LLMs.txt for AI SEO: What Google Actually Said in June

Google’s John Mueller and Martin Splitt spent a full episode of the Search Off the Record podcast in mid-June pushing back on the idea that publishing a stripped-down Markdown version of your site or a /llms.txt file is a real AI SEO strategy. Google’s own AI optimization guide, updated May 15, 2026, lists llms.txt for AI SEO as the first item in its mythbusting section. Server-log telemetry across 137,000 domains confirms the practical read: 97% of llms.txt files received zero requests in May 2026. You need a clear answer for clients who keep asking about llms.txt for AI SEO.

What Google actually said about llms.txt for AI SEO

According to Search Engine Journal’s coverage of the Search Off the Record episode, Mueller said he spoke directly with one of the people who created the LLMs.txt proposal. The creator told Mueller the file was never intended to make a site discoverable by AI systems. It was meant for an AI assistant that already knew about your site and wanted a curated map of what else was on it. Discovery was never part of the proposed standard. That is the gap most operators have not noticed when they ship a /llms.txt file.

Mueller’s second argument against llms.txt for AI SEO is more direct. The file is a self-report. The site owner writes it and decides what it says. He compared it to the old keywords meta tag that search engines have ignored for more than a decade for the same reason. A signal a site controls about itself does not help an AI system tell two sites apart. In his exact words, telling an AI system “I have the best website ever, and here are all of the pages that everyone must go to” does not differentiate one site from another.

Martin Splitt picked up the second half of the argument on Markdown rewrites. The Markdown version is meant to be token-efficient for an LLM, which sounds smart in isolation. The problem is that a stripped-down Markdown copy removes navigation, internal links, hierarchical cues, and accessibility markup. Those are the same signals search systems use for discovery and ranking. Trading them away for token efficiency moves your content further from how the platforms actually evaluate it, not closer.

Why this matters for your AI SEO program

The Google guidance pushes a clear position. Foundational SEO is the same investment that pays for AI search visibility. There is no parallel pipeline that requires a different file format, a separate ruleset, or a Markdown twin of your site. That is the same direction the rest of the platform has been moving for months. The AI search ranking signals framework for GEO and AEO we use already treats those acronyms as labels on the same discipline, not as separate practices.

The telemetry backs the official line on llms.txt for AI SEO. Ahrefs analyzed 137,000 domains and reported that 97% of /llms.txt files received zero requests in May 2026, according to Search Engine Journal’s roundup of the Ahrefs data. Only 28% of domains had even published the file. Of the requests that did happen, named AI retrieval bots like ChatGPT and Perplexity made up only about 1% of the fetches. The rest were audit tools and generic crawlers. If you shipped the file to influence ChatGPT or Perplexity citations, the read traffic is not there to support that assumption.

There is one nuance worth keeping. Anthropic has confirmed that Claude Desktop and claude.ai read /llms.txt in their retrieval flows, and coding agents like Cursor and Continue do the same. That is an application-layer use of the file by an agent that already has your domain in hand, not a discovery mechanism that gets you cited in the first place. Mueller’s point still holds: nothing about the file gets your site found by an AI that does not already know it exists.

If you also publish content for Google AI Mode or AI Overviews, the Munich AI Overviews liability ruling changes some of the publisher posture around AI citations. That is a separate set of policy decisions from this one. The two stack: the policy posture covers what you let AI systems do with your content once they read it, and the AI SEO posture covers whether you ship a /llms.txt file in the first place.

A practical decision tree for llms.txt for AI SEO

Use the following sequence to settle the question for any client account in under thirty minutes.

Step 1. Pull your server logs for the last 90 days. Look at requests to /llms.txt and any /llms-full.txt variant. Filter by user agent. Count requests from named AI retrieval bots like GPTBot, PerplexityBot, ClaudeBot, and Google’s various crawlers. If the named bot fetch count is in the single digits, you are inside the 97% group from the Ahrefs data and llms.txt for AI SEO is not driving discovery for you.

Step 2. Identify which AI surfaces matter most. If your priority surface is Google AI Mode or AI Overviews, Google has stated directly that the file is ignored. Maintaining it adds zero ranking signal. If your priority surface is Claude or Cursor, llms.txt for AI SEO has some application-layer utility because those products read it in their retrieval flows. Decide where the citations need to land before you decide whether the file is worth maintaining.

Step 3. Compare the maintenance cost against the upside. A well-maintained /llms.txt is not free. Someone has to keep it in sync with your real site structure as you publish, redirect, and retire pages. Pair that against the bot traffic you saw in step one. For most lead-gen and ecommerce sites, the maintenance cost outweighs the application-layer utility. For documentation-heavy SaaS sites that get real retrieval from coding agents, the file pays for itself.

Step 4. Use the saved time on foundational HTML signals. The same hour you would spend updating a /llms.txt file would do more if you used it to improve internal linking, write a better H1, or fix structured data. The four-pass GEO content audit workflow is the version of this work that we run on client accounts. Every signal it touches ranks on the same HTML graph the discovery pipeline reads. There is no parallel format the same hour could chase that would outperform that work.

Step 5. Write a one-line client policy on llms.txt for AI SEO. Pick a sentence and stick to it across accounts. Ours is “We will not maintain a /llms.txt file as an AI SEO tactic unless the account has a documented retrieval use case from a Claude or coding-agent surface.” Put it in your client report template so the question stops coming back. Add a link to the Google guidance change so a client who asks why has an answer they can read.

Step 6. Audit any account that already has the file. If a previous agency shipped a /llms.txt file you inherited, leave it in place. Removing it does no good and Mueller has been clear it does no harm to maintain it. The llms.txt for AI SEO decision matters for accounts where keeping the file in sync is consuming hours you could spend on signals the platforms actually use.

Where llms.txt for AI SEO sits next to AI Mode and AI Overviews

The wider story is consistency. Google’s VP of Search position from a separate June Think with Google post says the same thing: good SEO is good GEO, and you do not need a parallel optimization track for the AI features Google is shipping. The May 15 mythbusting section of the AI optimization guide names llms.txt, AI-specific markup, content chunking, and AI-only rewrites as practices Google’s own systems do not use. The Search Off the Record episode is the long-form version of the same message.

That consistency matters because the AI side of the platform is still shipping fast. The June 17 AI search console toggle changes what data you see for AI Mode and AI Overviews performance. The GA4 AI Assistant channel rollout from May 13 changes how the resulting clicks get bucketed in your reports. Neither change rewards a /llms.txt file. Both reward the same set of HTML, schema, and linking signals you have always optimized for.

The same logic applies to surfaces other than Google. If your account already runs into Facebook AI Mode brand monitoring or the Google AI search opt-out posture for publisher properties, your decision tree is the same. Optimize the HTML the platforms read, and decide separately about the consent or opt-out posture you want for AI systems. A /llms.txt file is not the lever for either decision. The lever is the structural quality of the pages you already publish.

One application use case to keep on your radar

There is a narrow case where /llms.txt has real upside. If you run a documentation-heavy SaaS product or a developer platform, coding agents like Cursor, Continue, and Claude Desktop do read the file in their retrieval flows. For those audiences, the file gives an in-product agent a faster map of your docs. That is application-layer help inside a tool the user is already using, not a discovery channel that brings new users in. If your account fits that profile, build the file carefully, keep it small, and treat it the same way you treat your sitemap.

The Chrome mismatch worth flagging on llms.txt for AI SEO is Lighthouse’s behavior. Chrome’s Lighthouse audit now checks for /llms.txt at the root of a site, even though Google Search ignores the file. The audit makes the file look like a missing best practice when it is not one for discovery. If a client reads a Lighthouse report and asks why their score dropped, the answer is that the audit is an application-agent signal, not a search signal. The two systems live inside Google but optimize for different things.

If you want a second set of eyes on whether /llms.txt belongs on your specific account before you spend another hour maintaining it, book a free consultation and we will walk through your server logs, your priority AI surfaces, and your real options together. Let’s Grow!