llm.txt or llms.txt? The Correct Filename, Settled

John Walter -
Quick answer. The correct filename is llms.txt — plural, with the s. llm.txt is a common misspelling and no spec, tool, or crawler looks for it. If you have already deployed /llm.txt, rename it to /llms.txt and nothing will be reading the old path, because nothing ever was.

Updated October 2026.

Search volume for llm.txt is substantial, which means a lot of people are about to ship a file at a path nothing requests. This page exists to settle the naming, explain why the confusion is so persistent, and point you at the right reference.

Want the format, an example, or llms-full.txt?

This page only settles the naming question. For the v2 format spec, a real annotated production file, the difference between llms.txt and llms-full.txt, and 15 days of server-log data on which crawlers actually fetch these files, read llms.txt Examples, Format, and llms-full.txt Explained.

Is it llm.txt or llms.txt?

It's llms.txt. The proposal — published by Jeremy Howard of Answer.AI on 3 September 2024 and revised to v2 in August 2026 — is named llms.txt throughout, the spec site is llmstxt.org, the reference Python package is llms-txt, and every file that real adopters serve is at /llms.txt: Stripe, Cloudflare, Vercel, Perplexity, Mintlify, Supabase, Anthropic.

There is no llm.txt specification. There is no tool that falls back to the singular path. Agents and crawlers that look for this file at all request the exact string /llms.txt, which means a file at /llm.txt is invisible by construction.

Why does everyone write llm.txt?

Four reasons, and they compound:

  • Every neighbouring filename is singular. robots.txt is the mental template, but you also have humans.txt, security.txt, ads.txt — and the acronym itself is singular. "LLM" reads as one thing, so "llm.txt" feels right.
  • The plural is semantically odd. The file isn't about multiple LLMs; it's addressed to language models in general. English doesn't make that distinction visible in a filename.
  • Google treats them as the same query. Search either spelling and you get results titled with the other, so nothing corrects you.
  • Google's own public comment used the singular. The Reddit thread John Mueller replied to was titled "LLM.txt – where are we at?", and his widely-quoted answer is still reproduced under that spelling.

None of that makes llm.txt correct. It just explains why the mistake is so evenly distributed.

What about llms.txt and llms-full.txt?

Those are the two filenames that actually exist in the wild:

PathReal?What it is
/llms.txtYes — specifiedA curated Markdown index of the pages you want agents to read
/llms-full.txtYes — convention, not specEvery page's full text concatenated into one file
/llms-ctx.txt, /llms-ctx-full.txtDeprecatedWere described in v1; removed from the proposal in v2
/llm.txtNoMisspelling. Nothing requests this path
/.well-known/llms.txtNoExplicitly rejected by the v2 spec

I already deployed llm.txt — what now?

Three minutes of work:

  1. Rename the file to llms.txt in your static directory — public/ in Next.js, static/ in Hugo, Docusaurus or Astro.
  2. Optionally 301 the old path to the new one. This is cosmetic; no client was requesting /llm.txt, so nothing breaks either way.
  3. Verify the content type. curl -sI https://yoursite.com/llms.txt should return 200 with text/plain or text/markdown. If it returns text/html, a framework catch-all is intercepting the path — the single most common real deployment bug with this file, affecting 40% of files Common Crawl measured.

Does llm.txt — or llms.txt — actually work?

The naming is the easy question. The harder one is whether the file does anything, and the honest answer is: much less than its promoters claim.

In 15 days of our own nginx logs for codersera.com — 1.8 million requests — the named crawlers from OpenAI, Anthropic, Google, Perplexity, Meta and ByteDance made roughly 190,000 requests to the site and fetched /llms.txt three times between them. Ahrefs, surveying 137,210 domains against May 2026 server logs, found that 97% of published llms.txt files received no requests at all that month. The clients that do read them are AI-native search startups and coding agents, not the big answer engines.

Full data, including the per-crawler breakdown, is in the complete llms.txt guide.

FAQ

Is llm.txt a real standard?

No. There has never been an llm.txt specification. The real proposal is llms.txt, published at llmstxt.org in September 2024 and revised in August 2026. llm.txt is a misspelling that propagated through search results.

Will AI crawlers find a file at /llm.txt?

No. The clients that request this file at all ask for the literal path /llms.txt, and Ahrefs found that no AI bot ever speculatively probes for alternative filenames — zero requests arrived for files that didn't exist. A file at /llm.txt will be fetched by nothing.

Should I publish both llm.txt and llms.txt?

No. Publish /llms.txt only. A duplicate at the singular path adds a second file to maintain and a second thing to go stale, in exchange for zero additional reach.

Is llms.txt the same as robots.txt?

No, and the difference matters. robots.txt is enforcement — per-User-Agent allow and disallow rules that every major AI company honours. llms.txt is an inclusion file with no enforcement semantics; it cannot block, restrict, or opt your content out of anything. If your goal is control over AI crawlers, robots.txt is the only file that does it.

What is llms-full.txt?

A community convention — not part of the spec — in which a site concatenates the full body text of every page into one file. Real ones run from about 570 KB to 64 MB, so most are too large for any model's context window and are realistically bulk-ingestion artifacts rather than something an agent reads inline.

Where should the llms.txt file go?

At the root of your domain, next to robots.txt: https://yoursite.com/llms.txt. Since v2 you may also publish subpath files such as /docs/llms.txt, which cover pages beneath that path, with the most specific file winning. The spec explicitly rejects /.well-known/.