llms.txt: What It Is, Who Actually Reads It, and Whether You Need One in 2026

llms.txt is the most oversold file in AI optimization. Agencies list it as a deliverable, generators mint thousands of them, and yet the measured reality in 2026 is blunt: 97% of published llms.txt files never receive a single request, and no study has found a citation effect. We build AI visibility software and we still won't score it. Below: what the file is, who reads it, and the narrow case where it earns its keep.

What llms.txt is

llms.txt is a proposal published in September 2024 by Jeremy Howard of Answer.AI. The idea: place a markdown file at your site root (yourdomain.com/llms.txt) containing a one-paragraph summary of the site and a curated, annotated list of links to its most important pages. Context windows are limited and HTML is noisy, so the file hands a language model a clean map of what matters: a table of contents written for machines. A companion variant, llms-full.txt, goes further and inlines the full text of those pages into one large file.

Two things llms.txt is not. It is not access control: unlike robots.txt, it grants and denies nothing, and no crawler is obliged to look at it. And it is not a standard: it remains a community proposal that no major AI vendor has formally adopted. Both points matter for what follows.

What the file looks like

The format stays minimal: plain markdown, readable by you, parseable by a machine. A complete llms.txt for a small SaaS fits in a dozen lines:

# AcmeBooks
> Accounting software for freelancers in the EU. Invoicing, VAT returns, bank sync.

## Product
- [Features](https://acmebooks.example/features): what the product does, plan by plan
- [Pricing](https://acmebooks.example/pricing): current prices and billing terms

## Docs
- [Getting started](https://acmebooks.example/docs/start): setup in 10 minutes

An H1 with the site name, a blockquote summary, then sections of annotated links. You have now read the whole specification. The format is sensible. The question that decides whether you should bother is different: does anyone on the consuming side open the file?

Who reads it in 2026: the evidence

This question has been studied at scale, and the results are consistent.

97% of llms.txt files receive zero requests. Ahrefs analyzed 137,210 domains in 2026: about 28% had published a valid llms.txt, but almost none of those files were ever fetched. Of the requests that did arrive, 96% came from bots, and 77% of that bot traffic was SEO auditing tools checking whether the file exists rather than AI systems reading it. Named AI retrieval bots (OAI-SearchBot, PerplexityBot) accounted for roughly 1% of requests.

No measurable citation effect across ~300,000 domains. SE Ranking tested whether sites with llms.txt get cited by AI more often than sites without it, and found no statistically significant relationship. Removing the variable from their model improved its accuracy — the file behaved as noise, not signal.

The platforms' own statements point the same way. Google's Search Central documentation now says outright that Google Search doesn't use such files; Gary Illyes confirmed in July 2025 that Google has no plans to support llms.txt, and John Mueller compared it to the keywords meta tag — a self-declared signal search engines stopped trusting two decades ago. OpenAI, Anthropic, Meta, Perplexity and Mistral have made no public commitment to reading llms.txt in production. GPTBot has been observed fetching the file on some sites, but OpenAI has never said it influences answers. Even Anthropic — which publishes an llms.txt for its own documentation — has never committed to reading yours.

So who does read it? Coding assistants. Cursor, Windsurf, Claude Code and GitHub Copilot can pull a documentation site's llms.txt or llms-full.txt to load accurate, current docs into context. That is a real, observable use case — and it is essentially the only one. If you run developer documentation, llms.txt has a genuine audience. If you run a dental clinic, a law firm or an online store, effectively nobody is fetching your file.

Why CitePulse detects llms.txt but doesn't score it

A CitePulse audit checks whether your llms.txt exists and is well-formed, and reports it as information, with zero weight in your AI Reputation Score. That is a deliberate choice, and it separates us from tools that list llms.txt as a ranking factor.

The reasoning is simple: we score what the evidence says influences AI answers. Crawler access, JavaScript-free content, schema.org, concrete claims and third-party corroboration all have observable mechanisms and measurable effects — they're what our technical checklist for AI citability is built on. llms.txt currently has neither: the files aren't fetched, and their presence doesn't correlate with citations. Scoring it would inflate reports with a checkbox that feels productive and changes nothing. We sell the measurement and the fixes, not guaranteed AI citations — and honest measurement means refusing to score theater. If the evidence changes — if a major assistant starts consuming llms.txt at answer time — we will start weighting it, and we'll say so publicly.

Should you create one anyway? The verdict

If you publish developer documentation: yes. Coding assistants consume it, it takes minutes, and docs platforms like Mintlify generate it automatically. This is the one clear win.

For everyone else: it's optional and last on the list. A correct llms.txt costs about fifteen minutes and carries no known penalty, so there's no reason to fear it — there's just very little reason to expect anything from it. Make it accurate, keep it consistent with your site, and treat it as a lottery ticket on future adoption, not a fix. If any vendor tells you llms.txt is why AI doesn't cite you, that tells you more about the vendor than about your site.

If you want one, don't pay for it: our free tools include an llms.txt generator (alongside the schema and robots.txt checks that move your score). Spend the time you save on the blockers with evidence behind them — starting with whether AI crawlers can read your site at all, which is the first thing a citation check reveals.

FAQ: llms.txt in practice

Does llms.txt help SEO or Google rankings?

No. Google has stated plainly that Google Search does not use llms.txt, Gary Illyes confirmed in July 2025 that Google has no plans to support it, and John Mueller compared it to the long-dead keywords meta tag. Whatever llms.txt may become, it is not an SEO signal today.

Does llms.txt replace robots.txt?

No. They do opposite jobs. robots.txt controls access (which crawlers may fetch which paths) and major AI crawlers honor it. llms.txt is an invitation: a curated reading list with no access-control power at all. If you only maintain one of the two, maintain robots.txt.

Will adding llms.txt get my business cited by ChatGPT or Perplexity?

The evidence says no. Ahrefs found 97% of llms.txt files receive zero requests, and SE Ranking's ~300,000-domain analysis found no statistically significant relationship between having the file and being cited by AI. Citation depends on readable content, concrete claims and third-party corroboration — not on a file assistants don't fetch.

What is the difference between llms.txt and llms-full.txt?

llms.txt is a short markdown index: a summary of the site plus links to its most important pages. llms-full.txt inlines the full content of those pages into one large file, so a coding assistant can load an entire documentation set in a single request. Both live at the site root; both are part of the same proposal.

I already have an llms.txt file. Should I delete it?

No. There is no evidence of harm, and a correct file costs nothing to keep. Verify it doesn't contradict your actual site, update it when your key pages change, and don't let it displace the work that measurably matters: crawler access, server-rendered content, schema.org and third-party presence.

Related reading: the technical checklist for AI citability · why ChatGPT doesn't recommend your business · what is answer engine optimization (AEO).