Skip to content
PublishSense

llms txt AI Search: What Google Actually Says It Reads

Hands sort handwritten call notes into three piles on a wooden worktable. (AI-generated image)

No. An llms txt AI search file will not get your practice quoted, because Google's John Mueller says no AI system fetches it. The chatbots that send patients crawl your ordinary pages instead. So spend that afternoon publishing the answers your front desk gave by phone this week.

What is llms.txt, and what is your agency actually selling?

An llms.txt file is a plain Markdown list of links to the pages you think matter most. It is a proposal, not an adopted standard, and it adds no new answers to your website.

The idea came from Answer.AI, which proposed a /llms.txt file with brief background and links to key pages. The authors had software documentation in mind. The llms.txt site itself says they expected it to help when a model assembles context, not in training. Neither page promises visibility in ChatGPT, Gemini or Google.

The difference between llms.txt and a sitemap is simple. A sitemap lists every URL so crawlers can discover your pages. An llms.txt file is a short, hand-picked list of what you say is important. Answer.AI pitched it as a curated alternative to crawling every page in the sitemap.

So the line on the invoice is a table of contents. It is cheap because a person, or an LLM, just reads your pages and writes a list. If your site has a few thin service pages, the list stays short and the answers stay missing.

Does an llms txt AI search file get a practice quoted?

No. Google’s John Mueller wrote that “no AI system currently uses llms.txt”. He added that consumer chatbots fetch your pages for training and grounding, but none of them fetch the file.

He made a sharper point on a Search Off the Record episode. The file is self-reported, Mueller argued, so an LLM cannot trust it to tell one website from another. Every practice in town could claim to be the best choice. The file gives the system no reason to believe any of them.

He also asked an awkward question. If an agency uses an LLM to build the file, Mueller pointed out, the same model could simply create it for itself. That is the whole case against paying for it. The file restates pages a machine can already read and understand.

What do ChatGPT, Claude and Perplexity actually read?

They read your ordinary HTML pages, the same ones a patient sees. What gets quoted is the answer you published, not a manifest listing answers you never wrote.

OpenAI says so in its crawler documentation. OAI-SearchBot surfaces websites in ChatGPT search, and OpenAI recommends allowing it in your robots.txt file. Perplexity works the same way. PerplexityBot follows robots.txt directives and builds an index of pages for its answers. Neither describes a step that looks for llms.txt.

Mueller says the proof sits in your own data: it shows up plainly in server logs. Chatbots request pages and never the file. For a practice, the useful work sits in two places:

  • Access: your robots txt rules and your host let the search bots in.
  • Answers: each page answers something a patient asked, in clear words an engine can lift.

What is WebMCP, and does a practice need it?

WebMCP is a Google-backed proposal that lets a site publish actions an AI agent can call, like booking or sending a form. Most practices don’t need to build it, but it points to the future of agents far better than llms.txt.

The Chrome team frames it as a way to expose structured tools for AI agents instead of leaving them to scrape screens. Without it, Search Engine Land explains, an agent hunts for a button and guesses at the fields. With it, the site tells the agent exactly what it can do.

Mueller said he likes the WebMCP approach because it has clear goals. He also noted that discovery and ranking are still tied to HTML. WebMCP helps an agent once it is on your site, not before. So ask your scheduling vendor whether they plan to support it. That is their product work, not an afternoon on your website.

How do you test your agency's llms.txt pitch?

Ask the agency to show you one AI platform that requested the file. If they cannot point to a single request in your server logs, the file is doing nothing.

The test follows Mueller’s own advice: make one when an AI platform that brings you clients asks for it. Until then, use this checklist with any agency or freelancer:

  • Logs: which bot fetched /llms.txt, and which URL did it request next?
  • Access: does robots.txt allow OAI-SearchBot and PerplexityBot?
  • Content: does each listed page answer a real patient question?
  • Gaps: what did patients ask this week that no page answers?
  • Measure: where will we see the result in Search Console or referral traffic?

A good agency will welcome these questions. If you run an agency, they make a stronger pitch than the file. The agency report puts researched topics under your own name.

What should you do with that afternoon instead?

Publish the three questions your front desk answered by phone this week. Each one becomes a page that ChatGPT, Gemini and Google can read, quote and send patients to.

I was a strength and conditioning coach for the Toronto Blue Jays, the Tampa Bay Buccaneers and the University of South Florida. Anyone who has coached knows a written plan earns nothing until the reps happen. A manifest of your pages is the plan. The published answer is the rep.

The afternoon, step by step

  • Ask the front desk for the three questions they repeated most, in the caller’s words.
  • Answer each one in the first sentence, then add what you would say in the room.
  • Link any figure to its source, and link the page where a patient can book.
  • Use the caller’s phrasing as the heading, because that is how people search.

The front desk questions guide covers which ones to answer first. The call log content plan turns a week of calls into a list. Our first app, Blog & Social Posts Done for You, has run on my own four sites since September 2026. It was built around that habit, the same work AI answer engines reward.

Frequently asked questions

Will Google penalize my site for having an llms.txt file?

Nothing Google has said points to a penalty. The file simply goes unread. Mueller suggests making one when an AI platform that brings you clients asks for it. Keeping an existing file costs nothing, and paying for a new one buys nothing. Treat it as neutral and move the budget to pages that answer patients.

Nothing Google has said points to a penalty. The file simply goes unread. Mueller suggests making one when an AI platform that brings you clients asks for it. Keeping an existing file costs nothing, and paying for a new one buys nothing. Treat it as neutral and move the budget to pages that answer patients.

My WordPress SEO plugin already made an llms.txt. Should I delete it?

You can leave it. Yoast added a one-click llms.txt generator, so many sites have one without asking. A file a plugin built for free does no harm. The mistake is calling it AI search optimization and assuming the work is done. The pages it lists may answer very few questions patients actually type.

You can leave it. Yoast added a one-click llms.txt generator, so many sites have one without asking. A file a plugin built for free does no harm. The mistake is calling it AI search optimization and assuming the work is done. The pages it lists may answer very few questions patients actually type.

Should I convert my practice website to Markdown for AI engines?

No. On Google’s podcast, Mueller and Martin Splitt agreed HTML is still the foundation for crawling and discovery. Generative systems read normal web pages well. A Markdown copy of a thin page is just as thin. Those hours are better spent answering one more question a patient asked.

No. On Google’s podcast, Mueller and Martin Splitt agreed HTML is still the foundation for crawling and discovery. Generative systems read normal web pages well. A Markdown copy of a thin page is just as thin. Those hours are better spent answering one more question a patient asked.

Is GEO different from SEO for a private practice?

For a practice, generative engine optimization and SEO come down to the same work. Both reward a site that answers real questions clearly. Both need pages a crawler can reach, written by someone who knows the subject. New acronyms tend to arrive with new invoices, so ask what will be published, not what the service is called.

For a practice, generative engine optimization and SEO come down to the same work. Both reward a site that answers real questions clearly. Both need pages a crawler can reach, written by someone who knows the subject. New acronyms tend to arrive with new invoices, so ask what will be published, not what the service is called.

Should agencies still offer llms.txt to clients?

An agency can include it at no charge, but selling it as AI visibility promises something no platform has confirmed. Clients eventually check server logs or ask why nothing changed. The safer position is to sell published answers and measured results. Tell clients plainly that the file is a speculative proposal.

An agency can include it at no charge, but selling it as AI visibility promises something no platform has confirmed. Clients eventually check server logs or ask why nothing changed. The safer position is to sell published answers and measured results. Tell clients plainly that the file is a speculative proposal.

The Bottom Line

Skip the llms.txt line item, or take it only as a harmless extra, and put the afternoon into answers. Your checklist right now is short:

  • Confirm robots.txt lets the search bots in.

If finding the hours is the real problem, get help choosing what to publish next, in your own voice. Try it.

Start free for 14 days

Start free for 14 days

Sources

Related on this site

Jim Goetz, Founder of PublishSense

All posts