What is generative engine optimization (GEO)?
How AI assistants choose which sources to cite, and what you can change.
On this page
Short answer
llms.txt is a small Markdown file at the root of a website that gives AI tools a short map of the site: whose site it is, what it covers and which pages to read first. It is a proposal that anyone may follow, not a rule that search engines enforce. For a company that sells services, it is a quick task with an unproven payoff, so it belongs after the work on your real pages.
Where the file came from
Jeremy Howard of Answer.AI published the idea on llmstxt.org in September 2024. His reasoning is simple. A model that reads a site while doing a task has limited room, and an ordinary web page buries its useful text under menus and scripts. A short file written for machines saves that room and points the tool at the right pages.
The proposal goes one step further: it suggests offering a clean Markdown copy of each useful page at the same address with “.md” added. Few business sites do this, and you can skip it.
What goes inside
The spec makes only one line mandatory: a top heading with the name of the site or project. Everything else is optional, but it has a fixed order.
| Part | What it holds | Example for a dental clinic |
|---|---|---|
| H1 heading | the name | # Example Dental Clinic |
| Quote block | a one or two sentence summary | > Family dentistry and implants, two locations |
| Plain text | extra notes, no headings | languages spoken, insurance accepted |
| H2 sections with links | groups of pages, one link per line, a note after a colon | [Implants](https://example.com/implants/): prices and steps |
| Section named “Optional” | links a tool may skip when short on room | blog archive, careers |
The file can also sit in a folder, such as /docs/llms.txt, and then describes only that part of the site. When several files apply, the spec says the most specific one wins.
Who reads it today
Here is what we could confirm in first-hand sources as of October 2026.
- Google Search. Google’s documentation on AI features says you don’t need new machine-readable files, AI text files or special markup to appear in AI Overviews or AI Mode. In an informal Bluesky post in June 2025, Google’s John Mueller wrote that no AI system currently uses llms.txt. That is his observation, not a statement from the AI companies.
- Chrome Lighthouse. Lighthouse now has a group of experimental audits for agentic browsing, and one of them looks at llms.txt. According to Chrome’s documentation, a site without the file is marked “not applicable”. The audit flags a server error when the file is requested. So Lighthouse checks that the file works if you have one. It does not reward you for having it.
- AI companies. Anthropic, OpenAI and Google’s Gemini API team publish llms.txt files for their own developer documentation. We opened all three on October 29, 2026. That makes sense for docs that coding tools read. None of these companies says its assistant uses other sites’ llms.txt to pick sources for answers.
- Site builders and plugins. llmstxt.org lists tools that create the file for you, including Yoast SEO, AIOSEO and Wix.
Our reading: the file has real use on software documentation. For a clinic, a law firm or an agency, we know of no public evidence that it brings more mentions or citations in AI answers.
The proposal goes one step further: it suggests offering a clean Markdown copy of each useful page at the same address with “.
What this means for a service business
When a buyer asks ChatGPT with search, Perplexity or Google about your field, these systems run a search and read your normal pages. So the pages decide the outcome.
Our own site is a small example. On September 22, 2026, Google AI Overviews linked to aoe.kz in 0 of the 17 queries we track. By October 6 the count was 2 of 20 (three queries were added to the set). The changes behind it were on the pages: a direct answer in the first paragraph, comparison tables, a block of questions. The first two links were seen at the September 29 check. The llms.txt file entered the aoe.kz site code on October 5, according to the repository history, six days later, so it could not have caused them. The queries were Russian-language searches in Almaty, measured on desktop. How we measure this is described on our AI search optimization page.
If you add one: a checklist
- Write the H1 with your business name exactly as it appears on your profiles and invoices.
- In the quote block, say what you do, where and for whom, in the same words as your home page.
- List 5 to 15 pages: main services, prices, contact, about, and your strongest guides. Add one plain note to each link.
- Leave out old campaign pages, tag archives and thin pages.
- Open yourdomain.com/llms.txt in a private browser window. It should show plain text with status 200, not redirect to the home page and not return an error page.
- Run Lighthouse once to confirm the audit reports no server error.
- Put the file on your update list. When a price or address changes, the file changes the same day. A stale llms.txt is one more place that states a wrong fact.
Common mistakes
- Treating it as robots.txt. llms.txt does not allow or block any crawler. Access rules still live in robots.txt.
- Expecting a ranking boost. Google says the file isn’t needed for its AI features. Nothing in Google’s documentation ties it to rankings.
- Pasting the whole sitemap. A list of 800 URLs defeats the purpose of a short map. Your XML sitemap already does that job.
- Facts that disagree with the site. If the file names a service you dropped last year, you have added a source of errors.
- Paying for it. An “llms.txt optimization” package is a 20-minute job. Spend the budget on pages and on being named on other sites.
When a buyer asks ChatGPT with search, Perplexity or Google about your field, these systems run a search and read your normal pages.
FAQ
Does Google use llms.txt?
Not for Search. Google’s AI features documentation says no special AI text files are needed, and John Mueller said in a 2025 post that, in his view, no AI system uses the file. The only Google product we found that looks at it is Lighthouse, which checks whether the file loads without errors.
Will ChatGPT or Claude read my llms.txt?
OpenAI and Anthropic publish their own files for developers but have not said their assistants use other sites’ files to choose sources. A tool may fetch it when a user points it at your site. That is a narrow case, not a route into answers.
How is it different from robots.txt and the sitemap?
robots.txt tells bots what they may crawl. The sitemap lists every page you want indexed. llms.txt is a short, human-written summary for AI tools. None of the three replaces the others.
Should we add one anyway?
If your site platform can create it in a few clicks, yes, and keep it accurate. If it means a developer ticket, put it behind the page work in our AI citations guide. Short definitions of related terms are in the AI search glossary.
