What Is llms.txt and Do You Need One?

What llms.txt is

llms.txt is a proposal, not a standard. Jeremy Howard published it in September 2024, and the llmstxt.org page, last modified in August 2026, still calls itself a proposal open for community input. The idea is simple: put a Markdown file at the root of your site that gives a language model a short overview and links to the pages that matter, so an agent does not have to untangle cluttered HTML.

The format is strict about one thing and relaxed about the rest. It needs an H1 with the site or project name. After that it allows a blockquote with a short summary, optional free text, and H2 sections holding lists of links in the form [name](url) with an optional note. By convention, a section called "Optional" holds secondary links that can be skipped when space is short. The proposal also suggests clean Markdown versions of pages at the same address with .md added.

What the evidence shows in November 2026

The honest summary is that llms.txt has no confirmed effect on the major assistants, and nobody can guarantee a citation from adding one.

• Google. Its generative AI guide lists llms.txt among things you do not need. It says Google Search does not use such files, though creating them for other services will not harm your Search visibility. Search Engine Journal reports that John Mueller compared the file to the old keywords meta tag and said AI services were not using it. Chrome's Lighthouse 13.3 added an experimental audit for it, in a separate agent-focused category that does not claim to improve rankings or citations.
• A large-sample study. SE Ranking looked at about 300,000 domains. Only 10.13% had the file. Using correlation tests and a machine-learning model, it found no relationship between having llms.txt and being cited by AI systems, and removing the variable made the model more accurate. The limits are real: the write-up does not name the platforms sampled or the exact collection method, and the authors say results depend on context. They still call the file optional for now.
• The assistant companies. None of the OpenAI, Anthropic or Perplexity crawler documentation we read says their search crawlers use llms.txt to choose citations. Absence of confirmation is not proof of absence, which is why we say "no confirmed effect" and not "no effect".

Everything above can change. If a major assistant documents support, this page will be updated.

Examples

1. The smallest valid file.

# Example Studio

That is valid under the proposal, because only the H1 is required. It is also useless to a reader.

2. A small-business file (illustrative). The business below is invented.

# Harbour Dental Limassol

> Private dental clinic in Limassol offering check-ups, hygiene, implants and emergency appointments. Prices published on the pricing page.

## Services
- [Check-ups and hygiene](https://example.com/check-ups): what a first visit includes and costs
- [Dental implants](https://example.com/implants): process, timeline, price range
- [Emergency appointments](https://example.com/emergency): same-day slots and how to book

## Practical
- [Prices](https://example.com/prices): current price list
- [Contact and opening hours](https://example.com/contact): address, phone, hours

## Optional
- [Blog](https://example.com/blog): background articles

Notice what the file does well: every note says what the reader will find, the prices are one click away, and nothing in it is a claim. It does not say "best clinic in Cyprus". An assistant that did read it would get a map of the site, not an advertisement.

3. A generated file. Our own file is produced automatically from the site's page list. It follows the same structure: the site name as H1, a blockquote with the description, a Services section, one section of guides per topic, and a Company section. Generation keeps the file from drifting out of date, which is the commonest way a hand-written one goes wrong.

Do you need one?

Use this decision rule.

• Add one if your site is documentation, an API reference or a knowledge base that developers or agents may read at answer time, or a plugin creates it in a few clicks and you accept that you should expect nothing in return.
• Skip it for now if you have not yet checked crawl access, indexing and the quality of your key pages. Those have documented effects. This does not.
• Never pay for it. A package that charges for an llms.txt file is selling something free and unproven.

If you add one, keep it honest. List pages that exist and are indexable. Rank Math, for one, leaves noindex pages out. Update it when pages change, and do not use it to hide or promote pages in ways that your visible site does not support.

llms.txt, robots.txt and sitemap.xml are different tools

People confuse the three because all of them sit in the root of a site. They do different jobs, and only two have documented support from the companies that run the engines.

If you must choose where to spend an hour, robots.txt wins. A wrong line there can remove you from ChatGPT search, Claude's search and Perplexity at once, and each of those companies says it reads the file.

A way to find out whether it matters for you

You do not have to take our word, or anyone else's. If you add the file, you can measure whether anything reads it.

1. Publish the file and note the date.
2. For 60 days, filter your server or CDN logs for requests to the llms.txt address.
3. Group the requests by user agent. Compare them with the list of known search and user-triggered bots from OpenAI, Anthropic, Perplexity, Google and Bing.
4. Add a line for each of your tracked prompts to your monthly log, as described in how to track your brand in AI answers, and compare the mention rate before and after.

Two caveats apply. A fetch shows that something read the file, not that it changed an answer. And a before-and-after comparison cannot separate the file from every other change you made. Treat any single-site result as an anecdote. If the log stays empty, that is itself a useful answer.

What to do before llms.txt

1. Confirm your key pages are indexed.
2. Check robots.txt and firewall rules for the search bots of OpenAI, Anthropic and Perplexity.
3. Rewrite your top pages so each section opens with its answer.
4. Record a baseline of what assistants say about you.

If, after that, you want the file, our WordPress how-to takes about five minutes. The LLM SEO explainer shows where the file sits among the other levers, and the AI search optimization guide ranks them by evidence. If you want a second opinion on priorities, our AI search service starts with an audit and quotes in writing. Check your robots.txt first.