a href="https://circleboom.com/twitter-management-tool"><strong>Open Circleboom Twitter</strong></a> <em>(this link takes you to Circleboom, outside Medium)</em>, then grant access to the X account whose audience you are reading.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*MtfZFmv1eBcctRFF.png" /></figure><ul><li><strong>Head for Follower / Following Management and Analytics</strong>, home to every audience-composition screen.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*r9PtASq3Oo0Ghiac.png" /></figure><ul><li><strong>Select Language Stats.</strong> It lands on Follower Language Stats first, under the title “Language Distribution Among Followers.”</li><li><strong>Read the left donut and write the non-English share down.</strong> That single number is your headline finding.</li><li><strong>Read the right donut second</strong>, and treat any slice under roughly 5% as a lead rather than a fact.</li></ul><h3>Cross-check against the accounts you follow</h3><ul><li><strong>Move across to Following Language Stats</strong>, titled “Language Distribution Among Following.” Same classifier, running on accounts you picked yourself.</li><li><strong>Compare it against what you know.</strong> You picked those accounts, so you can tell whether the classifier got them right in a way you never can with ten thousand strangers.</li><li><strong>Export each tab on its own.</strong> The three-dot control above the chart offers PNG, JPG and PDF, and it captures whichever tab is active, so repeat it for the other one.</li></ul><p>That last phase is the part most people never run, and it is the only cheap way to see the classifier being right or wrong on accounts whose real languages you already know. A reading you cannot audit is a reading you are trusting on faith.</p><p><strong>On screen: how the Following Language Stats tab renders the same charts for the accounts you follow rather than your audience.</strong></p><p><a href="https://www.youtube.com/watch?v=DI096oDeAcY">https://www.youtube.com/watch?v=DI096oDeAcY</a></p><h3>Using the accounts you follow on X as a control group</h3><p>Your following list is small, chosen, and known. That makes it the best test set you will ever have for free.</p><p>Pick fifteen accounts you follow whose languages you are certain about. Note what you expect, then read what the Following tab reports. You are not looking for perfection here. You are looking for the shape of the error.</p><h3>The three errors worth looking for</h3><p>Three patterns account for most of what goes wrong:</p><ul><li>A bilingual account you know posts mostly in one language being filed under the other.</li><li>An account with a sparse profile landing in Other rather than in its real language.</li><li>A named language appearing that you cannot account for anywhere in your list.</li></ul><p>A narrower version of this check is sharper still. Your <a href="https://circleboom.com/twitter-management-tool/twitter-circle-tool/twitter-mutuals">Twitter mutuals</a> sit on both lists at once, so they appear in both charts.</p><p>Any account the classifier files differently across the two tabs is a visible error you can point at.</p><p>Whatever you find on your following list is roughly what is happening on your follower list, at scale and out of sight.</p><p>If the classifier files three of your fifteen wrong, you have learned that the small slices on the follower side deserve real skepticism. If it files fifteen out of fifteen correctly, the breakdown has earned a lot more of your trust.</p><p>The two tabs exist so the audience and the network can be compared, and the comparison most people run is the strategic one: are my sources in the same languages as my readers. That is a good question.</p><p>The calibration read is the one I would run first, because it tells you how much to believe the answer to the strategic one.</p><p>The same logic applies to the other audience cuts sitting in that menu. Reading <a href="https://circleboom.com/blog/twitter-friends-demographics/">Twitter friends demographics</a> alongside your follower demographics gives you a known-answer check on inferences you would otherwise take at face value.</p><p>Circleboom is <a href="https://docs.x.com/enterprise/customer-directory#enterprise-customers-listing">listed on X’s Enterprise customer directory</a>, which is why the follower and following lists behind these charts are complete rather than sampled. That matters here specifically: a control group only works if the list you are checking against is the real list.</p><h3>What does it mean when your two tabs disagree?</h3><p>Sometimes the follower tab and the following tab tell stories that do not fit together, and the disagreement is informative.</p><p>An account whose followers read as 30% non-English while its following list reads as 95% English is a common shape, and it usually means the same thing. You built your reading habits in one language and your audience arrived from somewhere else. That gap is the actual finding, and no single chart shows it.</p><p>The reverse shape, a heavily multilingual following list against an overwhelmingly English audience, tends to mean your inputs are broader than your output. You are reading widely and publishing narrowly.</p><p>Neither is a problem on its own. Both are worth knowing before you decide what to write next.</p><h3>Place and timing sharpen a language slice</h3><p>Language is one axis of a picture with several, and the others sharpen it.</p><p>Where your audience sits geographically changes what a language slice means in practice. A Spanish segment concentrated in one city is a local audience you can write to directly; the same percentage spread across four continents is a diaspora, and it wants different content. <a href="https://circleboom.com/blog/the-benefits-of-using-twitter-geolocation-maps-for-your-business/">Twitter geolocation maps for your business</a> put that second reading beside this one.</p><p>Timing is the third axis, and it is the one people skip. A language segment that never sees your posts because you publish while it sleeps looks identical in your engagement data to a segment that saw them and did not care. Reading the <a href="https://circleboom.com/blog/what-are-the-most-engaged-time-zones-on-twitter/">most engaged time zones on Twitter</a> separates those two very different problems.</p><p>Launch the <a href="https://circleboom.com/twitter-management-tool/twitter-user-analytics/twitter-language-stats">Follower and Following language stats</a> <em>(this link takes you to Circleboom, outside Medium)</em> next to the rest of the <a href="https://circleboom.com/twitter-management-tool/twitter-user-analytics">Twitter user analytics</a> screens.</p><p>Read them as one picture rather than a stack of separate screenshots.</p><h3>Reading the two charts in the right order</h3><p>Language Stats gives you two charts and most people read one of them, in the wrong order, with the wrong level of confidence.</p><p>The order that works: the simple English against non-English split first, because it asks the least of the evidence and therefore holds up best. Then the detailed breakdown, treated as a set of leads. Then the Following tab, where you can check the classifier against accounts whose languages you already know.</p><p>That third step costs five minutes, and it separates a number you happen to have from a number you can defend.</p><p>Then act on the smallest thing the reading supports. Publishing once in the language you are unsure about will tell you more than another month of staring at donut charts.</p><p>Once you know which segment you are writing for, <a href="https://circleboom.com/blog/how-to-check-interests-on-twitter/">how to check interests on Twitter</a> tells you what to say to it, which the language chart never will.</p><p><strong>→ Check </strong><a href="https://circleboom.com/twitter-management-tool/twitter-user-analytics/twitter-language-stats"><strong>your X follower language stats</strong></a> <em>(this link takes you to Circleboom, outside Medium)</em></p><h3>What people ask after their first read</h3><h3>Can I trust a language slice that is only 3%?</h3><p>Treat it as a signal that something is there rather than as a measurement of how much. Small slices are where a classifier’s near-misses collect, because an account it could not confidently place still has to go somewhere, so read anything under roughly 5% as a prompt to look rather than a number to plan against.</p><h3>Why does my following list look so different from my follower list?</h3><p>Because you chose one and the other chose you. A gap between the two usually means your reading habits formed in one language while your audience arrived from elsewhere, which is worth knowing before you decide whether to change what you publish or change what you read.</p><h3>Does Language Stats let me unfollow or message a language segment?</h3><p>No. The screen reports and does not act, so switching tabs and exporting are the only two things available on it. Acting on a language segment happens through the follower list filters elsewhere in the product.</p><h3>How often should I re-run this?</h3><p>After anything that reshapes your audience quickly: a push into a new language community, a partnership with an account in another market, or a sudden spike in followers. Language proportions shift as composition shifts, so a reading taken before a growth event stops describing you fairly quickly.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=2786b0c88619" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Can you see an X following list without logging in?]]></title>
            <link>https://medium.com/@mariadunhum/can-you-see-an-x-following-list-without-logging-in-dcd1eeb4f87f?source=rss-d83cbad5dc9e------2</link>
            <guid isPermaLink="false">https://medium.com/p/dcd1eeb4f87f</guid>
            <category><![CDATA[privacy]]></category>
            <category><![CDATA[twitter]]></category>
            <category><![CDATA[following-list]]></category>
            <dc:creator><![CDATA[Maria Dunham]]></dc:creator>
            <pubDate>Wed, 28 Oct 2026 19:11:01 GMT</pubDate>
            <atom:updated>2026-10-28T19:11:01.815Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*qQjMZ3etm5iL2xXNJ9vFqQ.jpeg" /></figure><p>I used to treat a visible following tab as evidence that I could inspect the whole list. The mistake became obvious when the page stopped loading while the profile count still showed hundreds of accounts I had not seen.</p><blockquote><strong><em>What this access check tells you.</em></strong></blockquote><ul><li>A public profile may reveal a few accounts before X asks for a session or stops loading.</li><li>An authenticated Circleboom export can retrieve a public account’s following list as a structured CSV through <a href="https://docs.x.com/enterprise/customer-directory#enterprise-customers-listing">official X Enterprise APIs</a>.</li><li>Protected accounts remain outside that export boundary.</li></ul><blockquote><em>Circleboom uses that authenticated connection for the export. Start by choosing to </em><a href="https://circleboom.com/twitter-management-tool/twitter-search-tool/twitter-list-manager/export-twitter-following-list"><em>turn a public following list into a CSV</em></a><em>.</em></blockquote><p>That failure taught me to ask two separate questions. Can I look at part of the list, and can I obtain a working dataset I can verify?</p><h3>The browser result that looked complete</h3><p>The profile was public. Its following count was visible, and several familiar names appeared when I opened the tab. Nothing on the screen announced that I was looking at a sample.</p><p>Then scrolling stopped. Refreshing changed the result, and another browser produced a login prompt sooner. The list was visible, but visibility had been mistaken for completeness.</p><p><strong>A browser preview is an interface state, not a dataset.</strong></p><p>That is the practical answer to <strong>can you see following list on X without logging in</strong>. You might see some of it. You should not build a research method around getting all of it.</p><p>The distinction matters most when the task sounds simple: identify who an industry account follows, compare two networks, or save a list before it changes. Each task needs stable rows. A logged-out page gives no row count, retrieval record, or assurance that loading finished.</p><h3>What I now mean by “access”</h3><p>Access has layers. I separate them before choosing a method:</p><ul><li><strong>profile access:</strong> can I see the public account and its headline counts?</li><li><strong>preview access:</strong> can I open a following tab and inspect some visible accounts?</li><li><strong>session access:</strong> will X let me continue after I authenticate?</li><li><strong>data access:</strong> can I obtain structured rows for a permitted public list?</li><li><strong>use access:</strong> am I handling that file for a legitimate, limited purpose?</li></ul><p>The first two can sometimes happen without logging in. Export does not. It requires an authenticated connection, and privacy still controls which target accounts are eligible.</p><p>This model stopped me from asking whether a workaround could keep scrolling. The better question was which layer the job genuinely required.</p><p>For readers who want to understand the native view first, the guide to <a href="https://circleboom.com/blog/how-can-i-view-who-im-following-on-twitter-x/">viewing who you follow on X</a> explains the account-owner perspective.</p><h3>Public is not the same as anonymously unlimited</h3><p>A public account makes profile information available under X’s rules. It does not guarantee that X will render every following relationship to a logged-out browser indefinitely.</p><p>X can place login walls, limit pagination, or change its unauthenticated experience. Those controls can vary without the account becoming protected. A test that works today can fail in a later session.</p><p>Protected accounts have a firmer boundary. X’s explanation of <a href="https://help.x.com/en/safety-and-security/public-and-protected-posts">public and protected posts</a> describes the privacy setting, and Circleboom does not export protected accounts’ follower or following lists.</p><p>Logging in as myself establishes my identity. It does not grant me someone else’s protected relationships.</p><p>The article asking whether <a href="https://circleboom.com/blog/does-a-protected-twitter-x-account-hide-your-following-list/">a protected X account hides its following list</a> goes deeper into that distinction.</p><h3>The moment I stopped copying names by hand</h3><p>Manual copying looked harmless when the visible set was small. It became unreliable as soon as I needed repeatable work.</p><p>I could not tell whether I had skipped a name, copied the same account twice, or reached the end.</p><p>I also had only usernames. The attributes that would make the list useful, such as account age, location, bio, follower count, and verification state, required more profile visits.</p><p>The costs accumulated in quiet ways:</p><ul><li>missing rows changed the apparent shape of the network;</li><li>duplicate names inflated category counts;</li><li>profile changes made the notes stale;</li><li>manual formatting consumed more time than analysis;</li><li>no retrieval date made later comparisons ambiguous.</li></ul><p><strong>The work was not difficult because the list was hidden. It was difficult because the visible form was not portable.</strong></p><p>That was the point at which export became the correct task. The guide to <a href="https://circleboom.com/blog/export-twitter-following-followers-lists/">exporting Twitter followers and following lists</a> makes the same separation between an on-screen relationship and a usable file.</p><h3>My access-versus-export workflow</h3><p>I now begin with the least expensive question: will a quick preview answer the problem? If yes, I stop. If I need structured evidence, I use an authenticated export and document the result.</p><h3>Authenticate the account making the request</h3><ol><li><a href="https://circleboom.com/twitter-management-tool">Open Circleboom Twitter Management</a> and connect the X account you are authorized to use.</li><li>Check the connected username before searching for the target account.</li></ol><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*I27rUDzML8mm56DG.png" /></figure><p>Circleboom appears in X’s <a href="https://docs.x.com/enterprise/customer-directory#enterprise-customers-listing">Enterprise customer directory</a>. The connection uses supported X APIs, so I do not need to provide a target account’s password or hand browser cookies to an unknown script.</p><h3>Find the export feature</h3><ol><li>Open Essential Toolbox in the Circleboom dashboard.</li><li>Select Export Followers or Following.</li></ol><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*YfUCIPLiqGY_mmUL.png" /></figure><p>The menu contains the export operation rather than a browser imitation. That difference matters: the expected result is a file, not an endlessly scrolling page.</p><h3>Confirm the public target</h3><ol><li>Enter the public account’s X username.</li><li>Inspect the profile preview, including its display name, bio, age, post count, and relationship counts.</li><li>Stop if the account is protected or the preview identifies the wrong profile.</li></ol><p>Similar usernames make this confirmation step valuable. Export tokens are consumed according to the operation, so I verify the target before proceeding.</p><h3>Select following and export</h3><ol><li>Choose <strong>Export Following</strong>, not Export Followers.</li><li>Review the token balance and confirm the operation.</li><li>Download the CSV and add the retrieval date to its filename.</li></ol><p>“Followers” are the accounts that follow the target. “Following” are the accounts the target chose to follow. Mixing the two produces a perfectly formatted answer to the wrong question.</p><p><strong>→ </strong><a href="https://circleboom.com/twitter-management-tool/twitter-search-tool/twitter-list-manager/export-twitter-following-list"><strong>Export the correct public following list</strong></a></p><p><strong>See the handoff from access to analysis: this walkthrough shows the public-account export path and the resulting file.</strong></p><p><a href="https://www.youtube.com/watch?v=aHwKj1qtwr8">https://www.youtube.com/watch?v=aHwKj1qtwr8</a></p><h3>What changed when the list became a CSV</h3><p>The first improvement was not speed. It was confidence about what I had.</p><p>Each retrieved account became a row with fields that could include:</p><ul><li>profile ID, username, name, location, and bio;</li><li>tweet, following, and follower counts;</li><li>protected and verification indicators;</li><li>account creation timestamp;</li><li>quality flags for inactive, fake, egghead, or overactive patterns.</li></ul><p>I could sort, filter, deduplicate, and annotate without reopening every profile. A teammate could reproduce the same categories from the same snapshot.</p><p>The CSV also made uncertainty visible. For a very large target account, API pagination or retrieval windows may produce fewer rows than the profile’s displayed following count.</p><p>That caveat belongs in the analysis rather than being hidden behind the word “complete.”</p><p><strong>A structured file improves the audit trail, but it does not excuse an unverified completeness claim.</strong></p><h3>The row-count check I added</h3><p>I compare the profile metric with the exported row count whenever completeness affects the conclusion. The check takes little time and prevents an oversized claim.</p><p>My note includes:</p><ol><li>target username and public status;</li><li>displayed following count before retrieval;</li><li>number of data rows in the downloaded CSV;</li><li>date, time, and known processing limitation;</li><li>intended use and storage location.</li></ol><p>Relationship changes can occur during or after retrieval. A difference of a few rows may reflect live follow activity, while a large difference may point to a retrieval limit that deserves disclosure.</p><p>This habit changed the language in my reports. I write “the export returned 8,412 rows on September 16” rather than “this is everyone the account follows” unless I have evidence for the broader statement.</p><h3>What I do with the file, and what I avoid</h3><p>A following list can support legitimate network research. It can also reveal patterns that deserve careful handling even when the profiles are public.</p><p>I use exports to:</p><ul><li>map public sources followed by a credible industry account;</li><li>compare network composition across two dated snapshots;</li><li>find possible speakers, partners, or publications for human review;</li><li>prepare a deduplicated set before list organization;</li><li>document my own following list before making changes.</li></ul><p>I do not publish the raw file by default, infer sensitive traits from follows, or treat a quality flag as a final judgment about a person. Data reduction is part of responsible work: keep only the fields the purpose requires.</p><p>When I need the opposite relationship, I switch to the page for <a href="https://circleboom.com/twitter-management-tool/twitter-search-tool/twitter-list-manager/export-list-of-followers-on-twitter">exporting a list of followers</a>. The similar controls should not blur the different meanings.</p><h3>Export and X lists solve different problems</h3><p>An export freezes relationship data for external analysis. An X List organizes accounts for ongoing reading inside X. Neither replaces the other.</p><p>After reviewing an export, I may use Circleboom’s <a href="https://circleboom.com/twitter-management-tool/twitter-search-tool/twitter-list-manager">X List Manager</a> to organize selected accounts into a focused public or private list. A private list can keep the monitoring set out of public view, subject to X’s list behavior.</p><p>That does not hide my main following list, and it does not unlock protected accounts. The article on <a href="https://circleboom.com/blog/is-there-a-way-to-prevent-people-from-seeing-your-following-list-on-x/">preventing people from seeing a following list</a> explains why list privacy and follow privacy should not be confused.</p><p><strong>Export answers “what was in this public relationship set?” A List answers “which accounts do I want to monitor together?”</strong></p><p>Keeping those jobs separate makes the workflow easier to maintain.</p><h3>When the browser preview is still enough</h3><p>Export is not automatically better. It is unnecessary when I need to answer a tiny, immediate question.</p><p>A logged-out or native preview may be enough when I want to:</p><ul><li>confirm that a public organization follows a named partner;</li><li>recognize a few visible accounts;</li><li>inspect a profile before deciding whether deeper research is justified;</li><li>avoid storing data that has no continuing purpose.</li></ul><p>The decision rule is proportionality. If three visible profiles settle the question, collecting thousands of rows adds risk and work without adding value.</p><p>If the task requires sorting, filtering, repeatable counts, or a dated baseline, the preview has reached its limit. That is when I <a href="https://circleboom.com/twitter-management-tool/twitter-search-tool/twitter-list-manager/export-twitter-following-list">save the public following list as CSV</a>.</p><h3>The answer I give now</h3><p>When someone asks <strong>can you see following list on X without logging in</strong>, I do not answer with an unqualified yes. X may show a public profile and some visible accounts, but the result can stop, change, or demand authentication.</p><p>I ask what the person needs next:</p><ul><li>For a glance, try the public profile and accept the visibility limit.</li><li>For native browsing, log in when X requires a session.</li><li>For a structured file, authenticate Circleboom and export the public account.</li><li>For a protected target, stop at the privacy boundary.</li></ul><p>That sequence avoids both extremes. It does not pretend public pages are always inaccessible, and it does not pretend a few rendered rows equal a complete list.</p><p>Access is the doorway. Export is the documented handoff into analysis.</p><h3>FAQ about following-list access</h3><h3>Does an X login guarantee that I can see a complete following list?</h3><p>No. A login may restore native browsing, but the interface is still not a completeness guarantee or structured export. Use an appropriate authenticated export when the task requires a file and row-level analysis.</p><h3>Can I export someone’s following list if I do not control their account?</h3><p>Yes, when the target account is public. You connect your own authorized X account to Circleboom, confirm the public target, and use Export Following; protected target lists are not available.</p><h3>What is the safest way to save a following list?</h3><p>Use the official-API workflow, verify the target, record the date and row count, store the CSV with limited access, and refresh it before decisions that require current data.</p><h3>Why does the visible list look shorter than the following count?</h3><p>The profile metric and the rendered browser list are different outputs. Login walls, pagination, session limits, and interface loading can leave many relationships outside the visible preview.</p><p><strong>→ </strong><a href="https://circleboom.com/twitter-management-tool/twitter-search-tool/twitter-list-manager/export-twitter-following-list"><strong>Move from an incomplete preview to a dated CSV</strong></a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=dcd1eeb4f87f" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[I export a Twitter following list monthly, and the row count rarely matches]]></title>
            <link>https://medium.com/@mariadunhum/i-export-a-twitter-following-list-monthly-and-the-row-count-rarely-matches-d4c36148d733?source=rss-d83cbad5dc9e------2</link>
            <guid isPermaLink="false">https://medium.com/p/d4c36148d733</guid>
            <category><![CDATA[csv]]></category>
            <category><![CDATA[social-media-analytics]]></category>
            <category><![CDATA[export-data]]></category>
            <category><![CDATA[twitter]]></category>
            <dc:creator><![CDATA[Maria Dunham]]></dc:creator>
            <pubDate>Tue, 27 Oct 2026 13:06:01 GMT</pubDate>
            <atom:updated>2026-10-27T13:06:01.337Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*qJCGFgqW1I6b6gah9-jSyQ.png" /></figure><p>My profile says I follow a certain number of accounts. The CSV I just downloaded says something slightly different. That mismatch used to send me back to re-run the export before I understood it was not an error at all.</p><p>It is the shape of the X API showing through the file.</p><blockquote><strong><em>What a following list export actually gives you.</em></strong></blockquote><ul><li>Every account you follow on X as one CSV row instead of a scroll position.</li><li>Sixteen fields on each row, including account age, follower counts, and quality flags.</li><li>A dated snapshot you can hold next to the one you take in three months.</li></ul><blockquote><em>Circleboom builds that file from the accounts any public X account follows, through official API access. Pull yours here: </em><a href="https://circleboom.com/twitter-management-tool/twitter-search-tool/twitter-list-manager/export-twitter-following-list"><em>export a Twitter following list</em></a><em> </em>(this link takes you to Circleboom, outside Medium)<em>.</em></blockquote><h3>The number on the profile and the number in the file are two different numbers</h3><p>Let me be precise about the stakes here, because they are smaller than they sound and more annoying than they look.</p><p>Nobody loses data here. The export is a read operation, and running it twice costs you tokens, not follows.</p><p>What you lose is confidence.</p><p>If you are handing a client a follower-network audit and the row count in your file does not line up with the number on the public profile, the first question you get asked is whether the file is complete.</p><p>If you cannot answer that question, the audit is worth less than the hour you spent on it.</p><p>So I stopped treating the count as a checksum and started treating it as a range.</p><p><strong>The file is not a copy of the profile. It is a retrieval of the relationship at a moment.</strong></p><p>That reframe is what made the whole thing usable for me.</p><p>Anyone who has watched their own <a href="https://circleboom.com/blog/my-twitter-followers-and-followings-are-not-showing-in-my-followers-list/">followers and followings not show up in the followers list</a> already knows that X’s own surfaces disagree with each other. Often inside the same session. An export is not exempt from that.</p><p>Everything below is easier with a real file on screen. Run an export of <a href="https://circleboom.com/twitter-management-tool/twitter-search-tool/twitter-list-manager/export-twitter-following-list">your own following list on X</a> <em>(this link takes you to Circleboom, outside Medium)</em> and open the CSV in a spreadsheet before you read on.</p><h3>Why does the exported row count come back lower than X shows?</h3><p>Because very large lists come back in batches, and the exported row count reflects however many accounts the API returned inside the retrieval window.</p><p>That is the whole answer, and it is worth unpacking once.</p><p>X’s API does not hand over a follower or following list in a single response. It returns results in pages, each one carrying a token that fetches the next, and X’s <a href="https://docs.x.com/x-api/fundamentals/pagination">pagination documentation</a> notes that a page can come back holding fewer results than you asked for.</p><p>Circleboom walks that chain for you. For an account following a few hundred or a few thousand, the chain finishes and the row count matches what you would expect.</p><p>For an account following six figures, the retrieval takes longer and the exported row count can land under the account’s displayed count. That gap is pagination, not a broken export, and re-running it usually produces a file the same size.</p><p>Here is where I was wrong for months.</p><p>I assumed a short file meant Circleboom had failed and a full file meant Circleboom had succeeded, so I kept burning tokens re-running exports on large accounts.</p><p>The row count was never the success signal. Whether the rows I actually needed were present was.</p><p><strong>Check the segment you care about, not the total.</strong></p><p>If you exported a competitor’s following list to find the twelve industry accounts they pay attention to, and all twelve are in the file, the file did its job whether it holds 40,000 rows or 38,400.</p><h3>How to export a Twitter following list without guessing at the count</h3><p>The run takes about a minute once you know where the buttons live.</p><p>Circleboom keeps account exports under Essential Toolbox, separate from the tools that export posts, and the following list and follower list share one screen with two buttons. Reading the token balance before you commit is the step people skip and then regret.</p><h3>Reaching the Export Followers or Following screen</h3><ul><li><a href="https://circleboom.com/twitter-management-tool"><strong>Log in to Circleboom Twitter</strong></a> <em>(this link takes you to Circleboom, outside Medium)</em>. Authorization runs through the account whose exports you want.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*t3MS51lO1R4fsdJ7.png" /></figure><ul><li><strong>Go to the Essential Toolbox menu</strong> in the left navigation.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*pwHQlO-5L0OlTHwm.png" /></figure><ul><li><strong>Open Export Tools and choose Export Followers or Following</strong>, which is the account-list export rather than the post export.</li></ul><h3>Checking the account and the token balance</h3><ul><li><strong>Enter any public X username.</strong> Circleboom answers with a preview card: photo, handle, bio, how many days the account has existed, its post total, and both relationship counts.</li><li><strong>Compare the preview card’s following count against what you expect.</strong> This is your only pre-flight read on how big the retrieval will be.</li><li><strong>Read the remaining token balance on the page.</strong> Each export consumes tokens based on the number of accounts exported, so the cost of a 500-account list and a 50,000-account list are not comparable.</li></ul><h3>Writing the following list to CSV</h3><ul><li><strong>Select Export Following, not Export Followers.</strong> They sit next to each other and the wrong one costs you a second run.</li><li><strong>Click Export</strong>, and the file downloads on its own once the retrieval finishes.</li></ul><p>That sequence works because the preview card and the balance both sit upstream of the irreversible part. You confirm the account, you see what it will cost, and only then do tokens move.</p><p>Reverse the order and you find out you exported the wrong username after you paid for it.</p><p>One caution before you point this at other people’s accounts.</p><p>Following lists on X are public by default, which is exactly why the export works at all. There is a live question about whether you can <a href="https://circleboom.com/blog/is-there-a-way-to-prevent-people-from-seeing-your-following-list-on-x/">prevent people from seeing your following list on X</a>. Assume anything you can export from someone else, someone else can export from you.</p><p><strong>Watch it run: the preview card and the token balance in the two seconds before the Export Following button does anything.</strong></p><p><a href="https://www.youtube.com/watch?v=gOXfzG3jVz4">https://www.youtube.com/watch?v=gOXfzG3jVz4</a></p><h3>What I check in the file before I trust it</h3><p>Four passes, in this order, and none of them is the row count.</p><p>First, I sort by Username and scan for the handles I ran the export to find. If the specific accounts I care about are present, the file is fit for the question I asked it.</p><p>Second, I check IsProtected. Every row carrying that flag is an account whose own lists the API will never release, so I learn upfront which entries cannot be followed up on.</p><p>Third, I check CreatedAt for anything that looks recent. New accounts with large follower counts behave differently from old ones, and the timestamp is the cheapest way to spot them.</p><p>Fourth, I date the filename. Every value in the file was true at the instant of retrieval and not a second later, so an undated file turns into a liability inside a quarter.</p><p>The same discipline applies to every other CSV Circleboom writes. A <a href="https://circleboom.com/blog/twitter-dm-downloader-how-to-export-and-clean-up-your-messages/">Twitter DM downloader</a> file goes stale the same way, for the same reason.</p><p>Prefer to start from the follower side? Circleboom’s <a href="https://circleboom.com/twitter-management-tool/twitter-search-tool/twitter-list-manager/export-list-of-followers-on-twitter">export list of followers on Twitter</a> screen mirrors this one, and all four checks transfer without modification.</p><p>Launch the <a href="https://circleboom.com/twitter-management-tool/twitter-search-tool/twitter-list-manager/export-twitter-following-list">export Twitter following list</a> <em>(this link takes you to Circleboom, outside Medium)</em> page and run all four passes on your own file. It takes about a minute.</p><h3>When the gap is worth a second run, and when it is not</h3><p>The honest answer is that most of the time it is not, and I have wasted enough tokens learning that to be confident about it.</p><p>A second run is worth it when the first one returned a file that is obviously truncated rather than slightly short.</p><p>If you asked for the following list of an account with a few thousand follows and the CSV holds a few hundred rows, something interrupted the retrieval and running it again is reasonable.</p><p>That is a different situation from a 40,000-row file coming back at 38,400.</p><p>A second run is also worth it when time has passed. Nothing in the file updates after the retrieval ends, so an export from last month should be replaced rather than reconciled. Re-running is cheaper than being wrong about who you follow.</p><p>One balance covers every account export in the app, including <a href="https://circleboom.com/twitter-management-tool/twitter-search-tool/export-twitter-accounts">Export Twitter Accounts</a>, so a wasted re-run on a following list costs you a retweeter export later.</p><h3>When a re-run will not help</h3><p>A second run is not worth it when the accounts you needed are already in the file. This is the case that catches people, because a mismatch feels like an error even when the file answers the question perfectly.</p><p>And it is not worth it on protected accounts, where a second run returns exactly what the first one did, which is nothing. The API will not release a locked account’s lists no matter how many times you ask.</p><p><strong>The test is whether the missing rows would change what you do next.</strong></p><p>If you cannot name a decision that turns on them, you do not need them.</p><h3>Three things the columns settle that the count cannot</h3><p>The row count answers one question badly. The columns answer several well, and this is where a following list earns its keep.</p><ul><li><strong>IsInactive</strong> separates the accounts still posting from the ones you followed and never heard from again.</li><li><strong>CountFollowers and CountFriends</strong> together give you the follow ratio, which is how you spot the accounts following everyone versus the accounts being followed.</li><li><strong>Verified_Type</strong> carries four possible values, blue, business, government and none, which is a sharper filter than the single verified checkbox on a profile.</li></ul><p>Those three columns are why I export the following list rather than screenshot the Following tab.</p><p>Circleboom is <a href="https://docs.x.com/enterprise/customer-directory#enterprise-customers-listing">listed on X’s Enterprise customer directory</a>, so the flags come attached to the retrieval rather than being inferred from a rendered page. They also stay stable enough to compare across two files taken months apart.</p><p>They make the file portable in a way a screenshot never is. Segmenting a following list is the same motion as <a href="https://circleboom.com/blog/how-to-find-list-and-export-your-high-quality-followers-and-friends-on-twitter/">finding, listing, and exporting your high quality followers and friends on Twitter</a>. You are just applying it to the half of your network you built on purpose.</p><h3>The part worth keeping about the count</h3><p>Launch the export for the segment you need, not for a number that will match.</p><p>The row count was never the deliverable. The rows were.</p><p>If the accounts you ran the export to find are sitting in the file with their flags attached, you have what you came for, and a re-run will not improve it. If they are missing, the fix is a narrower question, not a bigger file.</p><p>Start with the <a href="https://circleboom.com/twitter-management-tool/twitter-search-tool/twitter-list-manager/export-twitter-following-list">Twitter following list to CSV</a> <em>(this link takes you to Circleboom, outside Medium)</em> export and let the columns do the arguing.</p><h3>Questions left over after the export</h3><h3>Can I export the following list of an account that is not mine?</h3><p>Yes, as long as it is public. You enter the username, confirm the preview card, and choose Export Following. This is the main reason the feature exists, since your own following list is only one of the networks worth reading.</p><h3>Does exporting someone’s following list notify them?</h3><p>No. Nothing is followed, unfollowed, blocked, or messaged. The export is a read against the API, and the accounts in the file have no signal that it ran.</p><h3>What happens if I run out of tokens partway through a big list?</h3><p>The export cannot proceed until the balance is restored, and the balance is shown on the page before you confirm, which is why the pre-flight read matters on large accounts. Tokens restore on your plan cycle.</p><h3>Is the CSV usable in Google Sheets, or do I need Excel?</h3><p>Either. It is a plain CSV with one header row and one row per account, so anything that opens a spreadsheet will read it, including the import step of most CRMs.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=d4c36148d733" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Find Twitter accounts tweeting in a language: the X field behind it]]></title>
            <link>https://medium.com/@mariadunhum/find-twitter-accounts-tweeting-in-a-language-the-x-field-behind-it-4afeb7586587?source=rss-d83cbad5dc9e------2</link>
            <guid isPermaLink="false">https://medium.com/p/4afeb7586587</guid>
            <category><![CDATA[audience-research]]></category>
            <category><![CDATA[twitter]]></category>
            <category><![CDATA[language]]></category>
            <category><![CDATA[multilingual-marketing]]></category>
            <dc:creator><![CDATA[Maria Dunham]]></dc:creator>
            <pubDate>Mon, 26 Oct 2026 13:01:02 GMT</pubDate>
            <atom:updated>2026-10-26T13:01:02.101Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*OHS-MUiKnZtlyRZybWhf4Q.jpeg" /></figure><p>Circleboom’s account export writes sixteen columns for every X profile it returns. Location is one of them. Language is not, and after running enough of these searches I have come to read that absence as the most useful thing on the screen.</p><blockquote><strong><em>Manual:</em></strong><em> open profiles one at a time, guess from the words, lose the list when you close the tab. </em><strong><em>Circleboom:</em></strong><em> filter public posts by detected language through Enterprise-level X API access, then pivot to the deduplicated accounts behind them.</em></blockquote><blockquote><strong><em>→ Pull a list of </em></strong><a href="https://circleboom.com/twitter-management-tool/twitter-search-tool/search-twitter-accounts"><strong><em>Twitter accounts tweeting in a language</em></strong></a><em> </em>(this link takes you to Circleboom, outside Medium)</blockquote><blockquote><em>The label those accounts carry is a machine guess about one post. Below is how I check it.</em></blockquote><h3>Where X keeps language, and where it does not</h3><p>I spent an embarrassing amount of time hunting for a language dropdown that was never going to be there.</p><p>The reason sits in <a href="https://docs.x.com/x-api/fundamentals/data-dictionary">X’s own field reference</a>. The lang attribute belongs to the Post object, and X&#39;s wording for it carries a condition: &quot;if detected by Twitter.&quot;</p><p>Look at the User object in the same document and count the fields. Name, username, description, location, created_at, public_metrics, verified_type, pinned_tweet_id, protected, and seventeen more.</p><p>Not one of them is a language.</p><p>So a “Spanish account” is not a thing X can hand back. A Spanish post is. Everything else you build has to be assembled from that.</p><p>That is why Circleboom’s two account-level screens, Deep X Account Search and Live X Account Search, filter by Location and never by Language. They read profiles. Profiles have no language on them.</p><p>The tweet-level screens do carry a Language filter, both the real-time one and the historical one, because posts are the only place the attribute exists.</p><p>The shape of that problem is not unique to X. Any network that classifies content rather than people forces the same detour. It is part of why it takes real work to <a href="https://circleboom.com/blog/how-to-search-and-find-people-on-bluesky/">search and find people on Bluesky</a> by anything other than a handle.</p><p>Once you accept the detour, the tooling is straightforward. <a href="https://circleboom.com/twitter-management-tool/twitter-search-tool/search-twitter-accounts"><strong>Find X accounts tweeting in a language</strong></a> <em>(this link takes you to Circleboom, outside Medium)</em> and the two hops happen on one screen.</p><h3>Why the two-hop route beats what most guides suggest</h3><p>Most articles on this query hand you the lang: operator and stop.</p><p>That works if you want to read posts. It falls apart the moment you want people, because a results feed is not a list. You end up opening profiles in tabs, copying handles into a note, losing half of them, and starting over next week with no record of what you already checked.</p><p>The <a href="https://docs.x.com/x-api/posts/search/integrate/operators">`lang:` operator in X’s search reference</a> works on posts classified into a given language. Posts. The step everyone skips is the one that turns those posts into a deduplicated, sortable, exportable set of accounts.</p><p>That step is a single button, and its behavior deserves precision. It reads “Display Profiles of this search (N)”, and N is smaller than your tweet count, often much smaller. Collect 500 posts on a keyword and you might get 280 unique authors, because the loudest accounts in any language community post several times each.</p><p>The number in that button is your real result. The tweet count was only the net.</p><h3>How I run a language-scoped account search on X</h3><p>I treat the tweet count as a probe, not a goal, and keep it low on the first pass, because collection consumes GetTweetTokens in proportion to posts gathered. A wide first run burns tokens teaching you that your keyword was ambiguous. I start small, read the result, then re-run.</p><p>Here is the sequence I settled on.</p><h3>Get to the screen that reads post language</h3><ul><li><a href="https://circleboom.com/twitter-management-tool"><strong>Open Circleboom Twitter</strong></a> <em>(this link takes you to Circleboom, outside Medium)</em> and link whichever X profile the search will run under.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*Y_SwljqnhwkUHPhd.png" /></figure><ul><li><strong>Head for the Advanced X Search menu</strong>, home to all seven discovery tools.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/915/0*E56mSwmYxQa0Rv0E.png" /></figure><ul><li><strong>Start Real-time Tweet Search</strong> and type what you want in your own words. The interface reads plain language and offers refined query variations, so nobody has to memorize operator syntax.</li></ul><h3>Set the filters that decide who reaches your list</h3><ul><li><strong>Set Language to your target</strong> in the Filters panel. Sitting beside it: a match-type selector, exclude terms, separate toggles for replies and for posts carrying links, hashtag and cashtag controls, a verified-only switch, media type, and engagement minimums.</li><li><strong>Choose a start date, and keep the tweet count low</strong> on a first pass. The presets run 24 hours, 7 days, and 30 days, with a custom range beside them. Read what comes back before spending more tokens.</li><li><strong>Hit “Display Profiles of this search”</strong> and work from the account view: sort by tweet count, filter by follower range, then follow, add to an X List, or export as CSV.</li></ul><p>That order works because every filter you set before collection reduces what the token spend has to pay for. Filtering afterward costs the same tokens and gives you a smaller usable result.</p><h3>The verification pass I run before acting on anything</h3><p>A language label is one observation of one post. Treating it as a property of a person is where the mistakes start.</p><p>The profile view carries the columns that turn an observation into a judgment:</p><ul><li>join date</li><li>follow ratio</li><li>the follower and following numbers side by side</li><li>lifetime post count</li><li>a three-way activity read of Active, Moderate Active, or Inactive</li></ul><p>Read those together and a weak match separates from a strong one quickly.</p><p>An account with 9,000 posts, an Active classification, and a matching tweet from yesterday is almost certainly writing in that language habitually. An account with 40 posts that matched once is a coin flip.</p><p>Three checks catch most of what slips through:</p><ul><li>Run the identical query across two different start dates and keep the accounts present in both.</li><li>Open five profiles from the middle of the result, not the top, and read their last ten posts.</li><li>Watch for accounts whose display name is in one script and whose posts are in another.</li></ul><p><strong>Overlap across two runs is the cheapest signal you will get.</strong> A one-off match can come from a quote, a joke, or a reply to a foreigner. A match on two separate days rarely can.</p><p>Circleboom is <a href="https://docs.x.com/enterprise/customer-directory#enterprise-customers-listing">listed on X’s Enterprise customer directory</a>, so the text being classified is the text X actually published. That removes one whole category of error, though not the classifier’s own.</p><h3>Bilingual accounts land in both language buckets</h3><p>I assumed a filter for one language would keep bilingual accounts out. It does the opposite, and I now think that is correct behavior rather than a flaw.</p><p>An account that writes Turkish most of the week and switches to English for a Friday thread lands in both buckets. A Turkish query catches them on Monday. An English query catches the same person on Friday. Neither result is wrong, because neither is a claim about a human being.</p><p>For outreach this matters more than it sounds. If you are writing to someone in their second language because a Friday post matched, you will read as a stranger who did not look. Reading five recent posts before contact costs a minute and changes the tone of everything after it.</p><p>The same gap opens whenever you <a href="https://circleboom.com/blog/how-can-i-find-people-with-the-same-interests-as-me-on-x/">find people with the same interests as you on X</a>. The signal points at someone. It does not describe them.</p><h3>Which posts can a language filter never see?</h3><p>Detection needs text. Some posts do not have enough of it.</p><p>A post that is one link, or four hashtags, or a lone emoji hands the classifier almost no text, and X’s field description keeps that risk visible with a single word: <em>if</em>. Those posts fall outside every language filter you will ever run.</p><p>This produces a quiet skew in your results. Accounts that write in sentences are overrepresented. Accounts that mostly share links, retweet, or post images with a hashtag are underrepresented, in every language, always.</p><p>If your target community leans visual, expect a thinner result than the community’s real size and adjust your read accordingly. That is a genuine limit of the method, not a setting you can change.</p><p>Where the skew hurts most is quality assessment.</p><p>A small return can mean a small community or a link-heavy one, and those need different responses. If it is the second, a bio-keyword pass reaches accounts the language filter structurally cannot see. Run it through <a href="https://circleboom.com/twitter-management-tool/twitter-search-tool/twitter-advanced-search-filters">Twitter advanced search filters</a> <em>(this link takes you to Circleboom, outside Medium)</em>.</p><h3>What the exported file actually contains</h3><p>The CSV writes ProfileId, Username, Name, Location, tweet count, following count, follower count, protected and verified flags, quality flags for egghead, fake, inactive and overactive, verification type, Bio, and the date the account was created.</p><p>Sixteen columns, and not one of them names a language. There was no such attribute on the accounts, so there is nothing for the file to record.</p><p>Whichever language that export represents is a fact about your search, not about its contents. Put it in the filename. I have opened enough unlabeled exports to know how fast that context evaporates.</p><p>Export draws on a token balance too. What is left of it shows beside the button, which is reason enough to filter a big result set down before downloading it.</p><p>The same discipline pays off on any Circleboom screen where you <a href="https://circleboom.com/twitter-management-tool/twitter-search-tool/export-twitter-accounts">export Twitter accounts</a> <em>(this link takes you to Circleboom, outside Medium)</em>.</p><p><strong>Watch it run: the keyword-to-account pivot that turns a filtered stream into a working list.</strong></p><p><a href="https://www.youtube.com/watch?v=hPPu49XaLGc">https://www.youtube.com/watch?v=hPPu49XaLGc</a></p><h3>Acting on the list without tripping anything</h3><p>Following is paced deliberately. The queue releases 50 follows per quarter-hour window and stops the day’s total at 400, then starts again on its own the following morning.</p><p>Nobody enjoys that constraint on day one. It is also the difference between a follow campaign that looks like a person with an interest and one that looks like a script.</p><p>On a first pass, a list beats a follow. An X List parks the segment somewhere you can watch it without touching your following graph, so a wrong call costs nothing. Promote the accounts that keep proving themselves; drop the rest silently.</p><p>That approach travels well. It is the same logic behind building a list when you <a href="https://circleboom.com/blog/find-the-intersection-of-followers-between-two-twitter-accounts/">find the intersection of followers between two Twitter accounts</a> rather than acting on the overlap immediately.</p><h3>Where to point this next</h3><p>A plan for the whole community is not the entry price. One keyword, one language, one week, and a small tweet count will tell you more about a language market than any amount of reading about it.</p><p>If the result is thin, widen the date range before widening the keyword. If it is noisy, add an engagement minimum before adding exclude terms. Those two moves fix most first attempts.</p><p>And if the accounts you find turn out to be a market rather than a curiosity, the same two-hop route scales without changing shape.</p><p>A language-scoped audience behaves like any other segment once it exists. The follow-up questions are the same ones that come up when you <a href="https://circleboom.com/blog/how-to-find-common-followers-between-twitter-accounts-without-coding/">find common followers between Twitter accounts without coding</a>.</p><p>Start narrow. <a href="https://circleboom.com/twitter-management-tool/twitter-search-tool/search-twitter-accounts">Find Twitter accounts tweeting in a given language</a> <em>(this link takes you to Circleboom, outside Medium)</em>. Let the first result tell you whether the market is there.</p><p><strong>→ Run your first </strong><a href="https://circleboom.com/twitter-management-tool/twitter-search-tool/search-twitter-accounts"><strong>language-filtered X account search</strong></a> <em>(this link takes you to Circleboom, outside Medium)</em></p><h3>FAQ on language-based discovery</h3><h3>Can I set up a search that keeps collecting on its own?</h3><p>Real-time Tweet Search keeps gathering from your chosen start date onward, so a query you leave running accumulates matching posts as they appear. The account view refreshes with new authors as the collection grows, which is what makes it useful during a live event rather than after one.</p><h3>Will this find accounts that have a language in their bio but post in another?</h3><p>No, and that is the point of using the tweet-level filter. A bio-based search matches profile text; a language filter matches what the account actually wrote. If you specifically want bio matches, the Find in Bio and Name field on the account-search screens does that job instead.</p><h3>Does the language filter work on replies and retweets?</h3><p>The Filters panel has separate toggles for including or excluding replies and posts with links, so you control that independently of language. Excluding replies usually raises quality, because replies are where code-switching happens most and where a one-word answer is most likely to be misclassified.</p><h3>How small a language community is still worth searching?</h3><p>Small enough that a keyword returns a few dozen accounts is still worth it, as long as those accounts are active. A language market with 50 engaged accounts writing daily is more useful than one with 5,000 dormant ones, and the activity classification in the profile view is what separates the two cases.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=4afeb7586587" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The X official way to see unfollowers is not the answer most people expect]]></title>
            <link>https://medium.com/@mariadunhum/the-x-official-way-to-see-unfollowers-is-not-the-answer-most-people-expect-72a86310ee99?source=rss-d83cbad5dc9e------2</link>
            <guid isPermaLink="false">https://medium.com/p/72a86310ee99</guid>
            <category><![CDATA[twitter]]></category>
            <category><![CDATA[followers]]></category>
            <category><![CDATA[unfollowers]]></category>
            <category><![CDATA[social-media-tool]]></category>
            <dc:creator><![CDATA[Maria Dunham]]></dc:creator>
            <pubDate>Sun, 25 Oct 2026 13:01:03 GMT</pubDate>
            <atom:updated>2026-10-25T13:01:03.393Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*rJAsMHaPMZ2JjGdOdWn4EQ.jpeg" /></figure><p>I have seen the same mistake in audience reviews more times than I can count: someone notices that their X follower count slipped, opens the follower list, and starts looking for a name that might explain it. The list feels like evidence.</p><p>It is not.</p><p>It is only a picture of the room after someone has already left.</p><blockquote><strong><em>The direct answer.</em></strong></blockquote><ul><li>X can show your current followers, but it does not keep a native, user-facing history of who unfollowed you.</li><li>A live follower list cannot reconstruct an account-level change after it happened.</li><li>Historical tracking has to begin before you need to review the loss.</li></ul><blockquote><em>Circleboom compares stored follower snapshots and identifies changes on Twitter/X using </em><a href="https://docs.x.com/enterprise/customer-directory#enterprise-customers-listing"><em>official X Enterprise APIs</em></a><em>. Start with </em><a href="https://circleboom.com/twitter-management-tool/twitter-circle-tool/twitter-audience-insights/who-unfollowed-me-on-twitter"><em>x official way to see unfollowers</em></a><em>.</em></blockquote><p>The expectation to reset is simple: <strong>you cannot recover a change log from a single live list.</strong> You either watched the change happen, or you started collecting the information before you needed it.</p><p>That is why <a href="https://circleboom.com/twitter-management-tool/twitter-circle-tool/twitter-audience-insights/who-unfollowed-me-on-twitter">start an X unfollower record</a> is a more honest next step than searching for a hidden X setting. It makes a promise a live list can actually support: future comparisons, not invented history.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*5AFtrbAdlqYmxWMW.jpg" /></figure><h3>The question behind the question</h3><p>People rarely search this because they want a spreadsheet of names. They want a reason.</p><p>Did a post alienate people? Did a new content direction miss the mark?</p><p>Did a follow-back account quietly leave after getting what it wanted?</p><p>Or did X clear a group of low-quality accounts from the platform?</p><p>Those are different questions, and the answer is not contained in a follower count. X’s own <a href="https://help.x.com/en/using-x/following-faqs">Following FAQ</a> documents the follow relationship, but it does not offer a native unfollower history or an individual unfollow notification.</p><p>That is the product boundary, not a setting you have failed to find.</p><p>Circleboom gives you a way to make future changes inspectable. <strong>Circleboom tracks follower-list changes on Twitter/X so you can review accounts that were present before and absent later.</strong> The value is not surveillance. It is context.</p><p>When I read a list of unfollowers, I do not begin with “Who should I punish?” I begin with “What changed around this window?” That small shift prevents a lot of bad social media decisions.</p><h3>Why the manual method creates false confidence</h3><p>The manual workaround sounds sensible: save a list, come back later, compare it, and find the missing names. It becomes impractical quickly.</p><p>Follower lists change constantly.</p><p>Usernames can change.</p><p>Accounts can deactivate.</p><p>An ordinary count drop gives you no clean way to distinguish a meaningful departure from platform noise.</p><p><a href="https://circleboom.com/blog/why-do-people-follow-then-unfollow-you-on-twitter-x/">Why people follow then unfollow you on Twitter/X</a> is a helpful reminder that some departures are behavioral, not personal. These can all produce the same visible result: a smaller number under your profile:</p><ul><li>follow-for-follow tactics;</li><li>topic mismatch;</li><li>dormant accounts; and</li><li>platform action.</li></ul><p>The false confidence comes from treating an available number as a complete explanation. It is the same mistake as judging a campaign from impressions alone. The metric is real. The story you attach to it may not be.</p><p><strong>Follower loss is not one event type. It is a category that needs sorting.</strong></p><p>That is why a record should include more than the account name. You want:</p><ul><li>the timing;</li><li>follower and following counts;</li><li>account age and activity classification;</li><li>follow ratio; and</li><li>language or location, where useful.</li></ul><p>These fields make it possible to see whether the change is concentrated in a segment.</p><h3>The failed shortcut: checking who does not follow back</h3><p>One approach I see people use is to compare their following list with their follower list and call every mismatch an unfollower. It is a tempting shortcut because the lists are visible today.</p><p>But “does not follow you back” and “previously followed you, then unfollowed” are not the same claim. The first describes a current relationship. The second requires history. An account may never have followed you at all.</p><p><a href="https://circleboom.com/blog/who-follows-whom-on-x-twitter-follower-tracker/">Who follows whom on X</a> helps with the reciprocity question, which is useful in its own right. It should not be used as a substitute for an unfollower log.</p><p>This was the point that changed how I think about the feature. A follower count is not a retention report. A non-mutual list is not a churn report. <strong>Retention starts with a before-and-after record.</strong></p><h3>What Circleboom records that X does not present natively</h3><p>Circleboom’s Who Unfollowed Me view compares follower snapshots and displays the accounts that dropped out of the relationship over a selected period. The list includes practical account context such as:</p><ul><li>username and location;</li><li>post count and join date;</li><li>following and follower counts;</li><li>follow ratio; and</li><li>an active or inactive classification.</li></ul><h3>A time window makes the evidence readable</h3><p>I find the distinction between account-level evidence and trend-level evidence especially helpful when a number looks alarming. The individual list tells you who left.</p><p><a href="https://circleboom.com/twitter-management-tool/twitter-user-analytics/twitter-follower-stats">Twitter Follower Stats</a> helps establish whether that movement sits inside a broader growth or decline pattern.</p><p>One view prevents vague conclusions; the other keeps a short-term event from being mistaken for a long-term direction.</p><p>Circleboom also lets you choose windows from one day through a year. That time selector is more useful than it first appears.</p><p>A long list tells you who disappeared. A short, event-based window gives you something you can compare with:</p><ul><li>a campaign calendar;</li><li>a content shift; or</li><li>an unusual posting burst.</li></ul><p>Circleboom uses <a href="https://docs.x.com/enterprise/customer-directory#enterprise-customers-listing">official X Enterprise APIs</a>, so the monitoring workflow is built around authorized access rather than a scraped list that you cannot audit. For a reader deciding whether to connect an account-management service, that distinction is not marketing decoration.</p><p>It is a reliability question.</p><h3>A history needs a stable measurement habit</h3><p><a href="https://circleboom.com/twitter-management-tool/twitter-circle-tool/twitter-follower-tracker">Twitter Follower Tracker</a> is another useful landing point when you want recurring audience monitoring rather than a single review. Tracking works best when the routine is quiet and regular.</p><p>The worst time to build the measurement habit is after a drop has already made you anxious.</p><p>For the account-level record itself, <a href="https://circleboom.com/twitter-management-tool/twitter-circle-tool/twitter-audience-insights/who-unfollowed-me-on-twitter">review X unfollower history</a> after the next daily update. The useful habit is a short weekly check with a longer review only when the timing or account mix points to a real audience question.</p><h3>Monitoring is a routine, not an emergency response</h3><p>If you are still deciding what kind of monitoring you need, <a href="https://circleboom.com/blog/twitter-follower-trackers/">Twitter follower trackers</a> offers the broader landscape. The important choice is not whether you want more dashboards.</p><p>It is whether the dashboard preserves the specific before-and-after evidence your question requires.</p><h3>How to turn a list of unfollowers into a useful review</h3><p>The best review does not begin with bulk actions. It begins with a small working hypothesis. For example: “Did the pricing announcement lose longtime followers?” or “Did the latest growth campaign mostly attract accounts that never intended to stay?”</p><h3>Start with the event, not the usernames</h3><p>Use the Who Unfollowed Me view to test that hypothesis in a defined time range. Filter for:</p><ul><li>engagement;</li><li>account quality;</li><li>follower count;</li><li>language, location, or join date; or</li><li>a bio term that identifies a customer group.</li></ul><p>The active-filter bar lets you see the segment you have built before you draw a conclusion.</p><p>The workflow is simple, but the order matters:</p><ol><li>Select the smallest period that contains the event you are investigating.</li><li>Check whether departures cluster around a post, campaign, or account change.</li><li>Filter to find out whether those accounts have meaningful shared traits.</li><li>Export the segment if the result will influence reporting or future content work.</li><li>Make a relationship decision only after the pattern is clear.</li></ol><p><strong>The list should produce a better question before it produces an action.</strong></p><p>Circleboom can export the selected or filtered result as a CSV. Exports use a token balance, displayed beside the export control. That is useful when a team needs to compare churn against campaign dates, customer records, or a longer-term audience report.</p><p>For a closer look at the basic relationship-history method, <a href="https://circleboom.com/blog/how-to-know-who-unfollowed-you-on-twitter-in-2020/">how to know who unfollowed you on Twitter</a> adds another practical entry point. Its older wording does not change the central rule: today’s list cannot prove yesterday’s departure without a prior record.</p><h3>What not to do with an unfollower list</h3><p>The most common overreaction is to automatically unfollow everyone who left. That may be reasonable for a deliberate reciprocity policy, especially when you had followed an account back and the relationship changed. It is not a good default for every departed account.</p><p>Circleboom offers these actions where the relationship exists:</p><ul><li>follow;</li><li>add to a list;</li><li>whitelist or blacklist; and</li><li>export or unfollow.</li></ul><p>If you choose to unfollow a group, Circleboom processes up to 50 unfollows every 15 minutes and up to 800 per day.</p><p>Those limits are a reason to be selective, not a target to hit.</p><p>There is a more subtle reason to pause. Some accounts on the list may be deactivated or affected by X’s spam handling rather than making a conscious choice about your content.</p><p>X notes in its <a href="https://help.x.com/en/using-x/common-following-issues">common following issues guidance</a> that a new follower who later disappears may have unfollowed, deactivated, or been hidden for spam activity.</p><p>That uncertainty is useful. It keeps you from blaming a post for a platform-level event, and it keeps a normal churn pattern from becoming a sudden strategy panic.</p><p>If your practical goal is cleaning a one-sided following list, <a href="https://circleboom.com/blog/best-unfollow-app-for-twitter/">best unfollow app for Twitter</a> is the more relevant next read. Cleanup and retention analysis can work together, but they should start from different evidence.</p><h3>The expectation reset that makes the tool valuable</h3><p>The x official way to see unfollowers does not exist as a retrospective X report. There is no hidden tab that reveals a complete list of everyone who left last month.</p><p>X gives you the present state.</p><p>A tracking service creates the comparison that turns present state into a history.</p><h3>A daily record is more useful than a false instant answer</h3><p>That is also why the service should be judged by how it handles limits. Circleboom’s daily data update means the record is timely enough for meaningful review without pretending to be a second-by-second notification feed.</p><p>The product notes that occasional X API discrepancies can temporarily put an account on the list and correct it after a subsequent fetch. A cautious team treats that as a prompt to inspect, not an instruction to act blindly.</p><p>There is a quiet operational benefit here too. Once the account history exists, the same time window can be revisited after a product launch, a repositioning decision, or a planned content experiment.</p><p>You do not have to reconstruct the baseline from screenshots or memory. You can compare the same type of audience change using the same filtering logic each time.</p><p>Circleboom updates follower data daily, which means an immediate departure may not appear until the next sync. It also means a small number of entries can be corrected after a later data fetch.</p><p>That is not a flaw to hide. It is the practical limit that separates a careful monitoring tool from a promise that it can know everything instantly.</p><p><strong>Watch the alert workflow for future follower changes.</strong></p><p><a href="https://www.youtube.com/watch?v=_WE4XrbVyz0">https://www.youtube.com/watch?v=_WE4XrbVyz0</a></p><h3>Your next move is to start before the next drop</h3><p>You do not need to obsess over every unfollower to manage an audience well. You need a history that lets you tell normal movement from a real retention signal, and a review habit that respects the difference.</p><p>The audience problem is not visibility. It is interpretation. A count tells you that something moved. A dated, filterable record lets you work out what moved and whether it matters.</p><p><strong>→ </strong><a href="https://circleboom.com/twitter-management-tool/twitter-circle-tool/twitter-audience-insights/who-unfollowed-me-on-twitter"><strong>track who unfollowed you on Twitter/X</strong></a></p><h3>Still wondering about unfollower tracking?</h3><h3>Does X have an official unfollower tracker?</h3><p>No. X provides current follower information but does not provide a native historical list of individual unfollow events. Circleboom builds that history by comparing follower snapshots over time.</p><h3>Can a tracker tell me why someone unfollowed?</h3><p>No tool can know a person’s private reason. Circleboom can show the time and account context, which helps you test whether the change aligns with a campaign, topic shift, or a pattern in audience quality.</p><h3>Will an unfollower alert appear instantly?</h3><p>Not necessarily. Circleboom updates follower data daily, so a recent change can appear after the next sync rather than at the exact moment it happens.</p><h3>Should I unfollow everyone who unfollows me?</h3><p>Only if that matches a deliberate relationship policy. Review the accounts first, because deactivation, spam handling, and normal churn can all create entries that do not require a reaction.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=72a86310ee99" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[How I block adult content on X, and where blocking stops working]]></title>
            <link>https://medium.com/@mariadunhum/how-i-block-adult-content-on-x-and-where-blocking-stops-working-f5ceb4eb77b2?source=rss-d83cbad5dc9e------2</link>
            <guid isPermaLink="false">https://medium.com/p/f5ceb4eb77b2</guid>
            <category><![CDATA[twitter]]></category>
            <category><![CDATA[content-moderation]]></category>
            <category><![CDATA[online-safety]]></category>
            <category><![CDATA[blocking]]></category>
            <dc:creator><![CDATA[Maria Dunham]]></dc:creator>
            <pubDate>Sat, 24 Oct 2026 16:01:02 GMT</pubDate>
            <atom:updated>2026-10-24T16:01:02.846Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*mQerQmR5_ncW3p_Dlj7hzA.png" /></figure><p>Blocking is the most-recommended answer to this and it is the wrong first move for most of what people are actually complaining about.</p><p>It works on exactly one of the three sources, and it is permanent in a way the other two fixes are not.</p><p>I use it. I use it fourth.</p><blockquote><strong><em>So when does blocking actually help?</em></strong></blockquote><blockquote><em>When an account can reach you and you want that closed. Not when a number is wrong, and not when your own feed is the problem, because blocking is a relationship control rather than a content filter. Circleboom pulls account lists through official X API access and stages a filtered, reviewable block queue for the cases that genuinely need it.</em></blockquote><blockquote><strong><em>→ </em></strong><a href="https://circleboom.com/twitter-management-tool/twitter-search-tool/mass-block-twitter-accounts"><strong><em>block the accounts posting adult content</em></strong></a><em> </em>(this link takes you to Circleboom, outside Medium)</blockquote><h3>The category error I was making</h3><p>I spent a long time looking for a way to block adult content and never found one, because it does not exist.</p><p>X lets you block accounts. It lets you mute words. It gates media behind a warning if the media has been marked.</p><p>None of those is a content block, and the difference is not pedantry, it changes what you should be doing.</p><p>A block is attached to an account and its access to you.</p><p>It has no relationship to what that account posts, which is why blocking one adult account does nothing about the next one and why people conclude the feature is broken.</p><p><strong>You can block who posts it. You cannot block what they post.</strong></p><p>Once that landed, the question stopped being how to block content and became which accounts are putting it in front of me. That has three separate answers, and only one of them is a blocking problem.</p><h3>The three sources, and where blocking applies</h3><p><strong>Accounts I follow.</strong> In my timeline because I subscribed, sometimes years ago, and the account drifted since. Blocking my own past decisions is heavier than the situation calls for. Unfollowing is the proportionate fix.</p><p><strong>Accounts that follow me.</strong> Costing me engagement rate by growing the denominator, sitting dormant, doing nothing. Removal fixes the number. Blocking adds permanence I do not need.</p><p><strong>Accounts that reach me without either.</strong> Replies, quote posts, messages, and content from the wider platform. This is the only category where blocking is the correct tool, because none of it depends on a follow relationship and nothing softer closes it.</p><p>Most people have all three and reach for blocking to solve all three. That is why it feels like it never quite works.</p><h3>Where blocking stops working entirely</h3><p>Two boundaries, and they are worth knowing before you spend an evening.</p><p><strong>Content from accounts you have no relationship with.</strong> For You, trending and reply threads under other people’s posts pull from the whole platform. You can block every account that has ever interacted with you and this channel stays completely open, because those accounts have not interacted with you yet.</p><p><strong>New accounts.</strong> A network operates a fleet and replaces it. Blocking a fleet is real work with a real result and it does not touch the operator, so your handle re-enters the target pool the next time a post of yours gets distribution.</p><p>That is not an argument against blocking. It is an argument for knowing what you are buying: a permanent boundary against specific accounts, not a filter against a category.</p><p><strong>Short demo: cutting non-verified accounts out of a follower list, which is a blunter version of the same triage.</strong></p><p><a href="https://www.youtube.com/watch?v=uahF3YhV5lY">https://www.youtube.com/watch?v=uahF3YhV5lY</a></p><h3>How I run the block pass when it is the right call</h3><p>Three phases. The middle one is the reason the first one is safe.</p><h3>Build the list somewhere other than the notifications tab</h3><ul><li><strong>Open </strong><a href="https://circleboom.com/twitter-management-tool"><strong>Circleboom Twitter</strong></a> <em>(this link takes you to Circleboom, outside Medium)</em> and connect the account being targeted.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*19qyC3-ZKckXiS-i.jpg" /></figure><ul><li><strong>Go to Essential Toolbox</strong> and open Mass Block X Accounts.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*GBILi2s4qH-Dim9O.png" /></figure><ul><li><strong>Pick the source deliberately.</strong> A public username pulls its followers or following. The retweeter tab takes post URLs and builds a cart from the accounts that reposted them, which is by far the tighter match when a specific post attracted the wave.</li></ul><h3>Cut it down before it becomes a block list</h3><ul><li><strong>Whitelist first, every time.</strong> Anyone I recognise gets protected before I touch a filter, and the label holds across every future session and every other feature.</li><li><strong>Apply Fake/Spam, then Find in Bio and Name</strong> for the phrases the network repeats. Adult promotion accounts have to advertise in the bio, so text matching is more precise than any behavioural score here.</li><li><strong>Add a Follow Ratio ceiling</strong> to catch what the text search missed.</li><li><strong>Read one full page by hand.</strong> A filter catching accounts I would not block individually is set too wide, and there is no later point where discovering that gets cheaper.</li></ul><h3>Stage, review, execute</h3><ul><li><strong>Select and click Add to Mass Block List.</strong> Nothing is blocked. This is a queue I can still edit.</li><li><strong>Open the Twitter X Mass Blocker extension and press Start</strong>, leaving the browser open while it works through the list with automatic pauses when X’s rate limit is reached.</li></ul><p>The staging step is the one people skip and the one that matters. Between adding to the list and pressing Start there is a complete, inspectable set in front of me.</p><p>After Start there is no batch reversal at all, so I would rather <a href="https://circleboom.com/twitter-management-tool/twitter-search-tool/mass-block-twitter-accounts">review the block queue before it runs</a> than find a mistake six months later.</p><h3>Why I default to removal instead</h3><p>The asymmetry is the whole reason, and it is not about severity.</p><p>A wrong removal reverses itself. The person can follow me again with no friction and will never know anything happened, so the worst case is somebody clicking follow twice over the course of a year.</p><p>A wrong block is permanent until I go find it in a list of thousands, has no batch undo, and the person on the other end never mentions it because they have no reason to. Nobody sends you a message to say they can no longer read you.</p><p>Given a review process I know is imperfect by construction, choosing the reversible action is arithmetic rather than caution.</p><p>So dormant accounts sitting in my follower count get <a href="https://circleboom.com/twitter-management-tool/twitter-circle-tool/remove-twitter-x-followers">removed rather than blocked</a>, and blocking waits for evidence that removal is not enough.</p><p>That evidence is specific. The same cluster returning within a week, or accounts putting links into my replies. Both mean the accounts are being actively driven rather than sitting idle.</p><p><a href="https://circleboom.com/blog/is-it-better-to-have-lots-of-twitter-followers-or-fewer-followers-who-are-more-engaged/">Is it better to have lots of followers or fewer engaged ones</a> is the wider version of the same trade-off.</p><h3>The order I work in now</h3><p>Four steps, and blocking is the last one for a reason.</p><p><strong>One: check the media setting.</strong> Two minutes. X permits marked adult content and gates it behind a warning, and accounts with no birth date on the profile cannot view marked content at all.</p><p>If it is displaying to me, my account permits it, and that is the cheapest thing to rule out.</p><p><strong>Two: audit what I follow.</strong> This is where most of what reaches my timeline actually comes from, and no setting touches it because I opened the door myself:</p><ul><li>Reciprocal follow-backs from years ago</li><li>one-post follows I forgot</li><li>and accounts that were fine when I followed them and have since been abandoned or repurposed</li></ul><p><strong>Three: clear my follower list.</strong> This fixes the engagement rate and the demographic reports rather than the feed, and removal is the right action because it is reversible.</p><p><strong>Four: block what is left.</strong> The accounts still reaching me after all of that are, by definition, the ones a follow relationship was never the mechanism for. Replies, quote posts, messages.</p><p>That is what blocking is for and it is the only category where the permanence is worth accepting.</p><p>Running them in that order means each step reduces the work of the next one, and the permanent action applies to the smallest possible set.</p><p>Doing it backwards, which is what I did for a year, means blocking a large number of accounts to solve problems that two settings and an unfollow pass would have handled.</p><p><strong>Blocking is the last resort, not the first tool. Everything above it is reversible.</strong></p><p>By the time I get to step four the list is usually small enough that <a href="https://circleboom.com/twitter-management-tool/twitter-search-tool/mass-block-twitter-accounts">running the block queue</a> takes minutes rather than an afternoon.</p><h3>Why reporting first wasted my time</h3><p>I reported before I blocked for months, on the assumption that adult content is obviously against the rules.</p><p>Marked content is compliant. X permits consensually produced adult media as long as it carries the sensitive marking, so a report filed on those grounds asks the platform to enforce a rule the account is following.</p><p>That is why so many of those reports appear to go nowhere.</p><p>The categories that land are unmarked explicit media, which violates the marking requirement, and platform manipulation, which covers mass registration and automated behaviour at scale.</p><p>Switching my report category improved the outcomes and did not make reporting a plan, because enforcement timing is not mine to control.</p><p><a href="https://circleboom.com/blog/how-to-report-a-twitter-account-and-get-it-deleted/">How to report a Twitter account and get it deleted</a> covers the flow properly if you want to do it well rather than often.</p><p><strong>Report the account, not the content. Then act as if the report will not land.</strong></p><h3>What the extension is actually doing, and the time it costs</h3><p>X has no bulk block endpoint. Any tool claiming otherwise is describing a capability the platform does not expose.</p><p>So the extension blocks sequentially, inside the browser, through X’s own interface, at X’s own pace. Circleboom queues and filters; the extension executes.</p><p>Two consequences worth planning for. The browser stays open for the whole run, because closing it stops the process with the remainder unblocked.</p><p>And large queues take hours rather than minutes, since a rate-limit pause of one to twenty minutes is normal and recurs as often as the queue length requires.</p><p>My last run of about 900 accounts took most of an afternoon in a background tab. The filtering took forty minutes and the reading took twenty-five, which is the ratio I now expect.</p><p>Circleboom handles the pacing as an <a href="https://docs.x.com/enterprise/customer-directory#enterprise-customers-listing">official X Enterprise Developer company</a>, inside X’s documented limits.</p><p>Running parallel sessions to go faster is the reliable way to get the account you were protecting restricted, and I would rather a cleanup take an afternoon than take my account with it.</p><h3>The objection I get most</h3><p>People tell me the review pass is over-cautious, that these are obviously bots and reading a page of them is wasted time.</p><p>Some of them are obviously bots. That is not the population the review is for.</p><p>The review is for the fifty or so rows in every few hundred that match three signals for entirely ordinary human reasons:</p><ul><li>A dormant account from 2013 with no avatar</li><li>An international user who reads and never writes</li><li>A small brand handle that tweets quarterly</li></ul><p>Those rows are indistinguishable from network accounts at the filter level and completely distinguishable at the eyeball level, in about four seconds each.</p><p>Ten minutes of reading against a permanent decision is not caution. It is the only part of the workflow doing quality control, and it gets cheaper every run because the whitelist compounds.</p><p>A <a href="https://circleboom.com/twitter-management-tool/twitter-circle-tool/twitter-follower-viewer">follower viewer</a> is a fast way to sanity-check a handful before trusting a filter across hundreds.</p><p>The second objection is that blocking feels disproportionate at scale.</p><p>Blocking one adult bot is uncontroversial; blocking four hundred is the same decision four hundred times, and the accounts did not become more sympathetic in aggregate.</p><p>What deserves hesitation is the source list, not the count. <a href="https://circleboom.com/blog/twitter-defamation-how-to-protect-your-reputation-in-the-age-of-viral-negativity/">Twitter defamation and protecting your reputation</a> covers the cases where the scale genuinely is the point.</p><p><strong>→ </strong><a href="https://circleboom.com/twitter-management-tool/twitter-search-tool/mass-block-twitter-accounts"><strong>stage your own block queue</strong></a> <em>(link goes to Circleboom, outside Medium)</em></p><h3>Questions readers ask me</h3><h3>Can I block a word instead of an account?</h3><p>You can mute words, which filters matching posts out of your timeline and notifications. It misses images with no caption entirely, which is most of this category.</p><h3>Does blocking stop them following me again?</h3><p>Yes, for those accounts. A blocked account cannot follow you, see your posts, reply, or message you. New accounts from the same network are unaffected.</p><h3>Is removing everyone an option?</h3><p>Technically, and it is a large hammer. <a href="https://circleboom.com/blog/how-to-delete-all-followers-on-twitter/">How to delete all followers on Twitter</a> covers what that actually involves and why it is rarely the right call.</p><h3>How long does a large block queue take?</h3><p>Hours above a few hundred accounts, because of the rate-limit pauses. The browser stays open but you do not have to watch it.</p><h3>What if I block someone real?</h3><p>You will not hear about it, which is the entire argument for the review pass. Reversal works, and it works account by account rather than in a single undo.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=f5ceb4eb77b2" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Find tweets from a specific date on Twitter: the date is your weakest filter]]></title>
            <link>https://medium.com/@mariadunhum/find-tweets-from-a-specific-date-on-twitter-the-date-is-your-weakest-filter-b43545bd80f9?source=rss-d83cbad5dc9e------2</link>
            <guid isPermaLink="false">https://medium.com/p/b43545bd80f9</guid>
            <category><![CDATA[twitter-search]]></category>
            <category><![CDATA[research]]></category>
            <category><![CDATA[twitter-history]]></category>
            <category><![CDATA[advanced-search]]></category>
            <dc:creator><![CDATA[Maria Dunham]]></dc:creator>
            <pubDate>Fri, 23 Oct 2026 15:01:05 GMT</pubDate>
            <atom:updated>2026-10-23T15:01:05.818Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*HxjSwABnl8zZ6NO2UoSZLA.jpeg" /></figure><p>Most advice on this topic treats the calendar as the answer. Set a start, set an end, and the post appears.</p><p>That is backwards, and I say so as someone who has burned collection budget proving it. A date range is a container, not a query. It tells the search where to look and nothing at all about what to look for, which is why a perfectly correct window on a busy phrase still buries the one post you wanted under three hundred you did not.</p><p>The date earns its keep only when a second control is standing next to it. Choose that control well and a crowded day collapses into a readable list. Choose it badly and you have narrowed your way past the answer.</p><blockquote><strong><em>Date alone:</em></strong><em> a correct window, a wall of results, and no way to tell which row is yours. </em><strong><em>Date plus one control:</em></strong><em> the same window cut down by a single exclude term or a minimum like count, which is how Circleboom returns X posts you can actually read through.</em></blockquote><blockquote><strong><em>→ </em></strong><a href="https://circleboom.com/twitter-management-tool/twitter-search-tool/twitter-advanced-search/twitter-historical-data"><strong><em>find tweets from a specific date on X</em></strong></a><em> </em>(this link takes you to Circleboom, outside Medium)</blockquote><blockquote><em>The second control is usually the one you are least confident about, which is exactly why it works.</em></blockquote><h3>Is a date range enough to find one specific tweet?</h3><p>Almost never, and the reason is arithmetic rather than technology.</p><p>A single calendar day on X carries hundreds of millions of posts. Narrowing to one day removes roughly 364 parts in 365 of a year’s archive, which sounds decisive until you remember what remains.</p><p>Whatever fraction of a day’s posts match your phrase is still a crowd, and the crowd is not sorted by how much you care about it.</p><p>This shows up most sharply on exactly the searches people run most: a phrase that trended, a hashtag from a launch, a reaction to something everyone was reacting to. The date is right. The result set is a transcript of the internet having a moment.</p><p>I hit this myself running a one-day window on a phrase I was certain about. The date was correct. What came back was mostly quote tweets of a single popular post, and my target was somewhere underneath them. The window was not the problem. The query was lazy.</p><p>There is a related trap on the other side of the same coin. A window so narrow it excludes the post entirely produces the identical experience of failure, and you cannot tell the two apart from an empty screen.</p><p>So I now <a href="https://circleboom.com/twitter-management-tool/twitter-search-tool/twitter-advanced-search/twitter-historical-data">find a tweet from a specific date on X</a> <em>(this link takes you to Circleboom, outside Medium)</em> with the date set generously. The narrowing happens somewhere else.</p><p>The general shape of that habit, applied beyond dates, runs through <a href="https://circleboom.com/blog/more-targeted-twitter-x-account-search/">more targeted Twitter X account search</a>.</p><h3>The second axis that turns a date into an answer</h3><p>Pick one narrowing control to stand beside the date. One, not four.</p><p>Circleboom’s Historical Tweet Search offers several narrowing controls for an X date window, and they are not interchangeable:</p><ul><li>Exclude terms, which are the strongest cure for a result set dominated by quote tweets and replies.</li><li>Keyword match type, switchable between exact phrase, contains, and partial.</li><li>A floor on likes, retweets, or impressions, which sorts a crowded day by whether anyone reacted at all.</li><li>Language, which removes whole regions of unrelated conversation in a click.</li><li>Media type, narrowed to images, videos, or text only, for a post you recall by its picture rather than its wording.</li></ul><p><strong>Choose the axis you are most confident about, not the one that sounds most precise.</strong></p><p>If you remember the post traveled, a minimum like count is worth more than a tighter window. If you remember the wording, exact-phrase matching earns its place. If you remember it had a picture, the media filter does more work than either.</p><p>The failure mode is stacking. Four filters means four separate memories all being correct at once, and memory does not perform that reliably.</p><p>Each control you add is another chance to exclude the thing you are hunting.</p><h3>Why the engagement floor is the most underused control</h3><p>Set a minimum like count and you are no longer reading a day, you are reading the part of that day anyone reacted to.</p><p>Say a launch day produced four thousand matching posts. A minimum of twenty-five likes usually leaves a few dozen rows, and if you remember the post because it traveled, your target is inside that shorter list by definition.</p><p>The move fails in one specific case, and it is worth naming. A post you remember because it was yours, or because it was quiet and personal, may have collected almost no engagement at all.</p><p>Set a floor on that search and you will exclude the exact row you came for.</p><p>So the floor belongs on searches where the memory itself is public. Use it when other people made the post memorable, and leave it off when you did.</p><p>There is a control worth turning off rather than on. Replies and links can both be excluded, and on a viral day excluding replies alone often cuts the result set by more than any date adjustment would.</p><h3>Anchor the window to a public event, not to your memory</h3><p>Here is the move that changed my hit rate more than any filter.</p><p>Stop asking when you posted. Start asking what you were posting about, then date that instead.</p><p>A product launch has a press date. A conference has a schedule. An outage has a status page.</p><p>A sports result, an election, a keynote, a platform policy change: all of them carry a public timestamp that somebody else has already verified, and all of them are more reliable than your recollection of a Tuesday.</p><p>The technique is to date the event precisely and then place your window around it asymmetrically. Reaction posts cluster in the 48 hours after a public moment, not before it.</p><p>So a window from the event date to three days later beats a symmetric window centered on it.</p><p>X’s own date-range interface has supported this since 2014, when the From and To fields arrived in the search screen.</p><p>Before that, you had to know the underlying operators to do it at all, as TechCrunch reported in <a href="https://techcrunch.com/2014/04/01/twitter-now-lets-you-easily-search-for-tweets-by-date-range/">Twitter now lets you easily search for tweets by date range</a>. The capability is old. The discipline around it is what people skip.</p><p>It works the other way too. The same anchoring logic, applied to an account’s behavior rather than a single post, appears in <a href="https://circleboom.com/blog/how-can-i-find-out-the-exact-date-that-i-started-following-someone-on-twitter-x/">how can I find out the exact date that I started following someone on Twitter</a>.</p><h3>What the results table tells you that the date field cannot</h3><p>The date field is an input. The table is evidence, and it carries more than most people read.</p><p>Each row prints impressions, likes, retweets, quotes, bookmarks, and replies alongside a Created At value showing when the post went up, down to the hour, plus how old it is now. That combination is what lets you separate the row you want from four near-identical neighbours.</p><p>Each Posts cell links straight out to the tweet itself on X, which settles anything the table cannot.</p><p>Circleboom retrieves all of this as <a href="https://docs.x.com/enterprise/customer-directory#enterprise-customers-listing">one of the companies on X’s official Enterprise developers list</a>. That access is what makes a window from last year behave the same as a window from last week.</p><p>Native search leans toward recent activity. You can still tailor results to a date range there, and X’s page on <a href="https://help.x.com/en/using-x/x-advanced-search">how to use advanced search</a> walks through the fields.</p><p>The depth behind that range is a different question from the interface in front of it.</p><h3>Read the rows, then pivot to the people</h3><p>One pivot is worth knowing before you start. A button marked Display Profiles of this search turns the tweet list into the accounts that produced it. When you are hunting a post you half-remember, recognizing an author is often faster than recognizing a sentence.</p><p>Open a <a href="https://circleboom.com/twitter-management-tool/twitter-search-tool/twitter-advanced-search/twitter-historical-data">date-scoped tweet search on X</a> <em>(this link takes you to Circleboom, outside Medium)</em>. The two views are one result set seen from different ends.</p><p>Once the right accounts are identified, they can be filed with the <a href="https://circleboom.com/twitter-management-tool/twitter-search-tool/twitter-list-manager">Twitter List Manager</a> for ongoing monitoring.</p><p>Or they can leave the platform entirely through <a href="https://circleboom.com/twitter-management-tool/twitter-search-tool/export-twitter-accounts">export Twitter accounts</a>, which turns a date hunt into a file you can work from later.</p><h3>How to find tweets from a specific date on Twitter with a second filter attached</h3><p>The order that keeps the window honest.</p><h3>Set the container wide enough to contain the answer</h3><ul><li><a href="https://circleboom.com/twitter-management-tool"><strong>Open Circleboom Twitter</strong></a> <em>(this link takes you to Circleboom, outside Medium)</em> and attach the account whose archive you want to reach.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*CP3_S0jo5MMchlad.png" /></figure><ul><li><strong>Move into the Advanced X Search menu</strong> and select Historical Tweet Search.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/915/0*rM0Lre2gws78f0k9.png" /></figure><ul><li><strong>Set a custom date range that brackets the event or day</strong>, generously rather than tightly.</li></ul><h3>Attach the one control you trust</h3><ul><li><strong>Write the search in plain language</strong>, then keep or edit one of the AI variations the interface offers.</li><li><strong>Add a single narrowing control</strong>, whichever of exclude terms, match type, engagement minimum, language, or media type you are most sure of.</li><li><strong>Choose a tweet count that matches the crowd</strong>, since the collection is sized in GetTweetTokens and a busy day will happily spend all of them.</li></ul><h3>Confirm before you act on the result</h3><ul><li><strong>Sort on Created At</strong> and read the clock time, not only the date.</li><li><strong>Follow the outbound link</strong> in the Posts column to the original tweet when the row looks right.</li></ul><p>That sequence works because it separates the two jobs a search is doing. The date decides which slice of the archive is even eligible, and the single control decides what survives inside that slice.</p><p>Collapse them into one step and every miss becomes ambiguous, because you cannot tell whether the window was wrong or the query was.</p><p>There is one more reason to keep the pair separate. A wide window with a single control can be tightened on either side afterwards, and a search you can tighten is a search you can save.</p><p>Quick recap: open the menu, set a generous range, attach one control, size the collection, then confirm on Created At.</p><p><strong>A short demo of the keyword-first move: how a search built on what people said resolves into the accounts that said it.</strong></p><p><a href="https://www.youtube.com/watch?v=hPPu49XaLGc">https://www.youtube.com/watch?v=hPPu49XaLGc</a></p><p>The same keyword-plus-window pairing drives ongoing tracking rather than one-off retrieval, which is the job of the <a href="https://circleboom.com/twitter-management-tool/twitter-search-tool/keyword-hashtag-tracker">keyword and hashtag tracker</a>.</p><p>The broader landscape of what search on the platform can actually reach is mapped in <a href="https://circleboom.com/blog/twitter-search-engine/">Twitter search engine</a>.</p><p>And readers coming at this with no account of their own should start from <a href="https://circleboom.com/blog/twitter-search-without-an-account/">Twitter search without an account</a>, because the constraints are different before you log in.</p><h3>The short version</h3><p>A date range narrows the archive. It does not find a post. Treating it as the whole query is the single most common reason a correct window returns the wrong screen.</p><p>The searches that land do three things:</p><ul><li>They pair a generous window with exactly one control you are confident about.</li><li>They anchor that window to a public event rather than a personal memory.</li><li>They confirm the hit on the Created At timestamp instead of on the date that was typed.</li></ul><p>The searches that fail almost always did the opposite: a narrow window, no second axis, and a conclusion drawn from an empty screen.</p><p>If your goal is removal rather than retrieval, that is a separate workflow, and <a href="https://circleboom.com/blog/how-to-delete-all-retweets-of-specific-dates-account-from-your-twitter/">how to delete all retweets of specific dates</a> covers it.</p><p>Begin with the window wide and one filter attached:</p><p><strong>→ </strong><a href="https://circleboom.com/twitter-management-tool/twitter-search-tool/twitter-advanced-search/twitter-historical-data"><strong>find a tweet from one exact date on X</strong></a> <em>(this link takes you to Circleboom, outside Medium)</em></p><h3>Two questions worth settling first</h3><h3>Should I search a single day or a range?</h3><p>A range, almost always. A single day is worth setting only when you hold a verified public timestamp for the event you are searching around. Even then, a two-day window costs little and covers a post that landed either side of midnight.</p><h3>What if I remember the picture but not a single word of the post?</h3><p>Set the media type to images and leave the date range wide. It is the one control that never asks you to recall any wording, which makes it the strongest thing you can attach to a window when the memory you are working from is visual rather than verbal.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=b43545bd80f9" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Delete all reposts on X (Twitter): why one click is really a queue]]></title>
            <link>https://medium.com/@mariadunhum/delete-all-reposts-on-x-twitter-why-one-click-is-really-a-queue-03f9079d0d62?source=rss-d83cbad5dc9e------2</link>
            <guid isPermaLink="false">https://medium.com/p/03f9079d0d62</guid>
            <category><![CDATA[retweets]]></category>
            <category><![CDATA[social-media-management]]></category>
            <category><![CDATA[twitter]]></category>
            <category><![CDATA[account-cleanup]]></category>
            <dc:creator><![CDATA[Maria Dunham]]></dc:creator>
            <pubDate>Thu, 22 Oct 2026 12:56:01 GMT</pubDate>
            <atom:updated>2026-10-22T12:56:01.594Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*EFdu3ebh-Yq3DZGKWB5syQ.jpeg" /></figure><p>I stopped believing the “one click” framing the first time I read what X’s API actually lets a tool do in a single request.</p><p>Nothing about a slow repost cleanup is broken. That is simply what the operation is underneath, and knowing it changes how you plan the job.</p><blockquote><strong><em>What you get from reading this.</em></strong></blockquote><ul><li>The reason no tool can remove your reposts in a single request.</li><li>The rate-limit behavior that decides how long your run takes.</li><li>The one thing I got wrong about the free tier before I checked the numbers.</li></ul><blockquote><em>Circleboom clears reposts from your X account through sanctioned API access rather than a browser script. Start with </em><a href="https://circleboom.com/twitter-management-tool/delete-retweets"><em>delete all reposts on X (Twitter)</em></a><em> </em>(this link takes you to Circleboom, outside Medium)<em>.</em></blockquote><h3>What actually happens when a repost disappears</h3><p>Each removal is its own API call.</p><p><a href="https://docs.x.com/x-api/users/unrepost-post">X’s unrepost API</a> is a DELETE against a path that carries exactly one post ID, and the successful response is a single object saying the post is no longer reposted.</p><p>One ID in, one confirmation back. There is no batch parameter, no array of IDs, no “clear all” flag anywhere in the shape of that request.</p><p>The authorization is worth noticing too. The same user-context token that lets an app repost on your behalf is what lets it undo one, so removal is not a special privileged operation.</p><p>It is the ordinary write permission, pointed the other way.</p><p>So a tool offering to delete all your reposts is doing what you would do by hand, only faster and without losing its place. It holds the list, walks it, and issues one call per entry.</p><p>That is not a criticism of the tools.</p><p>It is the only thing the platform allows, and it explains almost every strange thing people report about bulk repost cleanup: the progress bar, the pause, the browser window you are told not to close.</p><h3>The number nobody makes you look at first</h3><p>Most people underestimate their repost count by an order of magnitude. Reposting costs nothing in the moment, leaves no draft, and is forgotten within a day, so the mental tally never gets updated.</p><p>Then the list loads and it is 1,400 entries long.</p><p>I would rather you saw that number before you commit an evening to this.</p><p>Launch <a href="https://circleboom.com/twitter-management-tool/delete-retweets">bulk repost deletion</a> <em>(this link takes you to Circleboom, outside Medium)</em>, connect the account, and read the count before you touch anything else.</p><p>The count also tells you which cleanup you are actually doing. A few hundred is a tidy-up. Several thousand is an identity decision, and it deserves a different conversation with yourself than “this will take ten minutes.”</p><p>Worth knowing what a repost even is on the current platform, incidentally: <a href="https://circleboom.com/blog/repost-a-tweet/">repost a tweet</a> covers the naming change and what carried over from the retweet era.</p><p><strong>Phone-screen demo: the same paced repost queue running from the X mobile app instead of a desktop browser.</strong></p><p><a href="https://www.youtube.com/watch?v=PcICc7pKTYY">https://www.youtube.com/watch?v=PcICc7pKTYY</a></p><h3>How to delete all reposts on X (Twitter) from one queue</h3><p>Clearing every repost from an X account is a four-part loop: authorize, load, select, confirm.</p><p>Circleboom holds the whole repost record in one reviewable table, so the choosing happens off your timeline entirely.</p><p>The removals then run as a paced queue of individual calls, which is why the tab stays open until the run reports itself finished.</p><h3>Connect the account and load the history</h3><ul><li><a href="https://circleboom.com/twitter-management-tool"><strong>Open Circleboom Twitter</strong></a> <em>(this link takes you to Circleboom, outside Medium)</em> and authorize the handle that made all those reposts.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/500/0*-0GBjibSxB-XRZaF.png" /></figure><ul><li><strong>Go to the Essential Toolbox menu</strong>, which is where the delete utilities live next to export and list management.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*aMpScNWOnZxOPVEs.png" /></figure><ul><li><strong>Wait for the repost history to load</strong> as a list. Each row carries its context details, which is what makes review possible at all.</li></ul><h3>Select, confirm, and let the queue run</h3><ul><li><strong>Pick the entries to remove</strong>, singly or in groups. Selection is not deletion, and nothing moves until you confirm.</li><li><strong>Confirm the removal</strong> and leave the tab open. The queue works through your selection one call at a time against your connected account.</li><li><strong>Check the free daily allowance</strong> if you are not on a paid plan, because it caps any single day’s run at fifty entries.</li></ul><p>Each of those steps exists because of the constraint above it.</p><p>The list has to load before you can review, review has to happen before selection, and the confirmation is the point where a reversible choice becomes an irreversible one.</p><p>The queue is slow for a reason worth wanting.</p><h3>What to do in the ten minutes before you start the queue</h3><p>Two preparations save people from the version of this job that goes wrong, and both are boring.</p><p>Request your X data archive first, from X’s own settings.</p><p>X quotes a day or two for the file to be built, so asking for it now costs you nothing and hands you a permanent local record of everything the account held.</p><p>Deletion is irreversible, and the archive is the only copy you will ever have of what you removed.</p><p>Then decide the scope out loud.</p><p>Are you clearing every repost you ever made, or every repost from before a specific date, or every repost of one particular account?</p><p>Those are three different jobs with three different finishing conditions, and starting a queue without knowing which one you are running is how people stop halfway.</p><p>If the account belongs to a business, add a third step: tell whoever else has a stake in it. A profile that loses two thousand entries overnight is a visible change, and it is a much easier conversation before than after.</p><p><strong>None of this takes more than ten minutes, and all of it is unrecoverable if you skip it.</strong></p><h3>The rate limits that set your run time</h3><p>X publishes its limits and they are not generous for this kind of work.</p><p>The <a href="https://docs.x.com/x-api/fundamentals/rate-limits">X API rate limits</a> documentation describes windows measured in 15-minute blocks and applied per user token.</p><p>Spend a window and the next request comes back as a 429, with a header naming the moment the window resets.</p><p>Read that back in terms of your afternoon. A client cannot simply fire a thousand deletions and walk away. It fires what the window allows, waits for the reset, and continues.</p><p>That is the whole mechanism behind a bulk repost run taking real time on a big account. Nothing is stuck. The queue is holding.</p><p>If you have ever wondered why the same account can repost freely but clear reposts slowly, the answer is that both operations are drawing on the same kind of user-context budget.</p><p><a href="https://circleboom.com/blog/can-i-retweet-my-tweet-more-than-once/">Can I retweet my tweet more than once</a> gets at the same underlying limits from the creation side.</p><h3>Where I was wrong about the free tier</h3><p>I assumed the free allowance was a soft nudge toward a paid plan and that a determined person could work around it in an afternoon. That is not what it is.</p><p>Fifty a day, resetting each morning, with no way to talk your way past it on that plan. For a 1,400-entry history that works out at twenty-eight sittings.</p><p>What I had wrong was the framing, not the number.</p><p>Twenty-eight sittings sounds like a punishment until you notice the alternative it is being compared against, which is not “instant” but “never,” because almost nobody clears fourteen hundred reposts by hand.</p><p>Open <a href="https://circleboom.com/twitter-management-tool/delete-retweets">remove all your reposts on X</a> <em>(this link takes you to Circleboom, outside Medium)</em> and treat the daily ceiling as a schedule rather than an obstacle.</p><p>There is a genuine escape hatch for large histories, and it is not the API at all. The archive route works from a file instead, tweet.js, pulled out of your own X data download.</p><p>Because it is a file rather than a live query, it reaches history that the 3,200-post ceiling keeps out of reach.</p><p>The <a href="https://circleboom.com/twitter-management-tool/delete-old-tweets-archive/twitter-archive-eraser">Twitter archive eraser</a> path is the one to look at when the daily math stops working.</p><h3>What the queue leaves behind on other people’s timelines</h3><p>Undoing a repost leaves the original post completely alone.</p><p>What comes off is your boost of it, taken from your own timeline and activity record, while the post itself sits where it always did on the account that wrote it.</p><p>Content other accounts created also survives. Quotes, replies and reposts made by other people are theirs, not yours, and no cleanup you run touches them.</p><p>That boundary is worth holding onto, because it sets a realistic expectation. You are cleaning your own record, not editing the conversation.</p><p>It also means a repost cleanup is a much smaller act than deleting your own posts, and I think people misjudge that in both directions.</p><p>Some hesitate over reposts as though removing one is a public statement, when nobody except the original author has any way to notice.</p><p>Others assume clearing reposts wipes a controversy, when the quotes and screenshots built on top of it are untouched and always will be.</p><p>The honest description is narrower than either. You are removing your name from the list of accounts that boosted something, on your own profile, from that day forward.</p><p>A useful side effect is that your reposts stop being a research trail for anyone auditing your account.</p><p>To see what that trail looks like from the outside first, <a href="https://circleboom.com/blog/how-do-i-find-my-retweets-of-a-certain-account/">find my retweets of a certain account</a> shows how narrow the view can get.</p><h3>Is bulk repost deletion safe on X?</h3><p>Yes, when the removals go through authorized access.</p><p>Circleboom operates as a <a href="https://docs.x.com/enterprise/customer-directory#enterprise-customers-listing">verified Enterprise partner of X</a>, which is why a long run of deletions reads to the platform as a legitimate client rather than as automated abuse.</p><p>The practical safety points are ordinary ones. Deletion is permanent, so download your archive first if you want any record of what was there.</p><p>Large operations take time, so do not interrupt one mid-run.</p><p>And if the account is shared with a team or a client, tell them before you clear a few thousand entries out of a public profile.</p><p>On a business account, the safer first move is often a narrower one.</p><p>Working through <a href="https://circleboom.com/twitter-management-tool/delete-all-tweets/delete-tweets-by-date">delete tweets by date</a> lets you clear one defined period instead of the whole record.</p><p>Cross-posting habits leave their own trail as well, which <a href="https://circleboom.com/blog/how-do-i-automatically-repost-from-twitter-to-instagram/">automatically repost from Twitter to Instagram</a> traces.</p><h3>Which constraint is binding on your account</h3><p>Reposts come off one at a time because that is the only operation X offers, and every bulk tool is a queue built on top of that single-call reality.</p><p>Rate limits set the pace, the free plan sets a 50-a-day ceiling, and the archive route is the way past both when the history is genuinely large.</p><p>Knowing which of those three constraints is binding on your account is the entire planning problem. Everything else is clicking.</p><p>The value at the end is not a tidier feed.</p><p>It is a profile where every visible post is yours, which is a different thing to hand someone than a wall of other people’s opinions you happened to agree with once.</p><p><strong>→ </strong><a href="https://circleboom.com/twitter-management-tool/delete-retweets"><strong>delete every repost on your X account</strong></a> <em>(this link takes you to Circleboom, outside Medium)</em></p><h3>What people ask about repost cleanups</h3><h3>How long does a full repost cleanup actually take?</h3><p>Divide your repost count by the daily allowance on your plan, then add waiting time for rate-limit windows during each run. On the free plan that means 50 per day, so a 600-entry history is roughly a fortnight of short sessions rather than one long evening.</p><h3>Does removing a repost hurt the original poster?</h3><p>Their post is untouched and stays live on their own timeline. The only change is that their repost count drops by one, and on any post with real reach that is statistically invisible.</p><h3>Can I keep some reposts and clear the rest?</h3><p>Yes. Selection is manual by default, and the archive wizard adds a Protect List that exempts whatever you put in it from an otherwise total sweep.</p><h3>What happens if my connection drops mid-queue?</h3><p>The removals already confirmed by X stay removed, because each one was its own completed call. The remainder simply never ran, so the recovery is to reconnect, reload the list, and start a fresh selection from what is left.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=03f9079d0d62" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>