<?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 Simple Actor on Medium]]></title>
        <description><![CDATA[Stories by Simple Actor on Medium]]></description>
        <link>https://medium.com/@simple.actor.apify?source=rss-7e488d384d8b------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/0*nMEIeTdCKk2AVuGT</url>
            <title>Stories by Simple Actor on Medium</title>
            <link>https://medium.com/@simple.actor.apify?source=rss-7e488d384d8b------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Wed, 18 Nov 2026 03:02:57 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@simple.actor.apify/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[How to get LinkedIn follower counts for a list of companies (15 European airlines)]]></title>
            <link>https://medium.com/@simple.actor.apify/how-to-get-linkedin-follower-counts-for-a-list-of-companies-15-european-airlines-1550b51189ce?source=rss-7e488d384d8b------2</link>
            <guid isPermaLink="false">https://medium.com/p/1550b51189ce</guid>
            <category><![CDATA[linkedin]]></category>
            <category><![CDATA[b2b]]></category>
            <category><![CDATA[javascript]]></category>
            <category><![CDATA[data]]></category>
            <category><![CDATA[web-scraping]]></category>
            <dc:creator><![CDATA[Simple Actor]]></dc:creator>
            <pubDate>Thu, 05 Nov 2026 00:00:58 GMT</pubDate>
            <atom:updated>2026-11-14T05:38:45.726Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*52GBN4Yj0jo9fFIn.png" /></figure><p><em>I wrote this with an AI assistant. The runs, numbers and code are real, and I checked them.</em></p><p>British Airways has 1,298,508 followers on LinkedIn. Aer Lingus has 159,655, about an eighth as many. Yet Aer Lingus’s typical post drew more reactions: a median of 435.5, against 349 for British Airways.</p><p>I only saw that because I had both numbers in one table, for fifteen airlines, read on the same morning. That’s the job here: a list of companies in, one row per company out, with the follower count, the declared size, the headquarters and the latest posts.</p><h3>Fifteen rows, one request each</h3><p>I build and run a LinkedIn company scraper on Apify. It reads the page LinkedIn shows a logged-out visitor at linkedin.com/company/&lt;slug&gt;, so there&#39;s no login, no cookie and no API key. One page a company carries the record and its latest posts, about ten of them.</p><p>I gave it fifteen European airlines on the morning of 5 November, with outputFormat set to company so each airline came back as one row with its posts nested inside. In Node:</p><pre>import { ApifyClient } from &#39;apify-client&#39;;<br>const client = new ApifyClient({ token: process.env.APIFY_TOKEN });<br>const run = await client.actor(&#39;simple.actor/linkedin-company-posts&#39;)<br>  .call({ companies: [&#39;british-airways&#39;, &#39;lufthansa&#39;, &#39;klm&#39;],<br>          outputFormat: &#39;company&#39; });<br>const { items } = await client.dataset(run.defaultDatasetId)<br>  .listItems();<br>for (const c of items) {<br>  if (c.error) { console.warn(c.companySlug, c.error); continue; }<br>  console.log(c.companyName, c.followers, c.companySize);<br>}</pre><p>The first run took 86 seconds for fifteen slugs. Twelve were read. Three came back as not_found rows. Between them, the fifteen airlines I ended up with have 8,187,096 followers.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*UChwVikA3aNC7P8a.png" /></figure><p><em>The fifteen company records, from British Airways’ 1,298,508 followers down to Aer Lingus’s 159,655.</em></p><p>Each row also carries the tagline, the full About-us text, the logo, the website, the industry, the organisation type, the specialities and a postal address. Two size columns sit side by side and mean different things. companySize is the band the company declares, &quot;10,001+ employees&quot; for British Airways. employeesOnLinkedIn is how many profiles LinkedIn ties to it: 32,981.</p><h3>The slug is not the brand name</h3><p>Three of my first fifteen guesses were wrong in the safe way. virgin-atlantic, wizz-air and norwegian have no company page, so each came back as a not_found row and wasn&#39;t charged. The real ones are virgin-atlantic-airways, wizzair and norwegian-air-shuttle.</p><p>The fourth was wrong in the expensive way. sas is a real page. It belongs to SAS the software company in Cary, North Carolina, with 992,627 followers, and it came back as a normal, charged row. If I&#39;d only been reading the follower column, Scandinavian Airlines would have ranked third, ahead of Ryanair.</p><p>So I ran a second batch of eight candidate slugs. That one took 33 seconds. sas-scandinavian-airlines turned out to be a lookalike page with 1,059 followers, filed under Hospitals and Health Care. flysas is the airline: 208,563 followers, Airlines and Aviation, Solna.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*jv7cavT5H4u4oQ9n.png" /></figure><p><em>Three slugs, three different companies. The industry and website columns are what gave the first two away.</em></p><p>The check that catches this costs nothing: look at industry and website before you trust the follower count.</p><h3>Followers don’t predict reactions</h3><p>With the posts in the same row, I divided each airline’s median reactions per post by its followers. I left out reposts, because their reactions belong to whoever wrote the original, and posts where LinkedIn showed no count.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*LTjC2IR368fSO7Cy.png" /></figure><p><em>Reactions on a typical post for every 1,000 followers, from 158 posts.</em></p><p>Aer Lingus gets 2.73 reactions per 1,000 followers. Ryanair, with 920,161 followers, gets 0.16, seventeen times fewer. The two largest pages, British Airways and Lufthansa, sit near the bottom at 0.27 and 0.30.</p><p>How often they post varies just as much. KLM’s latest 10 posts cover 9.8 days, about seven a week. Aer Lingus’s 11 go back 149 days, and Virgin Atlantic’s 10 go back 356 days, nearly a year. A page that posts once a month may simply be posting its best material. I can’t tell that from ten posts, but I can see which pages to look at more closely.</p><h3>What it cost</h3><p>The scraper charges $0.0010 per company read on Apify’s Free and Starter plans, $0.0009 on Scale and $0.00085 on Business and above. Posts are free, and a company that can’t be read isn’t charged.</p><p>Across both runs, 17 companies were read: $0.017 on the Free plan. Two of those were the wrong SAS pages, so the fifteen airlines themselves come to $0.015. The six not_found slugs cost nothing.</p><h3>Where the rows need a second look</h3><p>The names arrive as LinkedIn writes them, and once as LinkedIn encodes them. Ryanair’s came back as Ryanair - Europe&amp;#39;s Favourite Airline, with the apostrophe still an HTML entity. Decode the names before they go into a spreadsheet.</p><p>website is whatever the company typed into LinkedIn. For thirteen airlines that was the main site. KLM&#39;s is a short link, klmf.ly/R05uLo, and easyJet&#39;s is its careers site. Three airlines, British Airways, SWISS and Wizz Air, give no founding year, so foundedYear is null.</p><p>The posts are only the latest ten or so: between 10 and 12 per airline here, because that’s what LinkedIn shows a logged-out visitor, and there’s no way to page further. Of the 158 posts, 18 had no comment count shown and 8 of the airlines’ own posts had no reaction count. Those are null, not zero. There’s no comment text and no breakdown of reactions by type.</p><h3>Make it your list</h3><p>The example repo, <a href="https://github.com/simple-actor/linkedin-company-followers-scraper">linkedin-company-followers-scraper</a>, takes any number of slugs and prints each company’s followers, size, headquarters, posts per week and median reactions, then writes them to companies.csv, in Node and in Python. Put in your own list: the accounts your sales team is chasing, the competitors in your trade, the suppliers in your town. Take each slug from the address bar rather than guessing, and check industry on the first run. The scraper is at <a href="https://apify.com/simple.actor/linkedin-company-posts">apify.com/simple.actor/linkedin-company-posts</a>.</p><p><em>Originally published at </em><a href="https://simple-actor.github.io/linkedin-follower-counts-list-of-companies-european-airlines/"><em>https://simple-actor.github.io</em></a><em> on November 5, 2026.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=1550b51189ce" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[How to get exact view counts for a YouTube channel’s videos (120 videos from NASA, ESA, SpaceX and…]]></title>
            <link>https://medium.com/@simple.actor.apify/how-to-get-exact-view-counts-for-a-youtube-channels-videos-120-videos-from-nasa-esa-spacex-and-b6ba0e83f199?source=rss-7e488d384d8b------2</link>
            <guid isPermaLink="false">https://medium.com/p/b6ba0e83f199</guid>
            <category><![CDATA[data]]></category>
            <category><![CDATA[youtube]]></category>
            <category><![CDATA[web-scraping]]></category>
            <category><![CDATA[data-science]]></category>
            <category><![CDATA[javascript]]></category>
            <dc:creator><![CDATA[Simple Actor]]></dc:creator>
            <pubDate>Wed, 04 Nov 2026 00:00:43 GMT</pubDate>
            <atom:updated>2026-11-13T06:21:46.588Z</atom:updated>
            <content:encoded><![CDATA[<h3>How to get exact view counts for a YouTube channel’s videos (120 videos from NASA, ESA, SpaceX and Blue Origin)</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*549YIUKMMhPNuK-j.png" /></figure><p><em>I wrote this with an AI assistant. The runs, numbers and code are real, and I checked them.</em></p><p>SpaceX’s channel page says its Starship SN8 recap has 9.5M views. The exact count, read from the video’s own page a few minutes later, was 9,599,850. That’s 99,850 views the label leaves out, on one video.</p><p>It matters as soon as you put two channels side by side. YouTube’s channel pages round every count down, and they date every video with “1y ago” or “2mo ago”. For a quick look that’s fine. For a spreadsheet that ranks videos, works out a like rate or plots uploads by month, it isn’t.</p><h3>Two runs, same 120 videos</h3><p>I build and run a YouTube channel scraper on Apify. It reads a channel’s latest videos, up to 30, from the page YouTube serves a logged-out visitor. With includeVideoDetails on it also opens each video&#39;s page for the exact view count, the like count, the description and the publish date.</p><p>I picked four public space channels, @NASA, @EuropeanSpaceAgency, @SpaceX and @blueorigin, and ran them twice on the morning of 4 November: once with details off, once with them on. Here’s the second in Node:</p><pre>import { ApifyClient } from &#39;apify-client&#39;;<br>const client = new ApifyClient({ token: process.env.APIFY_TOKEN });<br>const run = await client.actor(&#39;simple.actor/youtube-channel-videos&#39;)<br>  .call({ handles: [&#39;@NASA&#39;, &#39;@EuropeanSpaceAgency&#39;, &#39;@SpaceX&#39;,<br>          &#39;@blueorigin&#39;], maxVideos: 30, includeVideoDetails: true });<br>const { items } = await client.dataset(run.defaultDatasetId)<br>  .listItems();<br>for (const v of items) {<br>  if (v.error) continue; // a channel that couldn&#39;t be read<br>  console.log(v.channelName, v.viewCount, v.likeCount, v.title);<br>}</pre><p>The details-off run took 24 seconds and returned 120 videos. The details-on run took 4 minutes, one extra request per video, and returned the same 120, every one with an exact view count and a like count.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*TeIOkPbINc6P3y7a.png" /></figure><p><em>One video from each channel, with what the channel page said next to what the video’s page said.</em></p><h3>How far off the rounded numbers are</h3><p>119 of the 120 rounded counts were lower than the exact ones. The one that matched had 768 views, too few to round. The median gap was 518 views, 0.8% of the video’s count. On the biggest videos it ran to tens of thousands: SN8 was short 99,850, ESA’s “Astronaut Pikachu gets a private tour of the ISS” showed 1.7M for 1,799,423, and SpaceX’s “Starship — The Holy Grail of Rocketry” showed 2.8M for 2,898,289. The largest gap in percent was on a Blue Origin video, 11,000 for 11,977, 8.2%.</p><p>Rounding also makes different videos look the same. The channel pages gave the 120 videos 106 different view counts. The video pages gave 120. Two SpaceX videos both read 2.3M, and they were 70,677 views apart.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*qWbqShMQBDA6dE2O.png" /></figure><p><em>Every rounded figure in the run that covered more than one video.</em></p><p>Dates are worse. A channel page gives “1y ago” or “2mo ago”, so without the video’s page the scraper has to turn that into a date and flag it as approximate. Against the real publish days, those dates were off by a median of 19 days. 29 of 120 were off by more than a month, and five SpaceX videos by close to a year: “Starship’s Fifth Flight Test” reads “1y ago”, which works out to 3 November 2025, and was published on 6 November 2024.</p><h3>What the exact figures show</h3><p>With likes on every row you can compare channels that are nothing alike in size. NASA has 15.1M subscribers and Blue Origin 311,000, but per view Blue Origin’s audience likes more. The median video on SpaceX got 3.34 likes per 100 views, Blue Origin 3.24, NASA 2.48 and ESA 2.22.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*Bs0WESydNhsDxj-h.png" /></figure><p><em>Exact likes divided by exact views for every video in the run.</em></p><p>The exact counts also show how lopsided a channel can be. ESA’s 30 latest videos had 3,109,187 views between them, and 58% of those were the Pikachu video. Its next best had 124,432. NASA’s top video was its Moon Base update, 1,006,881 views and 11,716 likes. SpaceX’s was the five-year-old SN8 recap.</p><h3>What it cost</h3><p>The listing charges $0.0004 for each video returned and another $0.0008 for each video whose page was read for the exact figures, on the Free and Starter plans. The higher plans pay less. The details-on run was 120 × $0.0004 + 120 × $0.0008 = $0.144. The details-off run was 120 × $0.0004 = $0.048. A channel that can’t be read costs nothing.</p><p>So the exact figures are two thirds of the price. If all you want is to know what a channel posted this week, turn them off, add onlyVideosNewerThan: &quot;7 days&quot;, and pay for the cheap read. Turn them on when the numbers go into a report.</p><h3>Thirty videos, and a day rather than a time</h3><p>Thirty is the most a channel page shows, so it’s the most you get. For NASA that’s April to October 2026 out of the 6,100 videos YouTube lists for it. SpaceX posts far less, and its 30 go back to October 2020. Every row carries channelVideoCount so you can tell which case you&#39;re in.</p><p>Only 35 of the 120 rows had the publish time and a category. The other 85 had the publish day at midnight UTC, flagged with publishedAtIsApproximate: true, and category: null. YouTube sends the time and category in data it holds back from most requests from datacenter addresses. The day is right, and for a monthly chart it&#39;s enough. Subscriber counts are YouTube&#39;s rounded figure on every row, 15.1M for NASA, flagged the same way.</p><p>9 videos came back with no description. And the handle has to be the channel’s real one. My first attempt used @esa, which isn’t ESA’s handle: it came back as a not_found row, the run still succeeded, and the other three channels were read and charged as normal.</p><h3>Make it your channels</h3><p>The example repo, <a href="https://github.com/simple-actor/youtube-channel-stats-scraper">youtube-channel-stats-scraper</a>, takes any number of channels, prints each one’s top videos and median likes per 100 views, and writes every video to videos.csv, in Node and in Python. Swap in your own channel and the three you compete with, a set of creators you&#39;re thinking of sponsoring, or the brand channels in your trade, and run it once a month. Store channelId, not the handle: a handle can change, the id can&#39;t. The scraper is at <a href="https://apify.com/simple.actor/youtube-channel-videos">apify.com/simple.actor/youtube-channel-videos</a>.</p><p><em>Originally published at </em><a href="https://simple-actor.github.io/exact-youtube-view-counts-channel-videos-nasa-spacex/"><em>https://simple-actor.github.io</em></a><em> on November 4, 2026.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=b6ba0e83f199" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[How to see every Google ad a competitor started this month (136 new ads from three mattress brands)]]></title>
            <link>https://medium.com/@simple.actor.apify/how-to-see-every-google-ad-a-competitor-started-this-month-136-new-ads-from-three-mattress-brands-a7a7d83e936b?source=rss-7e488d384d8b------2</link>
            <guid isPermaLink="false">https://medium.com/p/a7a7d83e936b</guid>
            <category><![CDATA[data]]></category>
            <category><![CDATA[advertising]]></category>
            <category><![CDATA[marketing]]></category>
            <category><![CDATA[javascript]]></category>
            <category><![CDATA[web-scraping]]></category>
            <dc:creator><![CDATA[Simple Actor]]></dc:creator>
            <pubDate>Tue, 03 Nov 2026 00:00:04 GMT</pubDate>
            <atom:updated>2026-11-11T05:35:13.296Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*nUjja9dvt0u7xniz.png" /></figure><p><em>I wrote this with an AI assistant. The runs, numbers and code are real, and I checked them.</em></p><p>“Casper® King for a Queen Sale. Up to $1,000 off on Mattresses. Ends Nov 3.”</p><p>That’s one of Casper’s Google search ads, as Google archived it. I read it on 3 November. Google first showed it on 22 October, the same day Casper started nine other new ads and the same day its Friends &amp; Family sale ads stopped running.</p><p>You can’t see that from a search box. Google shows you the ads it picked for you, today. Its Ads Transparency Center keeps the rest: every ad an advertiser runs, with the day it was first shown and the day it was last shown. Those two dates are enough to rebuild a competitor’s month.</p><h3>What I asked for</h3><p>I build and run a scraper for the Ads Transparency Center on Apify. You give it a website, an advertiser ID or a name, and it returns one row per ad: format, first and last shown, days shown, a picture of the ad or a link to Google’s preview, and Google’s own count of how many ads it holds.</p><p>I picked three mattress brands that sell online, casper.com, purple.com and saatva.com, and ran them twice. The first run asked for every ad shown in the last 30 days ( shownSince: &quot;30 days&quot;). The second asked only for ads first shown in the last 30 days ( firstShownSince), with includeAdDetails on for the countries each one ran in. Here&#39;s the second, in Node:</p><pre>import { ApifyClient } from &#39;apify-client&#39;;<br>const client = new ApifyClient({ token: process.env.APIFY_TOKEN });<br>const run = await client.actor(&#39;simple.actor/google-ads-transparency&#39;)<br>  .call({ targets: [&#39;casper.com&#39;, &#39;purple.com&#39;, &#39;saatva.com&#39;],<br>          firstShownSince: &#39;30 days&#39;, maxAds: 2000 });<br>const { items } = await client.dataset(run.defaultDatasetId)<br>  .listItems();<br>for (const ad of items) {<br>  if (ad.type !== &#39;ad&#39;) continue; // error rows carry no ad<br>  console.log(ad.target, ad.firstShown.slice(0, 10), ad.format);<br>}</pre><p>The first run returned 695 ads in 14.5 seconds. The second returned 136 in 113 seconds, slower because it makes one extra request per ad for the details. No error rows in either.</p><h3>Purple shows the most, Saatva launched last</h3><p>Purple Innovation showed 461 of its own ads in the 30 days. Saatva showed 118, Casper 81. Google’s own counts for each website, every ad it holds, were 1,000 to 2,000 for Purple, 300 to 400 for Saatva and 600 to 700 for Casper.</p><p>The new ads came in waves, not a trickle. Purple started 79: 25 of them on 13 October and 26 on 22 October. Casper started 24, ten of them on 22 October. Saatva started 12, and every one of them in the last week, between 28 and 30 October.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*Sa8gs1Vh5hei8__4.png" /></figure><p><em>New ads by the day Google first showed them. The three brands’ own ads only.</em></p><p>The formats split differently too. 47 of Purple’s 79 new ads were video and 2 were text. Casper’s 24 were 9 video, 9 text and 6 image. Saatva’s 12 were 7 video and 5 image.</p><p>Saatva keeps old ads running longest. The median ad in its 30-day list had been shown for 363 days, and the oldest was first shown on 6 November 2021 and was still running on 2 November 2026. Casper’s median was 31 days.</p><h3>Casper’s month, ad by ad</h3><p>Casper is the one I could read best, because 14 of its 24 new ads have a picture. Google keeps most text ads as a picture of the ad, so I opened each one.</p><p>The story is in the dates. Eight ads started between 11 and 14 October and all eight were last shown on 22 October. Every one of them I could read was a Friends &amp; Family Sale ad: “Save 20% on The One”, “Save 25% on Snow Max”, “Save Up to 30%”. On 22 October ten new ads started. Five of the pictures say “King for a Queen Sale”, and two of those add “Up to $1,000 off on Mattresses. Ends Nov 3”.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*pxREaJVS7BcQuXfZ.png" /></figure><p><em>Each bar runs from the day Google first showed the ad to the day it last showed it. The labels are what each ad’s picture says.</em></p><p>That’s a sale handover you’d otherwise only catch by searching every day. Run the same lookup weekly and you get it for free: deduplicate on adId, which stays the same between runs, and anything new is a new ad.</p><p>Not every ad on a brand’s website is the brand’s. On the 30-day run, 17 of the 98 ads filed under casper.com came from 17 other advertisers, 7 of Purple’s from 6 others, and 11 of Saatva’s from 9. The website lookup finds every advertiser whose ads point at the site, so if you watch your own domain, that’s the list to check.</p><p>Every new ad ran in the US. One of Saatva’s also ran in Mexico. The median new ad had 3 variations.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*i87p0LHsPOFWXV0K.png" /></figure><p><em>Four rows as they came back from the new-ads run.</em></p><h3>What it cost</h3><p>The listing charges $0.0005 for each ad returned and another $0.0005 per ad for the details. Nothing else, no start fee. The 30-day run was 695 × $0.0005 = $0.3475. The new-ads run was 136 × $0.0005 + 136 × $0.0005 = $0.136. Together, $0.48. A website that can’t be looked up is never charged.</p><p>firstShownSince is the cheap way to get new ads. Only the ads it keeps are charged, but it still has to look down the list: one of Purple&#39;s new ads sat at position 456. So set maxAds high. It&#39;s how far down the scraper looks, not how many you pay for.</p><h3>You get the ad, not always its words</h3><p>Most rows aren’t text. Of the 115 new ads the three brands ran themselves, 31 had a picture and 84 had only previewUrl, a link to Google&#39;s script that draws the ad in a browser. Video ads are all like that. So the scraper gives you the date, the format and a link for every ad, but reading the headline means opening a picture or a preview.</p><p>Some pictures aren’t the ad. I opened all 31. 8 were a blank frame, a frame cut off mid-sentence, or Google’s own “500. That’s an error” page, archived as if it were the ad. The one Saatva ad with a picture was that error page, so none of Saatva’s 12 new ads could be read from a picture.</p><p>The dates are Google’s, in UTC. “Last shown” on an ad that’s still running is the last day Google had recorded when I ran it, here mostly 2 November. And an ad count is not a budget. Purple running nearly six times as many ads as Casper says nothing about who spends more.</p><h3>Make it your competitor</h3><p>The example repo, <a href="https://github.com/simple-actor/competitor-google-ads-scraper">competitor-google-ads-scraper</a>, takes a website and a number of days, prints the new ads by format and by the day they started, and writes them to ads.csv, in Node and in Python. Swap in the site you&#39;re up against (a rival gym, a local estate agent, a software tool in your niche), set the days to 7 and run it every Monday. Add region: &quot;GB&quot; or another country code to the input to see only what ran there. The scraper is at <a href="https://apify.com/simple.actor/google-ads-transparency">apify.com/simple.actor/google-ads-transparency</a>.</p><p><em>Originally published at </em><a href="https://simple-actor.github.io/see-every-google-ad-competitor-started-this-month-mattress-brands/"><em>https://simple-actor.github.io</em></a><em> on November 3, 2026.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=a7a7d83e936b" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[How to see which short videos rank on Google for a keyword (96 for “sourdough starter”, 4 of them…]]></title>
            <link>https://medium.com/@simple.actor.apify/how-to-see-which-short-videos-rank-on-google-for-a-keyword-96-for-sourdough-starter-4-of-them-6f008cbc1ba3?source=rss-7e488d384d8b------2</link>
            <guid isPermaLink="false">https://medium.com/p/6f008cbc1ba3</guid>
            <category><![CDATA[web-scraping]]></category>
            <category><![CDATA[javascript]]></category>
            <category><![CDATA[marketing]]></category>
            <category><![CDATA[video-marketing]]></category>
            <category><![CDATA[seo]]></category>
            <dc:creator><![CDATA[Simple Actor]]></dc:creator>
            <pubDate>Mon, 02 Nov 2026 00:00:10 GMT</pubDate>
            <atom:updated>2026-11-10T03:41:47.410Z</atom:updated>
            <content:encoded><![CDATA[<h3>How to see which short videos rank on Google for a keyword (96 for “sourdough starter”, 4 of them TikToks)</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*tvycFw-gTB6ZtwR8.png" /></figure><p><em>I wrote this with an AI assistant. The runs, numbers and code are real, and I checked them.</em></p><p>Ask people where short sourdough videos live and most will say TikTok. Ask Google, and it’s a different answer. Its Short videos tab showed me 96 videos for “sourdough starter” this morning. 47 were YouTube Shorts. 25 were Facebook videos, 16 Instagram Reels. TikTok had 4.</p><p>That matters if you make short videos and want them found from a search box, or if you’re trying to work out who owns a topic. The tab is a ranking, and a ranking can be read.</p><h3>What I asked for</h3><p>I build and run a scraper for Google’s Short videos tab on Apify. You give it a search, the way you’d type it into Google, and it returns every video the tab lists, in Google’s order: the link on its own platform, title, creator, platform, duration, a thumbnail, the description Google carries and, where Google shows one, the date.</p><p>I ran “sourdough starter” twice from the US, in English. Once with no time filter and 96 videos, once with timeRange: &quot;month&quot; and 48. Here is the first, counting platforms:</p><pre>import { ApifyClient } from &#39;apify-client&#39;;<br>const client = new ApifyClient({ token: process.env.APIFY_TOKEN });<br>const run = await client.actor(&#39;simple.actor/google-short-videos&#39;)<br>  .call({ queries: [&#39;sourdough starter&#39;], maxItems: 96 });<br>const { items } = await client.dataset(run.defaultDatasetId)<br>  .listItems();<br>const count = {};<br>for (const v of items) {<br>  if (v.type !== &#39;video&#39;) continue; // error rows carry no video<br>  count[v.platform] = (count[v.platform] ?? 0) + 1;<br>}<br>console.log(count);</pre><p>The type check is there because a search that can&#39;t be read comes back as a row with type: &quot;error&quot;, not as an empty result. Both runs came back clean: 96 videos and 48, no error rows.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*EaAM7KPkw3G_l_ll.png" /></figure><p><em>The platform mix for one search, with and without the past-month filter. Only 7 videos were in both lists.</em></p><h3>YouTube owns it, and Facebook is second</h3><p>Half of each list was YouTube: 47 of 96 any time, 24 of 48 in the past month. Facebook was a clear second both times, 25 and 15. TikTok was 4 and 2.</p><p>The top of the list was more mixed. The first twelve positions held 5 YouTube videos, 2 each from Instagram, Facebook and TikTok, and one Pinterest pin. So TikTok gets into the top twelve and then all but disappears.</p><p>Pinterest wasn’t something I’d asked for. It showed up three times, at positions 7, 70 and 91, and Reddit once. The scraper’s platform filter offers seven platforms. Google’s mix with no filter isn’t bound to those seven.</p><p>The list is spread thin. 72 different creators made the 96 videos and only 17 had more than one. If you’re working out who to partner with or who you’re up against, that’s the useful part: a long list of small channels, with a few brands and a TV programme near the top.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*xBLQT_2o7qaNB7cz.png" /></figure><p><em>Six of the first twelve rows, the ones posted by brands and businesses. Only the YouTube rows have a date.</em></p><h3>The past month is a different list</h3><p>I expected the month filter to trim the old videos off the same list. It didn’t. Only 7 of its 48 videos were among the 96 from the unfiltered run. Every dated video in it was posted between 2 October and 1 November.</p><p>The unfiltered list is where the old stuff lives. Of its 46 dated videos, 14 were posted before 2026 and the oldest went up in March 2024. 8 were from October or the first day of November. So if the question is “what is Google showing people for my keyword right now”, run both: the unfiltered list is what most searchers see, and the month list is what’s new and climbing.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*04hSBDHND3HY7URN.png" /></figure><p><em>Publish dates of the 46 dated videos. The rest had none.</em></p><p>“Short” also stretches. The median video in the unfiltered run was 80.5 seconds. 67 of the 96 ran past a minute, 27 past two minutes, and the longest was exactly three minutes. In the month list the median was 60 seconds.</p><h3>What it cost</h3><p>The listing charges $0.0008 for each video returned, $0.006 for each search Google answers, and a $0.00005 start fee. The 96-video run was 96 × $0.0008 + $0.006 + $0.00005 = $0.08285. The month run was $0.04445. Both together came to under 13 cents. A search that can’t be read is never charged.</p><p>It wasn’t quick. Google’s tab shows about twelve videos a page and repeats some from one page to the next, and the scraper keeps each video once, so 96 different videos took 14 pages and 6 minutes 21 seconds. The 48-video run took 3 minutes 16 seconds.</p><h3>What the rows won’t tell you</h3><p>There are no view counts. Google’s tab doesn’t show them, so the rows can’t carry them. If you want to know whether a video is popular rather than just ranked, you need another source.</p><p>Dates are YouTube’s alone. 46 of the 47 YouTube videos had one; none of the 49 from other platforms did. For anything not on YouTube, the month filter is your only handle on age.</p><p>Creator names aren’t joined across platforms. Natasha’s Kitchen is “Natashas Kitchen” on YouTube and “natashaskitchen” on Pinterest. Count creators per platform, or normalise the names yourself, before you call anyone the top creator.</p><p>And a run can work harder than its result shows. When I tried the example script on “cold brew coffee”, the scraper logged that 6 of 9 video links on Google’s first page couldn’t be followed to the video. It read a second page and still returned the 12 I asked for, but those six were never seen, and the tab is a ranking, so a missing video is a missing position.</p><h3>Make it your keyword</h3><p>The example repo, <a href="https://github.com/simple-actor/google-short-videos-scraper">google-short-videos-scraper</a>, takes a search, a count and a time range, prints the platform mix and the top five, and writes every video to videos.csv, in Node and in Python. Put in your product (&quot;standing desk review&quot;), your town (&quot;things to do in Leeds&quot;), or your trade (&quot;bathroom tiling tips&quot;), add month to see what&#39;s new, and give the scraper platform: &quot;tiktok&quot; or &quot;instagram&quot; in its input if you only care about one. The scraper is at <a href="https://apify.com/simple.actor/google-short-videos">apify.com/simple.actor/google-short-videos</a>.</p><p><em>Originally published at </em><a href="https://simple-actor.github.io/short-videos-rank-on-google-keyword-sourdough-starter/"><em>https://simple-actor.github.io</em></a><em> on November 2, 2026.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=6f008cbc1ba3" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[How to turn a website into clean text for an LLM (15 pages of Yosemite’s park site)]]></title>
            <link>https://medium.com/@simple.actor.apify/how-to-turn-a-website-into-clean-text-for-an-llm-15-pages-of-yosemites-park-site-bd146b8b0863?source=rss-7e488d384d8b------2</link>
            <guid isPermaLink="false">https://medium.com/p/bd146b8b0863</guid>
            <category><![CDATA[llm]]></category>
            <category><![CDATA[data]]></category>
            <category><![CDATA[retrieval-augmented-gen]]></category>
            <category><![CDATA[web-scraping]]></category>
            <category><![CDATA[javascript]]></category>
            <dc:creator><![CDATA[Simple Actor]]></dc:creator>
            <pubDate>Sun, 01 Nov 2026 00:00:29 GMT</pubDate>
            <atom:updated>2026-11-08T05:35:44.996Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*9UkxqM_vTX08lGwM.png" /></figure><p><em>I wrote this with an AI assistant. The runs, numbers and code are real, and I checked them.</em></p><p>The Half Dome permit page on Yosemite’s park site is 42,587 characters of HTML. The part a person actually reads, the rules, the lottery dates, the 300 hikers a day, is 6,844 characters. The other 35,743 are markup: menus, scripts, image markup and a footer.</p><p>If you paste a page like that into a language model, or load it into a retrieval index, you pay for all 42,587 and the model has to find the 6,844 itself. Across the 15 visitor pages I read from nps.gov/yose this morning, the HTML came to 842,006 characters and the text to 78,433. That&#39;s 9.3%. Nine characters in every ten were not the page.</p><p>This post is how I got the text out, what it cost, and what still came along that I didn’t want.</p><h3>Two runs: find the pages, then read them</h3><p>I build and run a web page text scraper on Apify. It has two modes. discover reads one page and returns its text plus every link to the same site, each with the words that were clickable. text reads the pages you name and returns their text alone. It doesn&#39;t crawl. It reads what you give it and nothing else.</p><p>So the first run was the park’s home page in discover mode. It came back in one row with 2,342 characters of text and 50 links. 27 of them went into Yosemite’s own pages, with anchors like “Fees”, “Camping”, “Half Dome Permits” and “Current Conditions”. 11 went elsewhere on nps.gov: the privacy policy, FOIA, the agency’s FAQ. The other 12 were photos, and their anchor was the image’s alt text: “A waterfall flowing down a granite cliff.”</p><p>The anchor text is what makes the second step easy. I didn’t have to guess that the fees page lives at /planyourvisit/fees.htm. I picked 15 visitor pages by their anchors and read them in text mode:</p><pre>import { ApifyClient } from &#39;apify-client&#39;;<br>const client = new ApifyClient({ token: process.env.APIFY_TOKEN });<br>const actor = client.actor(&#39;simple.actor/page-text-reader&#39;);<br>const run = await actor.call({<br>  urls: [&#39;https://www.nps.gov/yose/planyourvisit/fees.htm&#39;],<br>  mode: &#39;text&#39;,<br>});<br>const ds = client.dataset(run.defaultDatasetId);<br>for (const page of (await ds.listItems()).items) {<br>  if (!(&#39;text&#39; in page)) continue; // never read, not empty<br>  console.log(page.title, page.textLength);<br>}</pre><p>All 15 came back with status 200 in 11.9 seconds. None needed a browser: every row said rendered: false, which means the text was already in the HTML the server sent. The scraper only opens a browser when the served HTML turns out not to hold the page.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*YTdd2FxDvXRYiVoZ.png" /></figure><p><em>Served HTML against the text kept, page by page. Most pages were 50,000 to 65,000 characters of markup.</em></p><p>The share of text moved a lot from page to page. The FAQ was the wordiest, 10,041 characters of text out of 51,163, almost 20%. Basic Information was the thinnest, 1,448 out of 42,427, 3.4%: most of that page is links to other pages. The median page held 5,529 characters of text, which is a comfortable size for one chunk of a retrieval index or one paste into a chat.</p><h3>What a row looks like</h3><p>Each page is one row: the address asked for, loadedUrl (where it ended up after any redirect), statusCode, title, text, textLength, textTruncated, rendered and retrievedAt. I also switched on includeHtml for this run, which adds the raw markup as html. That&#39;s where the 842,006 came from. You don&#39;t need it to get the text.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*HXanXZ1Fr66jUCWm.png" /></figure><p><em>The Half Dome row, and its text after the site banner and the page heading.</em></p><p>The text is the page in full, not a summary. “A maximum of 300 hikers are allowed (about 225 day hikers and 75 backpackers) each day” is in there word for word, and so is the preseason lottery’s window, “March 1 through March 31”. A full stop goes at the end of every heading and paragraph, so a heading doesn’t run into the sentence under it.</p><p>One field matters more than it looks. A page that couldn’t be read comes back with an error and no text key at all. A page that loaded and genuinely had nothing on it has text: &quot;&quot;. That&#39;s why the code skips rows without text rather than rows with empty text: if you&#39;re filling an index, you don&#39;t want a failed fetch to look like a page with nothing to say.</p><h3>What it cost</h3><p>The listing charges $0.002 for each page read and $0.00005 a run as a start fee. The 15-page run was $0.03005, and the discover run before it $0.00205, so the whole site read came to $0.0321. A page that can’t be read is never charged; a 404 is, because “this page isn’t there” is an answer. A thousand pages cost $2.00 plus the start fee.</p><h3>The sentences that came along</h3><p>The scraper takes out navigation, headers, footers and cookie banners, and drops a line it has already seen on the same page. It doesn’t compare one page with another. So whatever the site repeats on every page in a way that doesn’t look like navigation stays in, once per page.</p><p>On nps.gov that was 17 sentences. The “.gov” banner at the very top (“Official websites use .gov. A .gov website belongs to an official government organization in the United States.”), the text of an “Exiting nps.gov” notice with its “Cancel” button, and the park’s footer: “Contact Info.”, “Stay Connected.”, “Site Index.”, “Español.”, the park’s phone number. Every one of the 15 pages started with that banner.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*okaS9vNliYy5rcta.png" /></figure><p><em>About 439 characters a page, 8.4% of all the text the run returned.</em></p><p>It’s 8.4% of the text, and in a retrieval index it’s worse than its size. Fifteen chunks that all open with the same five sentences about .gov websites look alike to a similarity search. The fix is a few lines of code: count how many pages each sentence appears on, and drop the ones that appear on all of them. The example repo does exactly that, and on the fees page it took 7,875 characters down to 7,436.</p><p>Two smaller things. Where a link sat at the end of a sentence, a space was left before the full stop: “the second Monday in October .” And one heading on the permit page did run into its paragraph, “Permits for Backpackers (including Camping in Little Yosemite Valley) Backpackers-including those who…”, so the full stops aren’t perfect either. Neither hurts a language model much, but don’t expect typeset prose.</p><p>The link list needs a filter too. Twelve of the home page’s 50 links were .jpg files, and eleven left Yosemite for nps.gov&#39;s agency pages. Filter by anchor text and by path before you feed the list back in, or you&#39;ll pay $0.002 to read a photograph.</p><h3>Make it your own site</h3><p>The example repo, <a href="https://github.com/simple-actor/website-to-text-for-llm">website-to-text-for-llm</a>, takes a start page and a pattern for the anchor text, reads the matching pages and writes them to one site.txt with the repeated sentences gone, in Node and in Python. Point it at your own product&#39;s help centre with &quot;pricing|billing|refund&quot;, your council&#39;s site with &quot;bins|parking|permits&quot;, or a supplier&#39;s documentation with whatever you&#39;d search it for. The scraper is at <a href="https://apify.com/simple.actor/page-text-reader">apify.com/simple.actor/page-text-reader</a>.</p><p><em>Originally published at </em><a href="https://simple-actor.github.io/website-clean-text-for-llm-yosemite/"><em>https://simple-actor.github.io</em></a><em> on November 1, 2026.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=bd146b8b0863" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[How to get a table at a booked-out New York restaurant on Resy (8 restaurants, 30 days)]]></title>
            <link>https://medium.com/@simple.actor.apify/how-to-get-a-table-at-a-booked-out-new-york-restaurant-on-resy-8-restaurants-30-days-13726aed6d5a?source=rss-7e488d384d8b------2</link>
            <guid isPermaLink="false">https://medium.com/p/13726aed6d5a</guid>
            <category><![CDATA[new-york]]></category>
            <category><![CDATA[restaurant]]></category>
            <category><![CDATA[data]]></category>
            <category><![CDATA[web-scraping]]></category>
            <category><![CDATA[javascript]]></category>
            <dc:creator><![CDATA[Simple Actor]]></dc:creator>
            <pubDate>Sat, 31 Oct 2026 00:00:29 GMT</pubDate>
            <atom:updated>2026-11-05T04:09:19.492Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*XblXXHmmMEJqRa8W.png" /></figure><p><em>I wrote this with an AI assistant. The runs, numbers and code are real, and I checked them.</em></p><p>Carbone had no table for two on any of the next 30 days. Its calendar on Resy showed 29 days sold out and one closed. Torrisi, Tatiana and 4 Charles Prime Rib looked the same. Misi, in Williamsburg, had a table on 25 of its 27 days.</p><p>I read those calendars on the morning of 31 October 2026, eight New York restaurants in one run that took 21 seconds and cost 0.8 cents. This post is what they showed, and what “available” turned out to mean once I looked at the actual times.</p><h3>The run</h3><p>I build and run a Resy scraper on Apify. Give it Resy venue URLs, switch on includeCalendar, and each restaurant comes back as one row with Resy&#39;s day-by-day booking status for your party size: calendar (every day, marked available, sold-out or closed), availableDates and availableDateCount. No Resy account, no login.</p><pre>import { ApifyClient } from &#39;apify-client&#39;;<br><br>const client = new ApifyClient({ token: process.env.APIFY_TOKEN });<br>const url = &#39;https://resy.com/cities/new-york-ny/venues/lilia&#39;;<br>const run = await client.actor(&#39;simple.actor/resy-availability&#39;)<br>    .call({ startUrls: [{ url }], maxItems: 20, partySize: 2,<br>        includeCalendar: true, calendarDays: 30 });<br>const { items } = await client<br>    .dataset(run.defaultDatasetId).listItems();<br>for (const v of items) console.log(v.name, v.availableDates);</pre><p>I gave it twelve URLs. Eight came back. Across those eight, 43 days had a table for two. Three restaurants held all 43.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*UxvraHrlAMcExmD1.png" /></figure><p><em>Each square is a day in the restaurant’s </em><em>calendar field. Dashed squares are days past the end of the calendar Resy returned.</em></p><p>Lilia had 17 open days and Misi 25. Both belong to the same group, which Resy names in venueGroup. I Sodi had one: that same evening. Carbone, Torrisi and Tatiana had none. Neither did 4 Charles Prime Rib, whose calendar stopped at 19 November, 20 days out, or Double Chicken Please, whose calendar was six days long.</p><p>That length is worth reading on its own. I asked for 30 days at every restaurant, and Resy returned 30 for Carbone and Torrisi, 27 for Lilia, Misi and Tatiana, 20 for 4 Charles, 13 for I Sodi and 6 for Double Chicken Please. A day past the end of a restaurant’s calendar isn’t sold out; Resy just doesn’t show it.</p><h3>“Available” is not 8pm</h3><p>The calendar said Misi was available on Saturday 7 November. So I ran the scraper again with includeAvailability, that date and a party of two, to get the actual tables. Misi had 13. Every one of them was lunch, between 11:30 and 14:30. Nothing for dinner.</p><p>Wednesday 4 November looked better on the calendar for both Lilia and Misi. Each had 9 tables. Lilia’s ran from 17:00 to 18:00 and then from 21:00 to 21:30, on the patio and at the bar. Misi’s were at 17:00 and 17:15, then 21:00 to 21:30. Neither had a table between 6pm and 9pm.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*-kEWgkP6uHWLPBZD.png" /></figure><p><em>Each mark is one row in </em><em>slots, from the </em><em>startTime field. The grey bands are </em><em>notifyWindows.</em></p><p>So a calendar day marked available means at least one table, at some hour, in some part of the room. If you want 8pm, read the slots for the day you care about. Each slot carries its startTime, its type (Dining Room, Bar, Patio, Kitchen Counter) and turnTimeSeconds. At Lilia each table was held for two hours, 7,200 seconds. None of the 31 slots I read needed a deposit.</p><h3>The table that opened while I watched</h3><p>Five minutes after the first calendar run, I tested the example script on Lilia, I Sodi, Carbone and Torrisi. Lilia now had 18 open days, not 17. That evening, 31 October, had gone from sold-out to available. Somebody had cancelled.</p><p>That’s the real use for a booked-out restaurant: run the same check every few minutes and look for a day that changes. Each run of this check costs a tenth of a cent a restaurant. When nothing is open, Resy also says whether it will take you on its notify list: notifyAvailable was true at 7 of the 8 restaurants for the Saturday, with the hours it covers in notifyWindows. Double Chicken Please, whose calendar didn&#39;t reach that Saturday, offered none.</p><h3>Two things I got wrong first</h3><p><strong>My first run returned two restaurants out of twelve.</strong> maxItems defaults to 3, and it caps the URL list too, not only search results. One of the first three URLs didn&#39;t answer, so I got two rows. Set maxItems to at least the number of URLs you give it.</p><p><strong>Four of my URLs left no row at all.</strong> Don Angie, Rezdora, Ci Siamo and Le B. Resy answers “not found” for those addresses, and the scraper treats that as “not on Resy” and writes nothing. I searched Resy for each name with the scraper’s search mode, and it returned other restaurants: Don Don and Don Ceviche, Fedora and Rhodora. So either my addresses were wrong or those places don’t take bookings on Resy. The scraper can’t tell the two apart, so check that every URL you give it brings back a row. The example scripts print the ones that didn’t.</p><h3>What it cost</h3><p>The scraper charges $0.001 for each restaurant row it returns, $1.00 per 1,000, or $0.98 per 1,000 on Apify’s Business plan and up. A URL that leaves no row costs nothing. The 30-day calendar for eight restaurants was 8 rows, $0.008. The Saturday check was another $0.008: I asked it to keep restaurants with no tables ( keepUnavailable), so all eight came back and all eight were charged. Everything behind this post, the mistakes and the searches included, was 35 rows: $0.035.</p><h3>Where it runs out</h3><p><strong>Thirty days at most.</strong> For Carbone and Torrisi I asked for a calendar to 31 December. I got the same 30 days, to 29 November, and then only 25 December, marked not available. A run asking for December alone returned that one day. Whatever Resy holds beyond its window, it doesn&#39;t show it.</p><p><strong>The calendar is per party size.</strong> Every number here is for two. A table for four is a different question; set partySize and run it again.</p><p><strong>It reads availability, it doesn’t book.</strong> Each slot comes with Resy’s booking token, but booking is Resy&#39;s app and your account. The scraper tells you which day to open it on.</p><p><strong>A day can change between runs.</strong> Lilia’s evening opened up inside five minutes. Every row has a scrapedAt time, and a calendar is only true at that moment.</p><h3>Making it yours</h3><p>Swap in the Resy URLs of the places you’ve been trying to get into, in your city, set partySize to the size of your group, and run it on a schedule. Keep each run&#39;s availableDates and compare with the last one. A new date is a cancellation. If you only care about dinner, follow up the new dates with includeAvailability and keep the slots after 7pm.</p><p>The full script, in Node and Python, prints each restaurant’s open days and writes every day of every calendar to a CSV. It’s in the <a href="https://github.com/simple-actor/resy-table-finder">example repo on GitHub</a>. The scraper is <a href="https://apify.com/simple.actor/resy-availability">Resy Scraper on Apify</a>.</p><p><em>Originally published at </em><a href="https://simple-actor.github.io/get-table-booked-out-restaurant-resy-new-york/"><em>https://simple-actor.github.io</em></a><em> on October 31, 2026.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=13726aed6d5a" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[How to compare Facebook engagement between pages (7 London museums, 56 posts)]]></title>
            <link>https://medium.com/@simple.actor.apify/how-to-compare-facebook-engagement-between-pages-7-london-museums-56-posts-297ecb408202?source=rss-7e488d384d8b------2</link>
            <guid isPermaLink="false">https://medium.com/p/297ecb408202</guid>
            <category><![CDATA[marketing]]></category>
            <category><![CDATA[javascript]]></category>
            <category><![CDATA[web-scraping]]></category>
            <category><![CDATA[social-media]]></category>
            <category><![CDATA[facebook]]></category>
            <dc:creator><![CDATA[Simple Actor]]></dc:creator>
            <pubDate>Fri, 30 Oct 2026 00:00:13 GMT</pubDate>
            <atom:updated>2026-11-03T11:21:11.213Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*LN3s3GjnKDvAUvXd.png" /></figure><p><em>I wrote this with an AI assistant. The runs, numbers and code are real, and I checked them.</em></p><p>The first page I asked for was the wrong museum. I wanted London’s Science Museum, typed facebook.com/sciencemuseum, and got eight posts about a polar vortex, a Halloween party and the Journal of Mammalogy. That address belongs to the Science Museum of Minnesota. London&#39;s is facebook.com/sciencemuseumlondon.</p><p>It’s the mistake worth checking for first when you compare your page with other pages: make sure each URL is the page you think it is. The rest of this post is what the comparison showed once the URLs were right.</p><h3>The run</h3><p>I build and run a Facebook page scraper on Apify. Give it public page URLs and it returns each page’s latest posts, one row per post: pageName, time, postType (photo, video, link or text), text, likes, comments, shares and, for videos, viewsCount. likes counts every reaction, not only thumbs-up; the seven types are in their own columns, reactionLikeCount to reactionCareCount. No Facebook login and no cookies.</p><pre>import { ApifyClient } from &#39;apify-client&#39;;<br><br>const client = new ApifyClient({ token: process.env.APIFY_TOKEN });<br>const urls = [&#39;https://www.facebook.com/tategallery&#39;];<br>const run = await client.actor(&#39;simple.actor/facebook-page-posts&#39;)<br>    .call({ startUrls: urls.map((url) =&gt; ({ url })), maxPosts: 8 });<br>const { items } = await client<br>    .dataset(run.defaultDatasetId).listItems();<br>for (const p of items.filter((i) =&gt; !i.error))<br>    console.log(p.pageName, p.likes + p.comments + p.shares);</pre><p>On 30 October 2026 I gave it eight page addresses I took for London museums, with maxPosts at 8, the most it returns. The run took 74 seconds and came back with 57 rows: 8 posts from each of seven pages, Minnesota included, and one error row. The Science Museum&#39;s own page went into a second run, 22 seconds, 8 more posts.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*XwPLZ9BLjESEQdNP.png" /></figure><p><em>Five of the 56 rows, the newest from each of five museums, with the fields as the dataset names them.</em></p><h3>The ranking, two ways</h3><p>For each museum I added up reactions, comments and shares on every post and took the median, so one viral post doesn’t carry a page. The National Gallery came first at 1,331 a post, then the V&amp;A at 1,086.5. Tate (454), the Natural History Museum (453.5) and the British Museum (393) sat together. Royal Museums Greenwich got 59 and the Science Museum 15.</p><p>The British Museum has the most followers of the seven, 1.8 million, and came fifth. So I read each page’s follower count too, with the scraper’s details mode, and divided. Per 1,000 followers the V&amp;A leads with 1.09, the National Gallery has 0.89, and the British Museum drops to 0.22. Royal Museums Greenwich, with 132,000 followers, goes from sixth to fourth at 0.45.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*VcGlhJ-P30ngX6-v.png" /></figure><p><em>Median reactions + comments + shares per post, and the same median per 1,000 followers. Follower counts are Facebook’s rounded figures.</em></p><p>The single biggest post was the V&amp;A’s on 23 October, a portrait of a sitter whose garment still survives: 7,656 reactions, 210 comments and 625 shares. Next came the National Gallery’s 13 October post about its Jan van Eyck portraits, 6,425 in all, and the V&amp;A’s opening of its Freddie Mercury show, 5,752. The British Museum’s best was a video called “Stop horsing around”, with 52,692 views and 2,754 reactions, comments and shares.</p><h3>Photos beat video, mostly</h3><p>29 of the 56 posts were photos and 27 were videos. The median photo drew 641 reactions, comments and shares, the median video 162. That holds inside five of the seven museums. At the V&amp;A, photos had a median of 2,513 against 228.5 for its two videos.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*K7YYSupF5NnZXCiQ.png" /></figure><p><em>Median engagement by </em><em>postType, each museum on its own. The number of posts is in brackets.</em></p><p>The two exceptions are the two smallest pages. Royal Museums Greenwich’s videos got a median of 182 against 19 for photos, and the Science Museum’s 21 against 10. The British Museum posted seven videos and one photo, so its photo median is one post.</p><h3>What it cost</h3><p>The scraper charges $0.003 a run and $0.001 for each post it returns. The first run returned 56 posts, $0.059, eight of them from Minnesota. The second returned 8, $0.011. The follower counts are $0.004 a page in details mode, and seven pages came to $0.028. The whole comparison cost a caller $0.098. Done right first time, it would have been one posts run at $0.059 and the same $0.028 for followers.</p><h3>Where the numbers stop</h3><p><strong>Eight posts is the most there is.</strong> Facebook serves a logged-out reader about eight posts in one request, and asking again returns the same eight, so maxPosts stops at 8. Eight posts reached back 9.1 days at Tate and 16 days at the Science Museum. A page that posts three times a day gives you three days. For a longer history, run it on a schedule and keep the rows.</p><p><strong>Not every page answers.</strong> facebook.com/museumoflondon came back as an error row, not_available, which means private, restricted or removed. So did facebook.com/londonmuseum, my guess at a newer address. Error rows aren&#39;t charged, but the page is missing from the comparison.</p><p><strong>Facebook rate-limits the read.</strong> When I tested the example scripts a few minutes later, Facebook refused one of two pages, then both, with “Rate limit exceeded”. Each came back as a read_failed row and the run still succeeded. Three tries over two minutes were refused; the same page read cleanly ten minutes later. If you track many pages, spread the runs out.</p><p><strong>Follower counts are rounded.</strong> Facebook shows “1M followers” for the V&amp;A and gives no exact number, so the per-follower column is only as precise as that. The row says so in followersAreExact, which was false for all seven.</p><p><strong>I read the follower counts one page per run.</strong> A details run over several pages stopped on its second or third page for me, with the rows read so far saved. One page at a time worked every time. I&#39;ve reported it.</p><p><strong>A post can have no text at all.</strong> One Natural History Museum video had no caption, and its title and text both came back empty. The V&amp;A&#39;s 20 October post was a share; the row marks it with isShare.</p><h3>Making it yours</h3><p>Put your own page first in the list and your three or four nearest rivals after it: the other cafés on your street, the other councils in your county, the other brands on your shelf. Check each pageName against the page you meant. Then run it every week, and you have the median, the best post and the photo-against-video split for each of them, week by week.</p><p>The full script, in Node and Python, ranks the pages and writes every post to a CSV. It’s in the <a href="https://github.com/simple-actor/facebook-page-engagement-scraper">example repo on GitHub</a>. The scraper is <a href="https://apify.com/simple.actor/facebook-page-posts">Facebook Page Scraper on Apify</a>.</p><p><em>Originally published at </em><a href="https://simple-actor.github.io/compare-facebook-engagement-between-pages-london-museums/"><em>https://simple-actor.github.io</em></a><em> on October 30, 2026.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=297ecb408202" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[How much more do Edinburgh hotels cost on New Year’s Eve?]]></title>
            <link>https://medium.com/@simple.actor.apify/how-much-more-do-edinburgh-hotels-cost-on-new-years-eve-9956683c3b6e?source=rss-7e488d384d8b------2</link>
            <guid isPermaLink="false">https://medium.com/p/9956683c3b6e</guid>
            <category><![CDATA[web-scraping]]></category>
            <category><![CDATA[travel]]></category>
            <category><![CDATA[javascript]]></category>
            <category><![CDATA[data]]></category>
            <category><![CDATA[hotel]]></category>
            <dc:creator><![CDATA[Simple Actor]]></dc:creator>
            <pubDate>Thu, 29 Oct 2026 00:00:27 GMT</pubDate>
            <atom:updated>2026-10-31T05:40:10.403Z</atom:updated>
            <content:encoded><![CDATA[<h3>How much more do Edinburgh hotels cost on New Year’s Eve? (Booking.com prices for the same hotels, two weeks apart)</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*1hPbEWhla9utJ-Yl.png" /></figure><p><em>I wrote this with an AI assistant. The runs, numbers and code are real, and I checked them.</em></p><p>The Novotel in Edinburgh’s centre wants £66.67 for two people on the night of Thursday 14 January. On Thursday 31 December, New Year’s Eve, the same hotel wants £428.57. Same hotel, same booking, two people, one night. That’s 6.4 times the price.</p><p>The Novotel isn’t unusual. I asked Booking.com for every place in Edinburgh on both nights, two weeks apart, and matched them up. The 32 hotels that were bookable on both nights cost a median £399.93 on Hogmanay and £71.27 in January: 5.2 times as much.</p><h3>Two searches and a join</h3><p>I build and run a Booking.com scraper on Apify. Give it a destination and dates and it returns a row for every property Booking would sell you: name, Booking’s own hotelId, property type, stars, review score and count, address, coordinates, and price, which is Booking&#39;s total for the whole stay. Every row also carries priceBasis, which says exactly what that total covers. Here it read &quot;total for 1 night, 2 adults, 1 room, as Booking quotes it&quot;. No Booking account and no key.</p><pre>import { ApifyClient } from &#39;apify-client&#39;;<br><br>const client = new ApifyClient({ token: process.env.APIFY_TOKEN });<br>const run = await client.actor(&#39;simple.actor/booking-hotels&#39;).call({<br>    queries: [&#39;Edinburgh&#39;], currency: &#39;GBP&#39;, adults: 2,<br>    checkIn: &#39;2026-12-31&#39;, checkOut: &#39;2027-01-01&#39;, maxHotels: 300,<br>});<br>const { items } = await client<br>    .dataset(run.defaultDatasetId).listItems();<br>for (const h of items) console.log(h.price, h.name, h.hotelId);</pre><p>I ran that twice on 29 October 2026, once for each night, and joined the two lists on hotelId. The New Year&#39;s Eve search took 6.5 seconds, the January one 16.5.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*Y2fL8iVdXV5sg9BJ.png" /></figure><p><em>The first six New Year’s Eve rows, with their positions, review scores and prices as Booking gave them.</em></p><h3>Most of the city is already gone</h3><p>The first thing the runs show is how little is left. Without dates, Booking lists 2,037 properties in Edinburgh. For one night on 14 January, 528 of them were bookable. For New Year’s Eve, two months ahead, 140 were.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*ijrJi_1FoqrYORP8.png" /></figure><p><em>Booking’s own count, </em><em>totalResults, for each search. The rest weren&#39;t offered for that night at all.</em></p><p>Of those 140, nothing was under £150. The cheapest was a guest house at £157.14, and the cheapest hotel was The Abbey Residence at £194.88. On 14 January, 86 of the 131 hotels were under £100, and the cheapest place in the city was a hostel at £23.81. The median across every kind of place went from £139.02 to £421.43.</p><h3>The same hotels, two weeks apart</h3><p>A median over two different sets of places can mislead, because the cheap ones may simply have sold out. So I took only the properties that appeared on both nights. There were 113, and the median one cost 3.92 times as much on New Year’s Eve.</p><p>Hotels went up the most. The 32 hotels on both lists had a median multiple of 5.2. Apartments, which are the biggest group, had 2.8, and six of them charged exactly the same on both nights.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*hLpgsCgEC7n4TGKB.png" /></figure><p><em>Thirteen well-known hotels, matched on </em><em>hotelId. The January dot is teal, New Year&#39;s Eve dark.</em></p><p>The budget hotels moved furthest. Britannia Edinburgh Hotel went from £35.71 to £308.73, 8.6 times. ibis budget Edinburgh Park went from £33.44 to £204.08. The five-star hotels started higher and rose less: the Sheraton Grand from £147.81 to £390.78, 2.6 times, and the W from £178.52 to £629.59. The extremes were 100 Princes Street at £2,458.73 against £252.98, and Stay Central Hotel at £726.66 against £59.96, 12.1 times.</p><p>So a cheap hotel in January is no guide to a cheap hotel on Hogmanay. On New Year’s Eve the gap between a budget room and a five-star one narrows: in January the Sheraton cost 4.4 times the ibis budget, on New Year’s Eve 1.9 times.</p><h3>What it cost</h3><p>The scraper charges $0.0015 for each property it returns, and nothing for a search that can’t be read. The New Year’s Eve search returned 140 properties, $0.21. The January one returned 520, $0.78. The count without dates was one property, $0.0015. A caller would pay $0.99 for the whole comparison.</p><p>Most of that is the January list, which I read in full so that every New Year’s Eve property had a chance to match. If you only care about hotels, Booking’s own filter ht_id=204 in the filters field leaves out the flats and rooms, and you pay for fewer rows.</p><h3>What the rows won’t tell you</h3><p><strong>Prices move while you watch.</strong> I ran the same New Year’s Eve search again four minutes later to test the example scripts. CODE The Court, a hostel in Parliament Square, had gone from £325.25 to £477.25, and the ibis on South Bridge from £369.84 to £403.97. A minute after that, the count had dropped from 140 to 139, and January’s from 528 to 516. Each row is what Booking quoted at retrievedAt, and for a night this busy that can be minutes old.</p><p><strong>The row doesn’t say which room.</strong> price is Booking&#39;s quote for the booking you asked for, but rooms aren&#39;t in the search results. On New Year&#39;s Eve the cheap rooms may be gone and the quote may be for a bigger one. The room list is on the hotel&#39;s own page, and every row links to it in url, but the scraper doesn&#39;t read it.</p><p><strong>The missing places don’t say why.</strong> 1,897 of the 2,037 weren’t offered for New Year’s Eve. Sold out, closed, or not selling a single night: the search doesn’t say, it just leaves them out.</p><p><strong>A full list can come back a few short.</strong> Booking counted 528 for January and the run returned 520 at a limit of 600. 27 of the New Year’s Eve properties weren’t in the January list, so they couldn’t be compared.</p><h3>Making it yours</h3><p>Swap Edinburgh for your city and the two dates for the ones you’re weighing: a festival weekend against the week after, a match day against a quiet Tuesday, the week your conference is on against the week before. Change adults and rooms to your own booking, because the prices are for exactly that. Run it every morning and you have a price history keyed on hotelId.</p><p>The full script, in Node and Python, runs both searches at once, joins them and writes a CSV with the multiple for every property. It’s in the <a href="https://github.com/simple-actor/booking-hotel-price-scraper">example repo on GitHub</a>. The scraper is <a href="https://apify.com/simple.actor/booking-hotels">Booking.com Scraper on Apify</a>.</p><p><em>Originally published at </em><a href="https://simple-actor.github.io/edinburgh-hotel-prices-new-years-eve-booking-com/"><em>https://simple-actor.github.io</em></a><em> on October 29, 2026.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=9956683c3b6e" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[How to find a business’s website, email and Instagram from just its name (8 New York bagel shops)]]></title>
            <link>https://medium.com/@simple.actor.apify/how-to-find-a-businesss-website-email-and-instagram-from-just-its-name-8-new-york-bagel-shops-d7d8c0ff9600?source=rss-7e488d384d8b------2</link>
            <guid isPermaLink="false">https://medium.com/p/d7d8c0ff9600</guid>
            <category><![CDATA[javascript]]></category>
            <category><![CDATA[lead-generation]]></category>
            <category><![CDATA[social-media]]></category>
            <category><![CDATA[web-scraping]]></category>
            <category><![CDATA[small-business]]></category>
            <dc:creator><![CDATA[Simple Actor]]></dc:creator>
            <pubDate>Wed, 28 Oct 2026 00:00:59 GMT</pubDate>
            <atom:updated>2026-10-30T05:36:48.489Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*a_9S8qvJcup8I6Rh.png" /></figure><p><em>I wrote this with an AI assistant. The runs, numbers and code are real, and I checked them.</em></p><p>What can you learn about a business if all you have is its name? I typed eight New York bagel shops into a scraper, each as a name and a neighbourhood, “Murray’s Bagels Greenwich Village New York” and so on, and nothing else. No website, no Maps link, no handles.</p><p>Eight rows came back in 5 minutes 51 seconds. Every one had a Google rating and a phone number. Seven had a website. Two had an email address on it, and two more only a contact form. Four had an Instagram account, read for its follower count and its three latest posts. A caller would pay 15.9 cents for the lot.</p><p>Three of the eight rows had something wrong in them, too, and those are worth as much of your attention as the good ones.</p><h3>One row, three sources</h3><p>I build and run a business profile scraper on Apify. Each name goes through three steps. First a Google search, whose knowledge panel holds the rating, review count, category, address, phone and website. Then the website itself, read for every email, phone number, contact form and social link it publishes. Then each linked social account, opened and read for followers and the latest posts. It all comes back as one row per business.</p><p>The neighbourhood matters. A bare name in a big city often matches the wrong branch, and one of my eight shows what that looks like.</p><pre>import { ApifyClient } from &#39;apify-client&#39;;<br><br>const client = new ApifyClient({ token: process.env.APIFY_TOKEN });<br>const queries = [<br>    &#39;Ess-a-Bagel Third Avenue New York&#39;,<br>    &quot;Murray&#39;s Bagels Greenwich Village New York&quot;,<br>];<br>const run = await client.actor(&#39;simple.actor/business-profile&#39;)<br>    .call({ queries, maxSocialProfiles: 3 });<br>const { items } = await client<br>    .dataset(run.defaultDatasetId).listItems();<br>for (const b of items) console.log(b.name, b.email, b.phone);</pre><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*77WjJpjvu17G4lad.png" /></figure><p><em>What came back for each name. Instagram follower counts as the rows carried them.</em></p><h3>What the names turned into</h3><p>Google was the strongest part. All 8 searches found a knowledge panel, all 8 had a rating and a phone number, and 7 had an address. Six of the eight had a review count, 14,885 reviews between them, from Ess-a-Bagel’s 154 to Russ &amp; Daughters’ 4,826.</p><p>The websites gave less. Six of the seven sites could be read, 21 pages in all. Murray’s Bagels publishes info@murraysbagels.com, and Black Seed Bagels two role addresses, catering@ and joinus@. Ess-a-Bagel and Tompkins Square Bagels publish no address but have a contact form, and the row says where it is in contactFormUrl. Five sites had phone numbers, 21 between them. Black Seed&#39;s had 9, all of them on the page for its Nolita shop.</p><p>The social step read 9 accounts on 4 businesses. Instagram came back for all four, from Murray’s 13,027 followers to Black Seed’s 91,506. Ess-a-Bagel also had TikTok, at 577 followers. Every Instagram and TikTok account had posted within the last week, and each row carries the three latest posts with their text, link and date.</p><p>Ess-a-Bagel also linked a LinkedIn page, its fourth account. With maxSocialProfiles at 3 it came back as a link and a handle, not read. That setting is a ceiling on the slowest and dearest part of the run, so a list of hundreds of businesses can&#39;t run away from you.</p><h3>What it cost</h3><p>This scraper charges $0.01 a business. It then calls one reader per website and per social account, and each of those bills its own listed price to the same caller. For these eight, at the prices listed on 28 October 2026, that came to $0.15875:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*9EjCEGUFUtDd6MaE.png" /></figure><p><em>Where the 15.9 cents went. Instagram, at $0.008 a profile, is the dearest account to read.</em></p><p>That’s about 2 cents a business. Instagram is the line to watch: four profiles cost $0.032, more than all six websites. Turn the social step off with includeSocialProfiles: false and these eight would have cost $0.10175, the place record and the website contacts only.</p><h3>Three rows that need a second look</h3><p><strong>Liberty Bagels.</strong> I asked for the Midtown shop. Google’s panel gave an address at 32 Broadway, New York, NY 10004, which is in the Financial District, nowhere near Midtown. It found no website, gave no review count, and the row’s name is my own query because the panel didn&#39;t title the business. My search wasn&#39;t specific enough. A street name would have fixed it.</p><p><strong>Absolute Bagels.</strong> Google lists its website as a rebrand.ly short link, and the link answered 404. The row says so in notes, and has no email, no site phones and no social accounts as a result. Its category reads &quot;More info&quot;, which is a button label on Google&#39;s panel, not a category. Black Seed&#39;s row has the same &quot;More info&quot;.</p><p><strong>Russ &amp; Daughters.</strong> The website was found and four of its pages were read, and the row came back with no email, no phone from the site and no social accounts, with nothing in notes to say why. Google&#39;s phone number is still on the row. Treat an empty website section with an empty notes as &quot;nothing found&quot;, not as &quot;nothing there&quot;.</p><p>Smaller things. Tompkins Square Bagels’ row has no review count, and its name is my query, the same panel gap as Liberty&#39;s. Black Seed&#39;s website gave a postal address for its Rockefeller Plaza shop rather than Nolita&#39;s: the site&#39;s own markup names that one. Facebook pages come back with their latest posts but no follower count, and the row says so; one of them came back with a display name that wasn&#39;t the shop&#39;s. Instagram posts come back without like counts. And each Instagram account carries a follower count next to a note saying Instagram publishes none to a logged-out reader. The number is what came back; the note contradicts it. I&#39;d check one by hand before quoting them anywhere.</p><h3>Making it yours</h3><p>The input is a list of names with a place on each. Swap the bagel shops for the gyms in your town, the dentists on your sales list, the competitors you watch. Put the street in when a chain has several branches. If you already have the websites, pass them as websites instead and the Google step is skipped. And if all you need is the phone and the email, turn the social step off and save the Instagram money.</p><p>The full script, in Node and Python, writes a CSV with one column per platform and is in the <a href="https://github.com/simple-actor/business-email-and-social-finder">example repo on GitHub</a>. The scraper is <a href="https://apify.com/simple.actor/business-profile">Business Profile Scraper on Apify</a>.</p><p><em>Originally published at </em><a href="https://simple-actor.github.io/find-business-website-email-instagram-from-name-new-york-bagels/"><em>https://simple-actor.github.io</em></a><em> on October 28, 2026.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=d7d8c0ff9600" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[How to tell how busy a restaurant is from OpenTable (50 steak restaurants in Chicago, ranked)]]></title>
            <link>https://medium.com/@simple.actor.apify/how-to-tell-how-busy-a-restaurant-is-from-opentable-50-steak-restaurants-in-chicago-ranked-430eb7c0cc13?source=rss-7e488d384d8b------2</link>
            <guid isPermaLink="false">https://medium.com/p/430eb7c0cc13</guid>
            <category><![CDATA[restaurant]]></category>
            <category><![CDATA[small-business]]></category>
            <category><![CDATA[web-scraping]]></category>
            <category><![CDATA[javascript]]></category>
            <category><![CDATA[data]]></category>
            <dc:creator><![CDATA[Simple Actor]]></dc:creator>
            <pubDate>Tue, 27 Oct 2026 00:00:42 GMT</pubDate>
            <atom:updated>2026-10-28T07:47:19.954Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*GJMBrTY3OQ2KT5vt.png" /></figure><p><em>I wrote this with an AI assistant. The runs, numbers and code are real, and I checked them.</em></p><p>At 10:37 on a Monday night, Chicago time, I searched OpenTable for “steak” in Chicago. The busiest of the 50 restaurants it gave back wasn’t a steakhouse. It was Gibsons Italia, booked 269 times. The next two, STK and Fioretta, had 111 each. OpenTable had listed Gibsons Italia 28th and Fioretta 43rd.</p><p>That number is on OpenTable’s listings, in small type: “Booked 269 times today”. It’s the cheapest honest measure of how busy a restaurant is that I know of. Nobody has to infer it. The restaurant’s own listing states it.</p><h3>Getting the number out</h3><p>I build and run an OpenTable scraper on Apify. Give it a search and a city and it reads OpenTable’s search page, which holds up to 50 restaurants, and returns one row for each: address, phone, coordinates, neighbourhood, cuisine, price band, rating, review count, and the booking figure as recentReservationCount. No OpenTable account, cookie or key.</p><pre>import { ApifyClient } from &#39;apify-client&#39;;<br><br>const client = new ApifyClient({ token: process.env.APIFY_TOKEN });<br>const run = await client.actor(&#39;simple.actor/opentable-restaurants&#39;)<br>    .call({ term: &#39;steak&#39;, city: &#39;Chicago&#39;, maxItems: 50 });<br>const { items } = await client<br>    .dataset(run.defaultDatasetId).listItems();<br>items.sort((a, b) =&gt;<br>    b.recentReservationCount - a.recentReservationCount);<br>for (const r of items) console.log(r.recentReservationCount, r.name);</pre><p>The full script, which writes a CSV, and a Python version are in <a href="https://github.com/simple-actor/opentable-scraper">the repo</a>.</p><p>The run took 69 seconds and returned 50 rows, every one with a phone number and a booking count. Between them they had been booked 2,796 times. The median was 53.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*l87Qt-Zd0ZgtEsn8.png" /></figure><p><em>The 12 busiest of the 50. Teal is a steakhouse. “Listed” is where the restaurant sat in OpenTable’s own order.</em></p><h3>Stars don’t tell you who is full</h3><p>The ratings ran from 3.9 to 4.9, a narrow band, and they said very little about bookings. STK Chicago had the lowest rating of all 50, 3.9, and was joint second busiest. The three rated 4.9 had 29, 42 and 55. Review counts said a bit more: Gibsons Italia’s 6,158 is a lot, but Smith &amp; Wollensky had 9,965 reviews and 72 bookings.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*rm51jHwY3A2V9u8A.png" /></figure><p><em>Six of the 50 rows. The last one needs reading twice.</em></p><p>OpenTable’s order didn’t follow bookings either. Of the seven restaurants with 100 or more, only STK was in the first ten it listed. Gibsons Bar &amp; Steakhouse in Oak Brook, with 103, was 49th of 50. If you only look at the top of the page, you miss most of the busy rooms.</p><h3>Watching the same places</h3><p>To track restaurants over time you don’t want a search at all. You want the same list every time. Give the scraper restaurantIds and it reads each restaurant&#39;s own OpenTable page instead, which carries the same record. I took ten of the Chicago IDs and read them twice: at 10:39 pm on Monday, and again at 12:06 am on Tuesday, just after Chicago&#39;s midnight.</p><p>Despite the “today” in the label, nothing reset. Not one of the ten dropped to zero, or anywhere near it. Gibsons Italia went from 269 to 264, Fioretta from 111 to 117, Prime &amp; Provisions from 107 to 112. Three didn’t move at all. The figure behaves like a rolling count over the last stretch of time, not a tally since midnight, so compare a restaurant with itself at the same hour and weekday rather than waiting for a daily zero.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*Qf16yQ-aeWkhCgcK.png" /></figure><p><em>The same ten restaurants read by ID, either side of midnight in Chicago.</em></p><h3>What it cost</h3><p>The listing charges $0.0125 a run and $0.0003 a restaurant, with Apify’s platform costs included. The search came to $0.0125 + 50 × $0.0003 = <strong>$0.0275</strong>. Each run of the ten IDs was $0.0125 + 10 × $0.0003 = $0.0155, and the two took 6 seconds and 39 seconds. The start charge isn’t made if the search page couldn’t be read.</p><h3>What the count won’t tell you</h3><p>Kinzie Chophouse is the row that needs care. It shows 2 bookings, which reads like an empty room, but its record type is OTConnectPlus, not the GuestCenter every other row carries. The scraper hasn&#39;t measured that type, so it says isBookable: null rather than guess. Check restaurantType before you treat a low number as a quiet restaurant.</p><p>The search is OpenTable’s, not mine. “Steak” in Chicago brought back 21 places that aren’t steakhouses (8 Italian, 6 American, 3 seafood, 4 others) and 17 outside the city itself, 9 of them in Oak Brook. Filter on cuisine and city if you want only steakhouses in Chicago.</p><p>One search is one page, so it stops at 50. A whole city’s restaurants take several searches, by cuisine or by neighbourhood.</p><p>The code is in <a href="https://github.com/simple-actor/opentable-scraper">simple-actor/opentable-scraper</a>, in Node and Python. Both take a search and a city on the command line and write a CSV, busiest first, or take IDS= for a fixed list. The scraper is <a href="https://apify.com/simple.actor/opentable-restaurants">OpenTable Scraper</a> on Apify. Put in your own cuisine and your own town, and if you run it on a schedule, read the same IDs at the same hour each week.</p><p><em>Originally published at </em><a href="https://simple-actor.github.io/how-busy-is-a-restaurant-opentable-chicago-steak/"><em>https://simple-actor.github.io</em></a><em> on October 27, 2026.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=430eb7c0cc13" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>