<?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 Camp on Medium]]></title>
        <description><![CDATA[Stories by Camp on Medium]]></description>
        <link>https://medium.com/@camp52401?source=rss-3436a8b94f52------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/0*Zae9rslNkQXoT8z1</url>
            <title>Stories by Camp on Medium</title>
            <link>https://medium.com/@camp52401?source=rss-3436a8b94f52------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Thu, 19 Nov 2026 07:49:11 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@camp52401/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[IndexNow, Explained: Fast Indexing for Bing and AI Search]]></title>
            <link>https://medium.com/@camp52401/indexnow-explained-fast-indexing-for-bing-and-ai-search-3d2e179b649e?source=rss-3436a8b94f52------2</link>
            <guid isPermaLink="false">https://medium.com/p/3d2e179b649e</guid>
            <category><![CDATA[seo]]></category>
            <category><![CDATA[ai-search]]></category>
            <category><![CDATA[indexnow]]></category>
            <category><![CDATA[bing]]></category>
            <dc:creator><![CDATA[Camp]]></dc:creator>
            <pubDate>Wed, 18 Nov 2026 17:13:11 GMT</pubDate>
            <atom:updated>2026-11-18T17:13:11.183Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="Hands typing on a laptop with a digital technology overlay" src="https://cdn-images-1.medium.com/max/1024/1*8R-It0nib0eXbX3SZB6oYQ.jpeg" /></figure><p>For most of the web’s history a search engine learned about a new or changed page whenever it next happened to crawl it, which could take weeks. IndexNow reverses that: your site tells participating engines the moment a URL changes, with one HTTP request. Because Bing’s index feeds several AI search features, it’s among the least expensive steps you can take for AI visibility. This guide explains the mechanics, setup, limits and what my own heavy use of it taught me.</p><p>The protocol is simple. Generate a key, host it as a text file on your domain, then send changed URLs to an endpoint with the key. One submission is shared with every participating engine, including Bing, Yandex, Seznam, Naver and Yep. Google doesn’t participate, so keep using sitemaps and Search Console there. I show single-URL and bulk submissions (up to 10,000 URLs per request), the response codes you’ll see, and setup options for WordPress, Cloudflare and custom PHP.</p><p>The rules that keep you out of trouble: submit only URLs that actually changed, only submit pages that return 200 and are indexable (or properly deleted ones), keep the key file in place permanently and back off on rate-limit responses.</p><p>Then the honest part. On my own site, hundreds of thousands of submissions led to only a small fraction of pages indexed. IndexNow gets a URL looked at sooner; it doesn’t decide whether it gets indexed. Fewer, stronger pages plus real links beat volume.</p><h3>Key takeaways</h3><ul><li>IndexNow pushes URL changes to participating engines in one request</li><li>One ping is shared across Bing, Yandex, Seznam, Naver and Yep</li><li>Google doesn’t use it, so keep your sitemap and Search Console</li><li>Submit only changed, indexable URLs</li><li>Faster discovery doesn’t override page quality or authority</li></ul><p>Related on BuiltToWinWeb: <a href="https://built2winweb.com/blog/where-to-submit-your-sitemap/">where to submit your sitemap</a></p><p><strong>Read the full guide on BuiltToWinWeb: </strong><a href="https://built2winweb.com/blog/indexnow-guide/"><strong>The complete IndexNow guide</strong></a></p><p><em>A summary by Jacob Campbell of an article first published on built2winweb.com on October 5, 2026.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=3d2e179b649e" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Where to List a Public MCP Server: Registry, Directories and Your Site]]></title>
            <link>https://medium.com/@camp52401/where-to-list-a-public-mcp-server-registry-directories-and-your-site-e285f5e2f855?source=rss-3436a8b94f52------2</link>
            <guid isPermaLink="false">https://medium.com/p/e285f5e2f855</guid>
            <category><![CDATA[ai-agent]]></category>
            <category><![CDATA[technical-seo]]></category>
            <category><![CDATA[model-context-protocol]]></category>
            <dc:creator><![CDATA[Camp]]></dc:creator>
            <pubDate>Tue, 17 Nov 2026 17:59:06 GMT</pubDate>
            <atom:updated>2026-11-17T17:59:06.809Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="Person working at a laptop at a wooden desk" src="https://cdn-images-1.medium.com/max/1024/1*MONhjURRtywelILIRFcaRQ.jpeg" /></figure><p>A public MCP server is only useful once people and assistants can locate it. Discovery happens in three places: the official registry, community directories and your own website. I cover all three, along with what to check before listing anywhere, since listing a broken or insecure server does more harm than not listing it.</p><p>The readiness check comes first: it handles initialize, tools/list and tools/call properly over Streamable HTTP, each tool is clearly described with an input schema and annotations (output schemas are a bonus), it passes a basic security review, and it has a stable public URL.</p><p>The official registry launched in preview in September 2025. You describe your server in a server.json file and publish it with the registry’s command-line tool. Names are namespaced and verified, either through GitHub or by proving you control a domain, and a domain namespace suits a business because the name matches your website. Community directories differ in what they ask for and whether they test servers, and some pull from the official registry automatically, which is a good reason to publish there first.</p><p>The listing that never goes stale is on your own site: a server card under /.well-known/, lines in llms.txt and agents.txt, and a human page explaining the tools. I finish with how to write the description and what to watch after listing.</p><h3>Key takeaways</h3><ul><li>Test protocol compliance and security before listing</li><li>Publish a server.json to the official MCP Registry</li><li>Use a domain namespace that matches your business</li><li>Some directories pull from the official registry</li><li>Keep a server card and llms.txt entry on your own site</li></ul><p>Related on BuiltToWinWeb: <a href="https://built2winweb.com/blog/mcp-server-security-checklist/">the MCP server security checklist</a></p><p><strong>Read the full guide on BuiltToWinWeb: </strong><a href="https://built2winweb.com/blog/list-your-mcp-server/"><strong>Where to list your MCP server</strong></a></p><p><em>A summary by Jacob Campbell of an article first published on built2winweb.com on October 5, 2026.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=e285f5e2f855" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Six llms.txt Examples for Different Kinds of Business]]></title>
            <link>https://medium.com/@camp52401/six-llms-txt-examples-for-different-kinds-of-business-b4163d85c6b6?source=rss-3436a8b94f52------2</link>
            <guid isPermaLink="false">https://medium.com/p/b4163d85c6b6</guid>
            <category><![CDATA[llm]]></category>
            <category><![CDATA[seo]]></category>
            <category><![CDATA[small-business]]></category>
            <category><![CDATA[ai-search]]></category>
            <dc:creator><![CDATA[Camp]]></dc:creator>
            <pubDate>Mon, 16 Nov 2026 17:08:48 GMT</pubDate>
            <atom:updated>2026-11-16T17:08:48.484Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="Person using a laptop with AI tools shown on screen" src="https://cdn-images-1.medium.com/max/1024/1*yuQoqaDI1oevLioxKX8FUQ.jpeg" /></figure><p>The llms.txt format is easy, yet most files I see are either a dump of every URL or a paragraph of sales copy, and neither helps a language model. A good one explains what the business does, where, who it serves and how to get in touch, notes what it doesn’t do, and links only to the pages that answer real questions. This post gives six complete, fictional examples you can adapt.</p><p>The six are a dentist, a restaurant, a SaaS product, a law firm, a plumber and an online shop. Each follows the same shape: an H1 name, a blockquote of hard facts (category, place, how long established, a price signal, one way to act, a phone number), a line stating what the business doesn’t do, a few H2 groups of carefully chosen links and, where useful, an Optional section a model can skip.</p><p>After the examples I pull out what they share, including why the negative line matters (it stops assistants recommending you for work you’ll turn down) and why links should be absolute. Then some industry-specific tips: state licences and jurisdictions for healthcare and legal, keep hours and dietary options current for restaurants, list service areas as town names for trades, put shipping and returns in the summary for shops, and link documentation for SaaS.</p><p>The final warning is about staleness. Generate the file from the same data as your pages, because an out-of-date llms.txt can do more harm than none.</p><h3>Key takeaways</h3><ul><li>Lead with a blockquote of hard facts, not marketing copy</li><li>Say what you don’t do</li><li>Link only the pages that answer real questions</li><li>Use absolute URLs so the file works anywhere</li><li>Generate it from your site data so it stays current</li></ul><p>Related on BuiltToWinWeb: <a href="https://built2winweb.com/tools/llms-txt-validator/">the free llms.txt validator</a></p><p><strong>Read the full guide on BuiltToWinWeb: </strong><a href="https://built2winweb.com/blog/llms-txt-examples/"><strong>Six ready-to-adapt llms.txt examples</strong></a></p><p><em>A summary by Jacob Campbell of an article first published on built2winweb.com on October 5, 2026.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=b4163d85c6b6" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[LocalBusiness Schema for Search and AI: What to Mark Up]]></title>
            <link>https://medium.com/@camp52401/localbusiness-schema-for-search-and-ai-what-to-mark-up-a65194297201?source=rss-3436a8b94f52------2</link>
            <guid isPermaLink="false">https://medium.com/p/a65194297201</guid>
            <category><![CDATA[schema]]></category>
            <category><![CDATA[structured-data]]></category>
            <category><![CDATA[local-seo]]></category>
            <category><![CDATA[ai-search]]></category>
            <dc:creator><![CDATA[Camp]]></dc:creator>
            <pubDate>Sun, 15 Nov 2026 18:44:00 GMT</pubDate>
            <atom:updated>2026-11-15T18:44:00.211Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="Palm trees against a bright blue sky" src="https://cdn-images-1.medium.com/max/1024/1*KqubuMnApyIoaD_vl1O-kQ.jpeg" /></figure><p>If a person asks an assistant to recommend a roofer for a leak in their town, the system needs to figure out which firms exist, what services they offer, the areas they cover and the hours they’re open. Much of that comes from search indexes built partly on structured data. For a local business, JSON-LD is the clearest way to spell out the details you want repeated, and this guide covers how to do it well.</p><p>I explain why structured data matters more now: beyond rich results, it supports entity understanding, an unambiguous record of who you are. Nobody can promise a particular assistant reads your markup directly, but Google and Bing do, and most AI search features stand on them.</p><p>The practical advice: use the most specific true type (RoofingContractor rather than LocalBusiness, for example), keep every value matched to your page and Google Business Profile, and pay attention to the properties that do the most work: a stable @id, address, phone and coordinates, opening hours including holidays, areaServed for service-area businesses, sameAs links to your profiles and your services.</p><p>I also cover the review rule many sites break: since 2019, reviews a business publishes about itself on its own site aren’t eligible for review stars on LocalBusiness or Organization results. Never mark up reviews that aren’t visible, never invent ratings, add breadcrumbs and test with the tools I list.</p><h3>Key takeaways</h3><ul><li>Use the most specific LocalBusiness subtype that’s true</li><li>Markup must match the page and your Business Profile</li><li>sameAs links connect your site to your other profiles</li><li>Self-serving reviews don’t earn stars for local businesses</li><li>Breadcrumb markup is an easy, reliable win</li></ul><p>Related on BuiltToWinWeb: <a href="https://built2winweb.com/local-seo-tips/">local SEO tips for the map pack</a></p><p><strong>Read the full guide on BuiltToWinWeb: </strong><a href="https://built2winweb.com/blog/local-business-schema-ai-search/"><strong>The LocalBusiness schema guide for AI search</strong></a></p><p><em>A summary by Jacob Campbell of an article first published on built2winweb.com on October 5, 2026.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=a65194297201" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Serving Markdown Versions of Your Pages to AI Agents]]></title>
            <link>https://medium.com/@camp52401/serving-markdown-versions-of-your-pages-to-ai-agents-98ffd665c17e?source=rss-3436a8b94f52------2</link>
            <guid isPermaLink="false">https://medium.com/p/98ffd665c17e</guid>
            <category><![CDATA[technical-seo]]></category>
            <category><![CDATA[markdown]]></category>
            <category><![CDATA[ai-agent]]></category>
            <dc:creator><![CDATA[Camp]]></dc:creator>
            <pubDate>Sat, 14 Nov 2026 17:11:32 GMT</pubDate>
            <atom:updated>2026-11-14T17:11:32.455Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="Hands typing on a laptop keyboard with a technology overlay" src="https://cdn-images-1.medium.com/max/1024/1*8R-It0nib0eXbX3SZB6oYQ.jpeg" /></figure><p>When an AI agent fetches a web page, much of what comes back isn’t content. It’s navigation, scripts, tracking tags, cookie notices and layout. The agent has to convert all that to text, and the conversion loses headings, mangles tables and sometimes keeps the navigation while losing the actual text. Offering a Markdown copy of every page avoids all of that.</p><p>In the guide I explain why agents prefer Markdown (fewer tokens, structure that survives and no extraction errors) and the two ways to provide it. The first is content negotiation: an agent sends an Accept header asking for text/markdown and gets Markdown from the same URL, while browsers still get HTML. The second is a .md twin URL for each page, advertised with a rel=alternate link in the page head. Most agent-ready sites do both.</p><p>The section that matters most for SEO covers headers: Vary: Accept on both versions so a CDN doesn’t serve Markdown to a browser (the most common mistake I see), a canonical Link header pointing at the HTML page, and optionally noindex on the .md twins.</p><p>I also cover what belongs in the Markdown (the real content plus facts that are only visual on the page, such as prices in a table), a minimal Apache rule for PHP sites, and three curl commands to test it.</p><h3>Key takeaways</h3><ul><li>Markdown gives agents your content without layout noise</li><li>Offer it via Accept-header negotiation and/or .md twin URLs</li><li>Always send Vary: Accept on negotiated responses</li><li>Point a canonical header at the HTML version</li><li>Test through your CDN, not just the origin</li></ul><p>Related on BuiltToWinWeb: <a href="https://built2winweb.com/blog/what-is-llms-txt/">what llms.txt is</a></p><p><strong>Read the full guide on BuiltToWinWeb: </strong><a href="https://built2winweb.com/blog/markdown-for-ai-agents/"><strong>How to serve Markdown to AI agents safely</strong></a></p><p><em>A summary by Jacob Campbell of an article first published on built2winweb.com on October 5, 2026.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=98ffd665c17e" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[MCP vs A2A: Tools for Assistants or Tasks Between Agents?]]></title>
            <link>https://medium.com/@camp52401/mcp-vs-a2a-tools-for-assistants-or-tasks-between-agents-e0022f121d81?source=rss-3436a8b94f52------2</link>
            <guid isPermaLink="false">https://medium.com/p/e0022f121d81</guid>
            <category><![CDATA[a2a-protocol]]></category>
            <category><![CDATA[mcp-protocol]]></category>
            <category><![CDATA[ai-agent]]></category>
            <dc:creator><![CDATA[Camp]]></dc:creator>
            <pubDate>Fri, 13 Nov 2026 17:10:24 GMT</pubDate>
            <atom:updated>2026-11-13T17:10:24.464Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="Palm trees against a bright blue sky with soft sun flare" src="https://cdn-images-1.medium.com/max/1024/1*KqubuMnApyIoaD_vl1O-kQ.jpeg" /></figure><p>Two open protocols come up in almost every discussion of agents and the web. MCP, the Model Context Protocol, links an assistant to tools and data. A2A, the Agent2Agent protocol, lets one agent pass a job to a separate agent. They’re often presented as rivals, but they answer different questions, and the same website can support both.</p><p>The one-line difference: MCP says “here are my tools, call them”, with the assistant in charge; A2A says “here is an agent, give it a task”, with the receiving agent deciding how to do the work. I cover each briefly. MCP was introduced by Anthropic in November 2024 and uses short, deterministic tool calls. A2A was introduced by Google in April 2025, moved to Linux Foundation governance, publishes an agent card under /.well-known/ and handles tasks with a lifecycle that can run for a while and stream progress.</p><p>A side-by-side comparison covers unit of work, who decides, duration, discovery and typical website use. They also fit together: an A2A agent often calls MCP tools internally, and Google’s Agent Payments Protocol is designed to extend both.</p><p>My recommendation for almost every business site is MCP first, because everyday assistants speak it and business facts are tool-shaped. A2A makes sense once you actually operate an agent that handles multi-step jobs for customers.</p><h3>Key takeaways</h3><ul><li>MCP exposes tools; A2A delegates tasks to another agent</li><li>MCP calls are short and deterministic; A2A tasks can be long-running</li><li>A2A agents often use MCP tools behind the scenes</li><li>Most business websites should add MCP first</li><li>Only publish an agent card that describes what actually exists</li></ul><p>Related on BuiltToWinWeb: <a href="https://built2winweb.com/blog/add-mcp-server-to-website/">how to add an MCP server to a website</a></p><p><strong>Read the full guide on BuiltToWinWeb: </strong><a href="https://built2winweb.com/blog/mcp-vs-a2a/"><strong>MCP vs A2A, compared side by side</strong></a></p><p><em>A summary by Jacob Campbell of an article first published on built2winweb.com on October 5, 2026.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=e0022f121d81" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[robots.txt for AI Bots: Training, Search and User-Triggered Explained]]></title>
            <link>https://medium.com/@camp52401/robots-txt-for-ai-bots-training-search-and-user-triggered-explained-4d5a3bc60c6f?source=rss-3436a8b94f52------2</link>
            <guid isPermaLink="false">https://medium.com/p/4d5a3bc60c6f</guid>
            <category><![CDATA[ai-crawler]]></category>
            <category><![CDATA[robotstxt]]></category>
            <category><![CDATA[technical-seo]]></category>
            <category><![CDATA[ai-search]]></category>
            <dc:creator><![CDATA[Camp]]></dc:creator>
            <pubDate>Thu, 12 Nov 2026 18:58:46 GMT</pubDate>
            <atom:updated>2026-11-12T18:58:46.059Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="Person working at a laptop at a home desk" src="https://cdn-images-1.medium.com/max/1024/1*MONhjURRtywelILIRFcaRQ.jpeg" /></figure><p>Blocking “AI bots” isn’t one decision, because AI crawlers do three different jobs. Training crawlers collect pages for future models. Search crawlers build the index an assistant consults when it cites sources. User-triggered fetchers load a page because a person asked. Blocking each has a different effect, and this guide explains the trade-offs.</p><p>I list the crawlers that matter from the providers’ own documentation: OpenAI’s GPTBot, OAI-SearchBot and ChatGPT-User; Anthropic’s ClaudeBot, Claude-SearchBot and Claude-User; Perplexity’s two agents; Google-Extended, which is a token controlling Gemini use rather than a separate crawler and doesn’t affect Search or AI Overviews; Applebot-Extended; and others such as CCBot.</p><p>Then three copy-ready policies. Policy A, the most common for businesses, keeps you visible in AI answers while opting out of training. Policy B is fully open, which is what I use on this site. Policy C is closed, suited to licensed content, with the trade-off that you’ll rarely be cited.</p><p>Two practical warnings follow. Your CDN can override your file through managed robots.txt or one-click blocking, so always check what’s actually served. And robots.txt is advisory and user agents can be faked, so if you treat bots differently, verify them with published IP ranges, reverse DNS or signed requests.</p><h3>Key takeaways</h3><ul><li>AI crawlers do three jobs: training, search and user-triggered fetching</li><li>Blocking search bots makes citations much less likely</li><li>Google-Extended doesn’t affect Google Search or AI Overviews</li><li>Check the robots.txt your CDN actually serves</li><li>Verify bots by IP range or DNS, not user agent alone</li></ul><p>Related on BuiltToWinWeb: <a href="https://built2winweb.com/blog/ai-crawler-user-agents/">the full AI crawler user-agent list</a></p><p><strong>Read the full guide on BuiltToWinWeb: </strong><a href="https://built2winweb.com/blog/robots-txt-ai-crawlers/"><strong>Robots.txt rules for AI crawlers</strong></a></p><p><em>A summary by Jacob Campbell of an article first published on built2winweb.com on October 5, 2026.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=4d5a3bc60c6f" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[llms.txt Explained: Format, Example and Honest Expectations]]></title>
            <link>https://medium.com/@camp52401/llms-txt-explained-format-example-and-honest-expectations-716a96513aed?source=rss-3436a8b94f52------2</link>
            <guid isPermaLink="false">https://medium.com/p/716a96513aed</guid>
            <category><![CDATA[ai-search]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[technical-seo]]></category>
            <dc:creator><![CDATA[Camp]]></dc:creator>
            <pubDate>Wed, 11 Nov 2026 18:01:48 GMT</pubDate>
            <atom:updated>2026-11-11T18:01:48.105Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="Hands typing on a laptop keyboard with a technology overlay" src="https://cdn-images-1.medium.com/max/1024/1*8R-It0nib0eXbX3SZB6oYQ.jpeg" /></figure><p>An llms.txt file is a short Markdown document in a site’s root folder that explains to language models what the site does and which pages matter most. It lives beside robots.txt and sitemap.xml, but it’s written for a reader with a limited context window and no patience for menus or scripts. This guide covers the whole format.</p><p>I start with where it came from: a proposal published in September 2024 by Jeremy Howard of Answer.AI, adopted first by documentation sites and since spread widely. Then the structure, which is deliberately small: an H1 with the site name (the only strictly required part), a blockquote summary, optional free text, H2 sections listing links, and an Optional section for links a model can skip. I include a complete example for a fictional roofing company and point out the detail people forget, which is saying what you don’t do.</p><p>I explain how llms.txt (the map) differs from llms-full.txt (the full text), and how both differ from robots.txt (permission) and sitemaps (completeness).</p><p>I’m honest about who reads it. As of this writing no major AI company has committed to using it for answers, and Google has said Search doesn’t. Agents and developer tools do fetch it, people paste it into assistants, and server logs show requests. So I treat it as cheap insurance and a useful discipline, with seven rules for writing a good one.</p><h3>Key takeaways</h3><ul><li>llms.txt is a short, curated Markdown summary at your site root</li><li>Only the H1 is strictly required; a summary blockquote is expected</li><li>Include what you don’t do to avoid wrong recommendations</li><li>llms-full.txt carries full text; llms.txt is the map</li><li>Treat it as low-cost insurance, not a ranking lever</li></ul><p>Related on BuiltToWinWeb: <a href="https://built2winweb.com/tools/llms-txt-validator/">the free llms.txt validator</a></p><p><strong>Read the full guide on BuiltToWinWeb: </strong><a href="https://built2winweb.com/blog/what-is-llms-txt/"><strong>What llms.txt is and how to write one</strong></a></p><p><em>A summary by Jacob Campbell of an article first published on built2winweb.com on October 5, 2026.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=716a96513aed" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[NLWeb and the /ask Endpoint: Conversational Access to a Website]]></title>
            <link>https://medium.com/@camp52401/nlweb-and-the-ask-endpoint-conversational-access-to-a-website-3bdd7888f337?source=rss-3436a8b94f52------2</link>
            <guid isPermaLink="false">https://medium.com/p/3bdd7888f337</guid>
            <category><![CDATA[ai-search]]></category>
            <category><![CDATA[ai-agent]]></category>
            <category><![CDATA[model-context-protocol]]></category>
            <category><![CDATA[schema-markup]]></category>
            <dc:creator><![CDATA[Camp]]></dc:creator>
            <pubDate>Tue, 10 Nov 2026 19:54:20 GMT</pubDate>
            <atom:updated>2026-11-10T19:54:20.676Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="Hands typing on a laptop with AI tools shown on screen" src="https://cdn-images-1.medium.com/max/1024/1*yuQoqaDI1oevLioxKX8FUQ.jpeg" /></figure><p>NLWeb is an open project Microsoft announced in May 2025 that adds a plain-language question endpoint to a website. A question sent to /ask comes back as structured Schema.org data built from the site’s own content. R.V. Guha, the creator of RSS and Schema.org, started it, and every NLWeb instance doubles as an MCP server, so one endpoint can answer both people’s questions and agents’ tool calls.</p><p>In the guide I explain the idea: NLWeb puts a site’s structured content into a vector database and responds to questions with results expressed as Schema.org types, so an agent can use the answer directly rather than parsing prose. I show what a request and response look like, the modes for listing, summarising or generating an answer, and streaming versus plain JSON.</p><p>I compare it with a conventional MCP server. MCP tools are precise (get the opening hours); /ask is open-ended (which services suit a restaurant?). Running it needs good structured data or feeds, a vector store and embeddings, a language model for the summary modes and somewhere to host the reference implementation.</p><p>I describe how we run /ask on this site, and give an honest recommendation for small businesses: not first. It shines on content-heavy sites like catalogues and publishers. A ten-page local site gets more from the basics, and should add NLWeb once people would genuinely ask it questions.</p><h3>Key takeaways</h3><ul><li>NLWeb gives a site a natural-language /ask endpoint</li><li>Answers come back as Schema.org-typed JSON</li><li>Every NLWeb instance also works as an MCP server</li><li>Its quality depends on your structured data</li><li>Small sites should do the agent-readiness basics first</li></ul><p>Related on BuiltToWinWeb: <a href="https://built2winweb.com/blog/mcp-vs-a2a/">MCP vs A2A explained</a></p><p><strong>Read the full guide on BuiltToWinWeb: </strong><a href="https://built2winweb.com/blog/what-is-nlweb/"><strong>What NLWeb is and how the /ask endpoint works</strong></a></p><p><em>A summary by Jacob Campbell of an article first published on built2winweb.com on October 5, 2026.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=3bdd7888f337" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[How the WebAgentScan Agent-Readiness Score Is Calculated]]></title>
            <link>https://medium.com/@camp52401/how-the-webagentscan-agent-readiness-score-is-calculated-aa2bfb1932a5?source=rss-3436a8b94f52------2</link>
            <guid isPermaLink="false">https://medium.com/p/aa2bfb1932a5</guid>
            <category><![CDATA[ai-agent]]></category>
            <category><![CDATA[ai-search]]></category>
            <category><![CDATA[web-development]]></category>
            <category><![CDATA[seo]]></category>
            <dc:creator><![CDATA[Camp]]></dc:creator>
            <pubDate>Mon, 09 Nov 2026 17:36:52 GMT</pubDate>
            <atom:updated>2026-11-09T17:36:52.740Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="Tall palm trees against a bright blue sky" src="https://cdn-images-1.medium.com/max/1024/1*KqubuMnApyIoaD_vl1O-kQ.jpeg" /></figure><p>WebAgentScan is the hosted scanner we built and run at Built2WinWeb. It rates a site, including any MCP server it runs, on a 100-point scale for how well AI agents can use it. A score only means something if you understand how it’s made, so this article explains every part of it.</p><p>The score is built from six pillars. Discovery (20 points) checks whether an agent can learn what the site offers from robots.txt, sitemaps, llms.txt, agent and server cards and the .well-known tree, and whether those files point at working URLs. Machine Content (15) checks whether content survives without JavaScript, whether Markdown versions exist and whether structured data parses. Agent Access (12) catches the self-inflicted damage: robots.txt rules, CDN blocking and firewalls shutting out AI search bots. MCP Quality (20) looks at protocol compliance, tool descriptions, schemas and annotations. Agent Security (18) tests the agent-facing surface, with findings that could help an attacker withheld until you verify you own the domain. Authority (15) reports link metrics per vendor under their own names, never blended into an invented figure.</p><p>A Commerce pillar adds up to 10 points only when a site sells something, and any pillar with nothing to grade is excluded rather than scored zero, so a brochure site isn’t penalised for not having a shop or server.</p><p>I finish with the order to fix things when a score comes back low, starting with agent access.</p><h3>Key takeaways</h3><ul><li>The score is built from six weighted pillars plus optional commerce</li><li>Pillars with nothing to grade are excluded, not zeroed</li><li>Agent Access problems are often self-inflicted by CDNs and firewalls</li><li>Sensitive security findings need domain verification to view</li><li>Fix access first; authority is the slow pillar</li></ul><p>Related on BuiltToWinWeb: <a href="https://built2winweb.com/blog/ai-agent-readiness-checklist/">the 30-point agent-readiness checklist</a></p><p><strong>Read the full guide on BuiltToWinWeb: </strong><a href="https://built2winweb.com/blog/what-is-webagentscan/"><strong>How the WebAgentScan score is built</strong></a></p><p><em>A summary by Jacob Campbell of an article first published on built2winweb.com on October 5, 2026.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=aa2bfb1932a5" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>