Does llms.txt actually do anything?
Google has now answered this in its own documentation. The answer is no, and the file has been sold to a lot of people.
- Google's own documentation says Search does not use AI text files.
- No major AI crawler documents consuming llms.txt.
- It spread because it is cheap to sell and impossible for a buyer to disprove.
- What governs assistant access is robots.txt, HTML content and non-contradiction.
Does llms.txt do anything for Google Search or AI crawlers?
For Google Search, including its AI features, llms.txt does nothing. Google’s own documentation says so directly. No major AI crawler documents consuming it either. If someone quoted you for adding one, you paid for a file nothing reads.
What did Google actually publish about llms.txt?
Google Search Central’s AI optimisation guidance, updated 10 July 2026, states: “You don’t need to create new machine readable files, AI text files, markup, or Markdown to appear in Google Search (including its generative AI capabilities), as Google Search itself doesn’t use them.”
That is about as unambiguous as platform documentation gets. Three further published statements bear on the same file, and each one is set out below beside the source that published it.
| Source | Subject of the statement | What that source published |
|---|---|---|
| Google Search Central, AI optimisation guidance, updated 10 July 2026 | AI text files and Google Search | “You don’t need to create new machine readable files, AI text files, markup, or Markdown to appear in Google Search (including its generative AI capabilities), as Google Search itself doesn’t use them.” |
| Google staff, stated separately from the guidance | llms.txt and the keywords meta tag | Google staff have separately compared llms.txt to the keywords meta tag, on the grounds that no AI service used it and bots did not request the file. |
| OpenAI crawler documentation | robots.txt and published IP ranges | OpenAI’s crawler documentation describes robots.txt and published IP ranges. |
| OpenAI documentation index. The conflation named in the final clause is a Mindflow observation, not a published source. | OpenAI’s own llms.txt file | OpenAI does publish an llms.txt for its own documentation index, which is a different thing from its crawler consuming yours, and the two get conflated in vendor pitches. |
What we read from that, and this part is our judgement rather than anything the sources state: not one of these four published statements describes an AI crawler reading an llms.txt file on your site.
Sources: Google Search Central AI optimisation guidance, updated 10 July 2026; Google staff; OpenAI crawler documentation.
Why did llms.txt spread if nothing reads it?
llms.txt is cheap to produce, easy to sell and impossible for a buyer to disprove. A file that costs an hour and produces no measurable change is the ideal upsell, because nothing happens either way. It also arrived at the right moment, when owners were being told AI visibility was a new discipline requiring new artefacts.
The proposal that a new file was needed felt plausible.
What actually governs whether an assistant can use your page?
Four conditions decide whether an assistant can use your page, and all four are properties of the pages you already have: crawler access, content that is present in the HTML, an answer given early, and facts that do not contradict each other.
- Whether the crawler is allowed in. That is robots.txt, and it is worth checking which AI user agents you are blocking by accident.
- Whether the content is in the HTML rather than appearing only after script execution.
- Whether the page answers a question in the first eighty words, in language a model can lift without ambiguity.
- Whether your facts agree with each other across your site, your profile and the directories. Contradiction is the most common reason a business gets skipped.
None of that requires a new file. All of it requires the existing pages to be right, which is slower and harder to sell.
“A file that costs an hour and produces no measurable change is the ideal upsell, because nothing happens either way.”
Questions owners ask
Rankings are never guaranteed. Anything we could not trace to a primary source is absent from this page, not estimated. The audit runs the same six layers described on the pricing page, and Share of Answer is scored quarterly.
Youssef runs visibility work for established local businesses: Google Maps, organic search and AI answers, measured quarterly against a published protocol. Where a number cannot be checked, it does not go in the report.
More about Youssef