Does llms.txt work?

• 97%of llms.txt files received zero requests of any kind in a monthAhrefs, ~38,000 files, May 2026
• 7llms.txt fetches by OpenAI crawlers in 12 weeks, against 3,990 robots.txt fetchesEZY.ai, 83 sites, 2026
• 10.13%of domains had an llms.txt, with no measurable link to AI citationsSE Ranking, ~300,000 domains, 2025

If “work” means getting your SaaS cited more often by ChatGPT, Perplexity or Google's AI features, the evidence so far says no. The file is cheap to make, so the real question is whether it deserves a place on your list ahead of the things that do move citations. Below is what the published studies and Google's own documentation say, where they are weaker than they look, and the narrow case where I think the file is worth having.

What llms.txt is meant to do

llms.txt was proposed by Jeremy Howard of Answer.AI in September 2024 [1]. It is a Markdown file at the root of a site (/llms.txt) that gives a language model a short map of the site: an H1 with the name, a one-paragraph summary in a blockquote, then H2 sections listing the most useful links with a note on each. A section called “Optional” marks links an agent can skip when it is short of context. The proposal also suggests publishing clean Markdown copies of pages at the same URL with .md added [1].

Two details matter for the “does it work” question. First, the proposal says the file is mainly meant for inference, when a model is answering a question, rather than for training [1]. Second, it is a proposal, not a standard: nothing obliges any AI company to fetch it, and robots.txt remains the file that controls access.

Adoption has grown fast anyway. Originality.ai, which monitors more than 3 million websites, counted 4,088 llms.txt files in June 2025 and 36,120 in May 2026, an 8.8-fold rise [6].

Do AI crawlers even request it?

This is the first test, and the easiest to measure: a file can only influence an answer if something reads it. Three independent log studies point the same way.

• Ahrefs, 137,210 domains, May 2026. 28% of domains in Ahrefs Web Analytics published an llms.txt, and 97% of those files received no requests at all during the month [2]. The remaining 3%, about 1,100 domains, shared roughly 22,000 requests. SEO audit tools made the most of them (21.7%). All AI bots together made 19.5%, and the bots that fetch pages to answer live questions, such as OAI-SearchBot and PerplexityBot, made 1.1%. Slackbot fetched llms.txt more often than PerplexityBot did [2].
• EZY.ai, 83 sites, 12 weeks in 2026. On the same sites over the same period, OpenAI's crawlers requested robots.txt 3,990 times and llms.txt 7 times [3]. The full comparison is in the table below.
• Flavio Longato, 1,000 Adobe Experience Manager domains, 30 days of CDN logs. No GPTBot, ClaudeBot or PerplexityBot requests for llms.txt were seen at all; Googlebot made 94.9% of the hits [4].

The pattern is consistent: the crawlers that decide what ChatGPT, Claude and Perplexity can see check robots.txt constantly and llms.txt almost never. Meta's crawler is the one exception in the EZY data. EZY.ai sells llms.txt generation, so this is a result that cuts against its own product, which makes it more credible rather than less.

One detail in the Ahrefs data is worth noticing. Among AI bots, Claude-Code, Anthropic's coding agent, was the third most frequent requester, ahead of every AI search and assistant bot except GPTBot and one indexer [2]. That fits the original idea of the file: a developer's agent looking for documentation, not a search engine choosing which vendor to recommend.

Does it change how often you are cited?

The largest test so far is SE Ranking's analysis of about 300,000 domains, reported by Search Engine Journal in November 2025. SE Ranking found an llms.txt on 10.13% of domains, spread fairly evenly across traffic tiers [5]. It then compared each domain's citation frequency in answers from major language models against a set of site factors, using simple correlation tests and a machine-learning model (XGBoost).

The result: no significant correlation between the file and citations, and removing the llms.txt variable from the model made its predictions slightly more accurate [5]. In other words, knowing whether a site had the file added nothing to explaining how often it was cited. SE Ranking's own wording was that it does not seem to affect citation frequency “at least not yet”.

The log studies explain why. If the AI search crawlers do not request the file, it cannot shape what they retrieve, and a correlation would be surprising.

What Google has said

Google has been unusually direct about this, more than once:

• April 2025. John Mueller of Google Search compared llms.txt to the keywords meta tag, a site owner's claim about what the site is about, and noted that server logs show AI services do not even check for it [7].
• June 2025. Mueller on Bluesky: “FWIW no AI system currently uses llms.txt.” [8]
• 2026, official documentation. Google's guide to generative AI features in Search lists llms.txt under things you do not need to do. Google Search “ignores them”, and creating one will neither help nor harm your visibility in Search, including its generative AI features [9]. The same guide says the best practices for SEO continue to apply, because the AI features are built on Google's core ranking systems.

There is one complication. By May 2026, Chrome's Lighthouse tool had an llms.txt audit in its “agentic browsing” category [10]. Read the detail, though: a missing file is marked “not applicable”, because providing one is optional, and the audit only flags a page when the server returns an error for the file [10]. Search Engine Journal described it as Google's guidance depending on which product you ask [11]. For Google Search, the documentation is clear. The Lighthouse audit is about agents browsing a site, not about rankings or citations.

How strong is this evidence?

Strong enough to act on, but not perfect, and it is worth knowing the weak points before you repeat any of these numbers.

• A request is not the same as use. Ahrefs counts fetches. A fetched file may never be used in an answer, so every figure is a ceiling on real use, as Ahrefs itself says [2].
• The samples are not the whole web. Ahrefs' customers skew technical, so its 28% adoption figure is an upper bound [2]. Longato's 1,000 domains all run one enterprise CMS [4]. EZY's 83 sites lean towards small businesses that had sought an AI visibility tool, and it publishes only aggregates [3].
• Correlation cannot prove a negative. SE Ranking shows no measurable relationship at domain level [5]. That does not rule out a small effect on particular sites, but the burden of proof sits with whoever is selling the effect.
• The SE Ranking figures come through Search Engine Journal's report. I could not read SE Ranking's own page when checking this article, so the numbers here are as SEJ reported them.
• This can change. If OpenAI, Anthropic or Perplexity announce that their search systems use llms.txt, the picture changes overnight. I have not seen any of them say so, and OpenAI's crawler documentation is built around robots.txt [5].

When an llms.txt is still worth having

My view: there is one real use case, and it is not marketing. If your product has public developer documentation, such as an API, SDK or integration guides, an llms.txt plus Markdown copies of the docs is a reasonable courtesy to developers whose coding agents read those docs. The Claude-Code requests in the Ahrefs data [2] suggest that this audience exists. The cost is an hour, and the benefit is to people already using your product, not to your share of AI recommendations.

For the marketing site of a B2B SaaS company, I would not bother, and I would be wary of anyone who sells it as a citation booster. If you already have one because a plugin made it, there is no need to remove it: Google says it neither helps nor harms [9]. Two things to check if you keep it. It should not contradict your pricing or positioning pages, because a stale summary is worse than none. And it should return a clean 200 or a 404, not a server error, which is the one thing the Lighthouse audit flags [10].

What to do instead

The time is better spent on things that change what AI assistants actually read:

1. Check crawler access. Make sure robots.txt and your CDN or firewall allow the search crawlers, such as OAI-SearchBot and PerplexityBot. The log studies show these bots read robots.txt all the time [3], and blocking them is the one file-level mistake that does remove you from answers.
2. Put the important facts in plain HTML. Pricing, features and who the product is for should be readable without JavaScript. That serves every crawler, with or without an llms.txt.
3. Work on the pages that get cited. For buyer questions, most citations go to third-party lists, comparisons and forums rather than your own site. The steps are in how to get your SaaS recommended by ChatGPT.
4. Measure before and after. If someone claims a file or tactic lifted your citations, ask for the prompt set, the number of runs and the margin of error. The method page shows how I measure.

Why this site has no llms.txt

My view: LiftInAI has no llms.txt on purpose. It has no developer documentation, so the one case where I think the file helps does not apply. Adding one anyway would suggest it matters for AI visibility, which is the claim this page argues against. The robots.txt does the work that matters: it names and allows each AI crawler individually. If the major assistants start using llms.txt for search, I will add one and update this page.

Questions

Does llms.txt help with ChatGPT?

There is no evidence that it does. OpenAI has not said its search uses the file, and in a 12-week log study across 83 sites, OpenAI's crawlers requested llms.txt 7 times against 3,990 requests for robots.txt. What controls whether ChatGPT search can see you is allowing OAI-SearchBot in robots.txt.

Does Google use llms.txt for AI Overviews?

No. Google's documentation on generative AI features in Search says Google Search ignores llms.txt and similar files, and that creating one neither helps nor harms visibility. Chrome's Lighthouse tool has an llms.txt audit for agentic browsing, but that is separate from Search rankings.

Can llms.txt hurt my site?

Not in Google Search, according to Google. The practical risks are smaller: a summary that goes out of date and contradicts your pricing page, or a server error when the file is requested, which the Lighthouse audit flags. If you keep one, keep it accurate.

What is the difference between llms.txt and robots.txt?

robots.txt tells crawlers what they may access, and AI crawlers check it constantly. llms.txt is an optional Markdown summary meant to help a model find a site's most useful pages. AI crawlers rarely request it, according to every log study published so far.

Should a SaaS company with an API publish llms.txt?

In my view, yes for the developer documentation, together with Markdown versions of the docs. Coding agents do request these files, and it helps developers who are already building on your product. It is not a way to be recommended more often.

Want to know what actually moves your citations?

The AI visibility audit runs your buyers' prompts across ChatGPT, Claude, Perplexity and Google AI Overviews, lists the pages that get cited, and turns it into a prioritised backlog. The scope is free and comes within one working day.

Get a free scope →

Sources

Checked on 2 November 2026. Figures are quoted as published; samples and periods differ between studies.

1. llmstxt.org — The /llms.txt file, proposed by Jeremy Howard (3 September 2024)
2. Ahrefs — We analyzed 137K sites: 97% of llms.txt files never get read (15 June 2026)
3. EZY.ai — Do AI bots read llms.txt? 83 sites, 27 April to 19 July 2026 (27 July 2026)
4. Flavio Longato — LLMs.txt: why AI crawlers ignore it, 1,000 domains (August 2025, updated June 2026)
5. Search Engine Journal — LLMs.txt shows no clear effect on AI citations, based on 300k domains (SE Ranking data, 20 November 2025)
6. PPC Land — llms.txt adoption rises 8.8x (Originality.ai data, 2 July 2026)
7. Search Engine Journal — Google says llms.txt comparable to keywords meta tag (17 April 2025)
8. Search Engine Roundtable — Google: no AI system currently uses llms.txt (18 June 2025)
9. Google Search Central — Optimizing your website for generative AI features on Google Search (last updated 10 July 2026)
10. Chrome for Developers — Lighthouse llms.txt audit (last updated 5 May 2026)
11. Search Engine Journal — Google's llms.txt guidance depends on which product you ask (20 May 2026)