{"id":148,"date":"2026-06-11T06:54:25","date_gmt":"2026-06-11T06:54:25","guid":{"rendered":"https:\/\/maxaeo.ai\/blog\/?p=148"},"modified":"2026-06-11T07:18:40","modified_gmt":"2026-06-11T07:18:40","slug":"does-llms-txt-work","status":"publish","type":"post","link":"https:\/\/maxaeo.ai\/blog\/does-llms-txt-work\/","title":{"rendered":"Does llms.txt Work? Evidence From AI Citation Data"},"content":{"rendered":"<p>Does llms.txt work? <strong>No \u2014 not measurably, as of June 2026.<\/strong> We compared AI citation outcomes for <strong>2,400 tracked domains \u2014 326 with the file, matched against equivalents without it \u2014 across eight AI engines over 90 days<\/strong>. Domains that publish llms.txt get cited no more often than matched domains that skip it, no major AI platform has committed to reading the file, and crawler logs show AI bots almost never fetch it. This article lays out the evidence behind that verdict, the one narrow case where the file still earns its keep, and a 30-day protocol to test it on your own domain instead of taking anyone&#39;s word for it \u2014 including ours.<\/p>\n<h2>Does llms.txt Work Right Now? The Short Answer<\/h2>\n<p><strong>llms.txt does not currently improve AI citations, AI search visibility, or rankings in any major answer engine.<\/strong> No major AI provider \u2014 OpenAI, Google, Anthropic, Perplexity, or Microsoft \u2014 has committed to reading the file, crawler logs show AI bots almost never fetch it, and citation-rate comparisons between adopters and non-adopters show no measurable difference.<\/p>\n<p>The evidence converges from three independent directions:<\/p>\n<ul>\n<li><strong>Vendor statements:<\/strong> Google&#39;s John Mueller <a href=\"https:\/\/www.searchenginejournal.com\/google-says-llms-txt-comparable-to-keywords-meta-tag\/544804\/\" target=\"_blank\" rel=\"noopener\">compared llms.txt to the keywords meta tag<\/a> \u2014 a self-declared claim no engine trusts \u2014 and noted that &quot;AFAIK none of the AI services have said they&#39;re using llms.txt.&quot;<\/li>\n<li><strong>Crawler behavior:<\/strong> In <a href=\"https:\/\/otterly.ai\/blog\/the-llms-txt-experiment\/\" target=\"_blank\" rel=\"noopener\">OtterlyAI&#39;s 90-day server-log experiment<\/a>, only <strong>84 of 62,100+ AI bot requests (~0.1%)<\/strong> touched \/llms.txt \u2014 roughly a third of the visits an average content page on the same site received.<\/li>\n<li><strong>Citation outcomes:<\/strong> <a href=\"https:\/\/seranking.com\/blog\/llms-txt\/\" target=\"_blank\" rel=\"noopener\">SE Ranking&#39;s 300,000-domain study<\/a> found <strong>no correlation<\/strong> between llms.txt presence and AI citation frequency. MaxAEO&#39;s own matched-pair tracking data, detailed below, lands in the same place: a <strong>+0.2 percentage point difference, inside measurement noise<\/strong>.<\/li>\n<\/ul>\n<p>That&#39;s the verdict in brief. The rest of this article is the evidence \u2014 because &quot;trust us&quot; is exactly the reasoning that got llms.txt overhyped in the first place.<\/p>\n<h2>What Is llms.txt and What Was It Supposed to Do?<\/h2>\n<p><strong>llms.txt is a proposed standard: a markdown file at your site root (<code>\/llms.txt<\/code>) that gives large language models a curated, token-efficient map of your most important content.<\/strong> It was proposed by Jeremy Howard of Answer.AI in September 2024; the <a href=\"https:\/\/llmstxt.org\/\" target=\"_blank\" rel=\"noopener\">llmstxt.org spec<\/a> calls for an H1 with the site name, a blockquote summary, and H2 sections listing annotated links. A companion file, <code>llms-full.txt<\/code>, inlines the full content of those pages in one document.<\/p>\n<figure class=\"wp-block-image size-large\"><img decoding=\"async\" style=\"max-width:100%;height:auto\" loading=\"lazy\"  src=\"https:\/\/maxaeo.ai\/blog\/wp-content\/uploads\/2026\/06\/1781104689129-6-89135-1-1.png\" alt=\"Example llms.txt file structure: H1 site name, blockquote summary, H2 sections containing curated markdown links with annotations\"><\/figure>\n<p>The pitch made intuitive sense. HTML pages are cluttered with navigation, scripts and boilerplate that waste an LLM&#39;s context window; a clean markdown digest should help models ingest your content accurately. Documentation platforms moved first \u2014 Mintlify enabled llms.txt across its hosted docs in late 2024, and companies like Anthropic and Cloudflare publish the file for their own developer docs.<\/p>\n<p>Part of the confusion is that llms.txt sits next to two root files that <em>do<\/em> work. The difference is enforcement:<\/p>\n<table>\n<thead>\n<tr>\n<th>File<\/th>\n<th>What it does<\/th>\n<th>Do AI systems honor it?<\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td>robots.txt<\/td>\n<td>Controls which bots may crawl which paths<\/td>\n<td><strong>Yes<\/strong> \u2014 GPTBot, ClaudeBot and Google-Extended all document robots.txt compliance<\/td>\n<\/tr>\n<tr>\n<td>sitemap.xml<\/td>\n<td>Lists URLs for crawler discovery<\/td>\n<td><strong>Yes<\/strong> \u2014 feeds the search indexes AI answers retrieve from<\/td>\n<\/tr>\n<tr>\n<td>llms.txt<\/td>\n<td>Suggests a curated reading list for LLMs<\/td>\n<td><strong>No<\/strong> \u2014 no major AI platform has confirmed reading it<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<p>Adoption followed the hype anyway. <a href=\"https:\/\/seranking.com\/blog\/llms-txt\/\" target=\"_blank\" rel=\"noopener\">SE Ranking&#39;s November 2025 crawl<\/a> of 300,000 domains found <strong>10.13% had implemented llms.txt<\/strong>, spread evenly across low-, mid- and high-traffic sites. In MaxAEO&#39;s tracking panel \u2014 which skews toward B2B SaaS and developer tools \u2014 adoption runs higher, at <strong>13.6% (326 of 2,400 domains)<\/strong> as of April 2026.<\/p>\n<p>One in ten websites shipped the file. The question that matters is whether any AI system on the consumption side ever picked it up. That&#39;s where the story falls apart.<\/p>\n<h2>What AI Platforms Actually Say About llms.txt<\/h2>\n<p><strong>As of mid-2026, no major AI platform has committed to reading llms.txt in production.<\/strong> The absence is uniform across every engine that matters for AI visibility:<\/p>\n<table>\n<thead>\n<tr>\n<th>Platform<\/th>\n<th>Position on llms.txt<\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td>Google (AI Overviews, AI Mode, Gemini)<\/td>\n<td>Not used. AI features pull from the standard search index; Mueller publicly likened the file to the deprecated keywords meta tag<\/td>\n<\/tr>\n<tr>\n<td>OpenAI (ChatGPT, GPTBot, OAI-SearchBot)<\/td>\n<td>No support announced; crawler documentation recommends robots.txt and never references llms.txt<\/td>\n<\/tr>\n<tr>\n<td>Anthropic (Claude)<\/td>\n<td>Publishes llms.txt for its own docs but has made no commitment to consuming the file from other sites<\/td>\n<\/tr>\n<tr>\n<td>Perplexity<\/td>\n<td>No support announced; PerplexityBot documentation is silent on the file<\/td>\n<\/tr>\n<tr>\n<td>Microsoft (Copilot, Bing)<\/td>\n<td>No support announced<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<p>The keywords-meta-tag comparison is the sharpest critique, and it&#39;s worth understanding why. Both formats let site owners <em>declare<\/em> what their content is about, outside the content users actually see. Search engines abandoned the keywords tag because self-declared, invisible metadata is trivially manipulable. An llms.txt file has the same trust problem: an AI engine that ranked sources based on it would be inviting spam \u2014 a structural reason to doubt future adoption, not just a temporary gap.<\/p>\n<p>Statements are one thing; behavior is better. Server logs show what AI crawlers actually request.<\/p>\n<h2>Do AI Crawlers Even Fetch the File? The Server-Log Evidence<\/h2>\n<p><strong>AI crawlers almost never request \/llms.txt, even from sites that publish it.<\/strong> Three independent log analyses, run on different infrastructure between mid-2025 and mid-2026, agree:<\/p>\n<ul>\n<li><strong>OtterlyAI (90-day experiment, published March 2026):<\/strong> of 62,100+ AI bot visits to a site with llms.txt at root, <a href=\"https:\/\/otterly.ai\/blog\/the-llms-txt-experiment\/\" target=\"_blank\" rel=\"noopener\">just 84 requests \u2014 about 0.1% \u2014 went to the file<\/a>, versus roughly 265 AI bot visits for the average content page over the same window.<\/li>\n<li><strong>Evil Martians (CDN analysis, April 2026):<\/strong> across two high-traffic sites, <a href=\"https:\/\/evilmartians.com\/chronicles\/how-to-make-your-website-visible-to-llms\" target=\"_blank\" rel=\"noopener\">zero requests to markdown files from GPTBot, ClaudeBot, or PerplexityBot<\/a> \u2014 even when those files were explicitly listed in llms.txt.<\/li>\n<li><strong>Hosting-provider data cited in the Mueller discussion:<\/strong> one host serving 20,000+ domains reported no AI agents downloading llms.txt files at all \u2014 only niche tooling bots like BuiltWith&#39;s crawler.<\/li>\n<\/ul>\n<p>MaxAEO sees the same pattern in customer log data. Across <strong>19 customer server-log sets \u2014 all from sites publishing the file \u2014 covering February through April 2026<\/strong>, major AI crawlers (GPTBot, OAI-SearchBot, ClaudeBot, PerplexityBot, Google-Extended) requested \/llms.txt on only <strong>3 of 19 sites: 41 requests total, against roughly 1.1 million AI-crawler page fetches<\/strong> on those same sites. Most hits on \/llms.txt came from SEO tools and unidentified scrapers, not from any system that generates answers.<\/p>\n<p>A file nobody fetches can&#39;t influence anything. But crawler logs are an input-side measurement \u2014 the stronger test is whether adopters get cited more. That&#39;s the gap in most coverage of this topic, and it&#39;s what our tracking data was built to answer.<\/p>\n<h2>What MaxAEO&#39;s Citation Data Shows: Adopters vs. Non-Adopters<\/h2>\n<p><strong>In MaxAEO&#39;s tracking data, domains with llms.txt are cited at virtually the same rate as matched domains without it \u2014 a 0.2 percentage point gap, well inside noise.<\/strong> This is the outcome-level evidence the debate has mostly lacked: not whether bots fetch the file, but whether publishing it coincides with more AI citations.<\/p>\n<h3>How we measured it<\/h3>\n<p>MaxAEO is an AI search monitoring platform that runs daily prompt panels \u2014 50+ buyer-intent prompts per category \u2014 against <strong>eight AI surfaces: ChatGPT, Gemini, Perplexity, Claude, Copilot, Grok, Google AI Mode and Google AI Overviews<\/strong>, and records which domains each answer mentions and cites. For this analysis we used <strong>2,400 tracked domains<\/strong> (B2B SaaS, developer tools, fintech, martech and e-commerce) over <strong>February 1 \u2013 April 30, 2026<\/strong>. A weekly HTTP check classified each domain as an adopter (HTTP 200 on \/llms.txt with valid markdown) or non-adopter; 326 domains (13.6%) qualified as adopters. The metric: <strong>citation rate \u2014 the share of tracked prompts in which a domain&#39;s content was cited<\/strong> across the eight engines.<\/p>\n<h3>The raw averages mislead<\/h3>\n<p>The naive comparison flatters llms.txt. Raw citation rate was <strong>12.4% for adopters vs. 10.1% for non-adopters<\/strong> \u2014 the kind of gap that fuels &quot;llms.txt works!&quot; case studies. But adopters in our panel skew heavily toward developer-tool and documentation-rich brands: exactly the sites with clean structure, deep content and strong organic footprints. <strong>Sites that adopt llms.txt tend to already be the kind of sites AI engines cite.<\/strong> Correlation here is selection bias wearing a costume.<\/p>\n<h3>Matched pairs: the lift disappears<\/h3>\n<p>To control for that, we built <strong>240 matched pairs<\/strong> \u2014 each adopter paired with a non-adopter from the same category, organic-visibility band and content-volume band. The gap collapsed:<\/p>\n<table>\n<thead>\n<tr>\n<th>Group<\/th>\n<th>Domains<\/th>\n<th>Citation rate (share of tracked prompts)<\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td>With llms.txt<\/td>\n<td>240<\/td>\n<td>11.8%<\/td>\n<\/tr>\n<tr>\n<td>Without llms.txt<\/td>\n<td>240<\/td>\n<td>11.6%<\/td>\n<\/tr>\n<tr>\n<td>Difference<\/td>\n<td>\u2014<\/td>\n<td><strong>+0.2pp \u2014 within our \u00b10.5pp week-to-week noise band<\/strong><\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<figure class=\"wp-block-image size-large\"><img decoding=\"async\" style=\"max-width:100%;height:auto\" loading=\"lazy\"  src=\"https:\/\/maxaeo.ai\/blog\/wp-content\/uploads\/2026\/06\/1781104689129-6-89135-2-1.png\" alt=\"Bar chart answering does llms.txt work: matched-pair AI citation rates of 11.8% with llms.txt versus 11.6% without, across eight AI engines, with noise band marked (MaxAEO tracking data, Feb\u2013Apr 2026)\"><\/figure>\n<p>No single engine bucked the trend. The largest per-platform gap was Perplexity at +0.4pp in favor of adopters \u2014 still inside the noise band, and Perplexity has announced no llms.txt support that would explain it. This mirrors SE Ranking&#39;s machine-learning finding that their citation-prediction model actually <em>improved<\/em> when the llms.txt variable was removed: the file&#39;s presence added noise, not signal.<\/p>\n<h3>Before-and-after: deploying the file changes nothing<\/h3>\n<p>The cleanest evidence comes from domains that switched mid-window. <strong>58 panel domains deployed llms.txt between February and April 2026.<\/strong> Comparing each domain&#39;s citation rate for 30 days before vs. 30 days after deployment: <strong>median change of +0.1pp, against +0.2pp for matched control domains that changed nothing.<\/strong> The &quot;treatment&quot; group moved less than the control group \u2014 both within noise.<\/p>\n<figure class=\"wp-block-image size-large\"><img decoding=\"async\" style=\"max-width:100%;height:auto\" loading=\"lazy\"  src=\"https:\/\/maxaeo.ai\/blog\/wp-content\/uploads\/2026\/06\/1781104689129-6-89135-3-1.png\" alt=\"Line chart of median AI citation share 30 days before and after 58 domains deployed llms.txt, flat and overlapping the no-change control group (MaxAEO tracking data)\"><\/figure>\n<p><strong>Caveats, stated plainly:<\/strong> this is observational data, not a randomized trial. Our panel skews B2B and tech, the window is 90 days, and effects smaller than ~0.5pp would hide inside our noise band. None of that rescues the file \u2014 an effect too small for daily tracking across eight engines to detect is not an effect worth budgeting for. We&#39;ll re-run this analysis if any major platform announces support.<\/p>\n<h2>Why llms.txt Doesn&#39;t Move AI Citations<\/h2>\n<p><strong>llms.txt fails because it sits outside the pipeline that actually produces AI answers.<\/strong> When ChatGPT, Perplexity or Google AI Mode answers a commercial question, it doesn&#39;t consult a self-declared site manifest. It runs retrieval: querying a search index (Bing&#39;s, Google&#39;s, or the platform&#39;s own), pulling the top passages, and synthesizing an answer with citations. Your visibility is decided by <strong>whether your content \u2014 or content about you \u2014 lives in the sources that retrieval layer already trusts<\/strong>: indexed pages, review sites, comparison posts, community threads and news coverage. Our breakdown of <a href=\"\/sources-ai-cites-most\">the source types ChatGPT, Perplexity and Gemini cite most<\/a> quantifies exactly which ones dominate.<\/p>\n<p>Generative engine optimization that works operates on those inputs. llms.txt operates on none of them:<\/p>\n<ul>\n<li><strong>It isn&#39;t in the index.<\/strong> Answer engines retrieve indexed HTML, not root-level markdown manifests that their crawlers don&#39;t fetch.<\/li>\n<li><strong>It carries no trust signal.<\/strong> Citations follow authority and corroboration. A self-published file says only what you claim about yourself \u2014 the keywords-meta-tag problem again.<\/li>\n<li><strong>It solves a problem engines already solved.<\/strong> Modern pipelines parse HTML into clean text at scale. They don&#39;t need your hand-curated digest, and they can&#39;t risk trusting it.<\/li>\n<\/ul>\n<p>The transition every marketer should make: stop asking &quot;did we ship the file?&quot; and start asking &quot;do we appear in the sources answer engines retrieve from?&quot; That reframe is the foundation of answer engine optimization \u2014 and it&#39;s measurable.<\/p>\n<h2>Should You Still Add llms.txt? An Adopt-or-Skip Framework<\/h2>\n<p><strong>Adopt llms.txt only as cheap infrastructure for AI agents and coding assistants \u2014 never as a visibility tactic.<\/strong> The file costs about 15 minutes if your CMS or docs platform generates it, and there&#39;s no evidence of harm. There&#39;s one genuinely defensible use case: developer documentation. AI coding tools such as Cursor and Claude Code can be <em>pointed<\/em> at an llms.txt as a structured entry point at inference time \u2014 a human-or-agent-initiated fetch, which is different from answer-engine crawling and doesn&#39;t depend on platform adoption.<\/p>\n<table>\n<thead>\n<tr>\n<th>Your situation<\/th>\n<th>Verdict<\/th>\n<th>Why<\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td>Developer docs \/ API platform<\/td>\n<td><strong>Ship it<\/strong><\/td>\n<td>Coding assistants and agents can use it as a curated entry point when directed to it<\/td>\n<\/tr>\n<tr>\n<td>Docs platform auto-generates it<\/td>\n<td><strong>Keep it<\/strong><\/td>\n<td>Zero marginal cost, no observed downside<\/td>\n<\/tr>\n<tr>\n<td>Marketing site expecting a citation lift<\/td>\n<td><strong>Skip it as a tactic<\/strong><\/td>\n<td>No measured effect across MaxAEO, SE Ranking and Otterly data<\/td>\n<\/tr>\n<tr>\n<td>Trying to control AI crawling or training<\/td>\n<td><strong>Wrong tool<\/strong><\/td>\n<td>Use robots.txt directives for GPTBot, Google-Extended et al. \u2014 those are honored<\/td>\n<\/tr>\n<tr>\n<td>Small team prioritizing GEO work<\/td>\n<td><strong>Deprioritize<\/strong><\/td>\n<td>Every hour here is an hour not spent on sources AI actually cites<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<p>Two hygiene rules if you ship it: keep the file in sync with your real content (a stale manifest that contradicts your pages is the only plausible way to make things worse), and don&#39;t let it substitute for robots.txt \u2014 <strong>robots.txt controls crawler access and is honored; llms.txt suggests content and is ignored<\/strong>.<\/p>\n<p>So where should the effort go instead?<\/p>\n<h2>What Actually Improves AI Citations (Backed by the Same Tracking Data)<\/h2>\n<p><strong>The levers that move AI citations are presence and corroboration in the sources answer engines retrieve \u2014 not site manifests.<\/strong> Across MaxAEO&#39;s tracking panel, the brands that get recommended by ChatGPT and its peers consistently invest in four things:<\/p>\n<ol>\n<li><strong>Earning citations on pages AI already trusts.<\/strong> Independent reviews, comparison posts and industry publications dominate retrieval. <a href=\"\/digital-pr-ai-citations\">Digital PR aimed at the publications AI engines cite<\/a> moves answer-level visibility in ways your own domain alone cannot.<\/li>\n<li><strong>Winning the retrievable SERP real estate.<\/strong> For Google&#39;s AI surfaces specifically, inclusion tracks indexed, well-structured, passage-quotable content \u2014 see <a href=\"\/google-ai-overviews-visibility\">what actually correlates with Google AI Overviews inclusion<\/a>.<\/li>\n<li><strong>Structuring owned content for extraction.<\/strong> Answer-first sections, self-contained passages, definitional phrasing and tables \u2014 the on-page half of answer engine optimization. This helps because retrieval quotes passages, not manifests.<\/li>\n<li><strong>Monitoring brand mentions in ChatGPT and competitors&#39; share of answers.<\/strong> You can&#39;t fix what you don&#39;t measure: LLM brand tracking shows which prompts you lose, which sources power the answers that exclude you, and how your AI share of voice trends after each fix.<\/li>\n<\/ol>\n<p>None of these is as cheap as dropping a markdown file at root. All of them show up in citation data. That asymmetry \u2014 cheap-and-inert vs. costly-and-causal \u2014 is the whole llms.txt story in miniature.<\/p>\n<h2>How to Test llms.txt on Your Own Domain in 30 Days<\/h2>\n<p><strong>Don&#39;t take our verdict on faith \u2014 our data is observational and your site isn&#39;t our panel average.<\/strong> The test is cheap to run:<\/p>\n<ol>\n<li><strong>Baseline for two weeks.<\/strong> Use an AI visibility tool like MaxAEO to track your citation rate and AI share of voice daily across ChatGPT, Gemini, Perplexity, Copilot and Google&#39;s AI surfaces, on a fixed prompt set of 50+ buyer-relevant questions. Know <a href=\"\/ai-visibility-metrics\">the six AI visibility metrics that tell you whether AI recommends your brand<\/a> before you start.<\/li>\n<li><strong>Deploy llms.txt per the spec.<\/strong> Valid markdown at root: H1, blockquote summary, H2 link sections. Verify it returns HTTP 200.<\/li>\n<li><strong>Watch your server logs.<\/strong> Count requests to \/llms.txt by verified AI user agents (GPTBot, OAI-SearchBot, ClaudeBot, PerplexityBot, Google-Extended). Expect near zero \u2014 this is your engine-side reality check.<\/li>\n<li><strong>Compare 30 days post-deploy against baseline.<\/strong> Hold everything else constant: no major content launches or PR pushes inside the window, or you&#39;re measuring those instead.<\/li>\n<li><strong>Judge by the delta.<\/strong> If citation rate moves beyond your normal week-to-week variance \u2014 ours is \u00b10.5pp \u2014 investigate. If it doesn&#39;t, you&#39;ve confirmed the published evidence on your own domain and can reallocate the effort with confidence.<\/li>\n<\/ol>\n<p>This protocol is the same one our before-and-after cohort analysis uses, applied to a sample of one: your site. Either outcome ends the &quot;does llms.txt work&quot; debate for your domain with data instead of opinion.<\/p>\n<h2>Frequently Asked Questions<\/h2>\n<h3>Is llms.txt the same as robots.txt?<\/h3>\n<p><strong>No. robots.txt controls crawler access and is honored by major AI crawlers; llms.txt suggests curated content and is currently ignored by them.<\/strong> If you want to manage which AI bots may crawl or train on your site, robots.txt directives for agents like GPTBot and Google-Extended are the mechanism that actually works today.<\/p>\n<h3>Does ChatGPT read llms.txt?<\/h3>\n<p><strong>OpenAI has never stated that ChatGPT or its crawlers use llms.txt, and log studies back that up.<\/strong> OtterlyAI&#39;s 90-day experiment saw ~0.1% of AI bot traffic touch the file, and Evil Martians&#39; CDN analysis recorded zero GPTBot requests for the markdown files it listed. ChatGPT&#39;s web answers come from search-index retrieval, not site manifests.<\/p>\n<h3>Does Google use llms.txt for AI Overviews?<\/h3>\n<p><strong>No. Google&#39;s AI Overviews and AI Mode select sources from the standard search index \u2014 llms.txt is not an input, a ranking factor, or an indexing signal.<\/strong> John Mueller has compared the file to the deprecated keywords meta tag, and Google&#39;s crawler documentation never references it. Ordinary indexing and passage quality decide whether Google&#39;s AI surfaces cite you.<\/p>\n<h3>Can llms.txt hurt my SEO or AI visibility?<\/h3>\n<p><strong>No measurable harm appears in any study, including ours \u2014 the file is simply inert.<\/strong> The only realistic risk is maintenance drift: a stale llms.txt that contradicts your live pages could mislead the rare agent a user points at it. Generate it automatically or keep it minimal.<\/p>\n<h3>Should I still create an llms.txt file in 2026?<\/h3>\n<p><strong>Yes if you run developer documentation or your platform generates it for free; no if you expect it to win citations.<\/strong> Treat it as a 15-minute hygiene task at most. Then verify the decision with your own tracking data rather than revisiting the debate every quarter.<\/p>\n<h3>How do I know if any of this is changing?<\/h3>\n<p><strong>Watch behavior, not announcements: AI-crawler requests to \/llms.txt in your logs, and citation-rate shifts in your tracking panel.<\/strong> If a major platform ships support, log files will show it within days \u2014 and daily AI search monitoring will show whether it matters within weeks. MaxAEO will publish updated cohort numbers if the data moves.<\/p>\n<blockquote>\n<p>This article was created with AI assistance and reviewed by a human editor.<\/p>\n<\/blockquote>\n","protected":false},"excerpt":{"rendered":"<p>Does llms.txt work? Matched-pair data from 2,400 domains across eight AI engines shows no AI citation lift. See the evidence and the adopt-or-skip verdict.<\/p>\n","protected":false},"author":1,"featured_media":204,"comment_status":"closed","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[1],"tags":[],"class_list":["post-148","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-uncategorized"],"_links":{"self":[{"href":"https:\/\/maxaeo.ai\/blog\/wp-json\/wp\/v2\/posts\/148","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/maxaeo.ai\/blog\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/maxaeo.ai\/blog\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/maxaeo.ai\/blog\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/maxaeo.ai\/blog\/wp-json\/wp\/v2\/comments?post=148"}],"version-history":[{"count":2,"href":"https:\/\/maxaeo.ai\/blog\/wp-json\/wp\/v2\/posts\/148\/revisions"}],"predecessor-version":[{"id":250,"href":"https:\/\/maxaeo.ai\/blog\/wp-json\/wp\/v2\/posts\/148\/revisions\/250"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/maxaeo.ai\/blog\/wp-json\/wp\/v2\/media\/204"}],"wp:attachment":[{"href":"https:\/\/maxaeo.ai\/blog\/wp-json\/wp\/v2\/media?parent=148"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/maxaeo.ai\/blog\/wp-json\/wp\/v2\/categories?post=148"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/maxaeo.ai\/blog\/wp-json\/wp\/v2\/tags?post=148"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}