Who really supports llms.txt?
The picture is less clear-cut than many vendors claim, and more nuanced than the headline “Google doesn’t want it”. Here is the status per system, every row with a source and a date.
The short version
Not one vendor guarantees it reads your llms.txt
Google says no, explicitly. Its official guide to generative AI features in Google Search (15 May 2026) states that additional machine-readable files “neither harm nor help your site’s visibility or rankings”.
The others say nothing but use the format themselves. Anthropic, OpenAI, Perplexity, Mistral and xAI all serve their own developer documentation as an llms.txt. A company that picks a format for its own docs presumably considers it useful.
What follows: an llms.txt does not improve your ranking. It is an offer to the systems that fetch it, and it helps you organize your content so that any machine processing benefits. Promises beyond that are not backed by evidence.
Status per system
Official stance and who runs an llms.txt themselves
Two questions that constantly get mixed up: does the vendor say it reads llms.txt? And does it use the format itself? We check the second column automatically every day, and you can open every URL yourself.
| System | Official stance | Publishes one? |
|---|---|---|
| Google SearchAI Overviews · AI Mode | Explicitly does not use llms.txt. In its official guide to generative AI features, Google states you do not need “new machine readable files, AI text files, markup, or Markdown” to appear in Google Search: such files “neither harm nor help your site’s visibility or rankings”. Google Search Central, generative AI guide · May 15, 2026 - publishes none (HTTP 404) |
- HTTP 404 |
| Anthropic · ClaudeClaude, Claude Code | Uses the format itself, but guarantees no fetching. Anthropic publishes its own developer documentation as an llms.txt and, in its engineering post “Writing effective tools for AI agents”, points to the format as the common place for LLM-friendly documentation. That is not a commitment to automatically fetch third-party llms.txt files. Anthropic Engineering · Aug 19, 2026 ✓ publishes an llms.txt · view |
✓ view |
| OpenAI · ChatGPTChatGPT search, OAI-SearchBot | No official commitment, but publishes one itself. OpenAI has never committed to consuming llms.txt. Retrieval runs through its own crawlers (OAI-SearchBot, ChatGPT-User) and through the Bing index. OpenAI’s own developer documentation, however, is served as an llms.txt. OpenAI developer docs · Aug 19, 2026 ✓ publishes an llms.txt · view |
✓ view |
| PerplexityPerplexityBot | No official commitment, but publishes one itself. Perplexity crawls sites in the classic way and makes no statement about evaluating llms.txt specifically. Its own documentation is available as an llms.txt. Perplexity docs · Aug 19, 2026 ✓ publishes an llms.txt · view |
✓ view |
| Microsoft Copilot · BingBing index, Copilot | No commitment: the Bing index is what counts. For Copilot and for ChatGPT’s web retrieval, what matters is whether a page is in the Bing index, not whether it has an llms.txt. Since February 2026 Microsoft has shown a dedicated AI citation report in Bing Webmaster Tools. Bing Webmaster Tools · Feb 10, 2026 - publishes none (HTTP 404) |
- HTTP 404 |
| MistralLe Chat | No commitment, publishes one itself. No public statement on consuming third-party files; its own documentation is provided as an llms.txt. Mistral docs · Aug 19, 2026 ✓ publishes an llms.txt · view |
✓ view |
| xAI · GrokGrok | No commitment, publishes one itself. No public statement on consuming third-party files; its own documentation is provided as an llms.txt. xAI docs · Aug 19, 2026 ✓ publishes an llms.txt · view |
✓ view |
Stance column editorially reviewed on Aug 19, 2026. Live column: 5 of 7 vendors checked so far serve an llms.txt on their documentation domain. Automated check, every 24 hours.
What to do instead
Four points with documented effect
What matters more than llms.txt is what the systems demonstrably use to pick sources:
The most common mistake: robots.txt blocks GPTBot or ClaudeBot, often unnoticed via a separate user-agent group. Without access, the other points have no effect.
AI crawlers do not execute JavaScript. They don’t see what is not in the raw HTML, even if the page ranks first on Google.
A machine-readable modification date and real updates. Answer engines prefer demonstrably fresh sources.
Clear headings, lists, tables, sourced numbers. RAG systems cite individual passages, and long unstructured text works poorly for that.
We check all four automatically: the AI crawler check for access, the AI check including citability for the rest. Both are free.
FAQ
llms.txt and the AI systems
Does Google read my llms.txt?
No. In May 2026 Google explicitly stated that additional machine-readable files are neither required nor helpful for visibility in Google Search and its AI features. So an llms.txt is not a way to gain visibility on Google.
So why have an llms.txt at all?
There are reasons that apply regardless of Google. Assistants and developer tools fetch the file when they know about it. A good llms.txt also forces you to structure your content clearly, which helps any kind of machine processing. And the AI vendors use the format for their own documentation, which suggests it will stay in use.
Where does the “publishes one” column come from?
From an automated check: once a day we request each vendor’s documentation domain at /llms.txt and record whether a valid file is served. You can open the linked URLs yourself to verify this.
How current is this page?
The stance column is editorially maintained and carries a source and date per row. The live column checks itself daily. If an official statement changes, we update the row.
Does llms.txt replace robots.txt?
No, it complements it. robots.txt governs what crawlers may not do; llms.txt offers what should be read. If your robots.txt blocks AI crawlers, even a good llms.txt won’t help. This is one of the most common mistakes.
First check whether AI crawlers have access
The crawler check reads your robots.txt against 20 AI crawlers, with real Google semantics, including the inheritance trap. It is free and needs no signup.