llms.txt: Is It Worth Creating? What We Know in 2026

llms.txt is a Markdown file at the root of a site, proposed in September 2024 by Jeremy Howard to point AI agents to the most useful pages. In 2026, Google says Search ignores the file, and public studies find no relationship between having one and being cited by AI. It makes sense for technical documentation and agents. For a blog that wants citations, it is low cost and an uncertain return.

Quick summary

• In the world of SEO, AEO and GEO, llms.txt is a community proposal, not a standard adopted by search engines. It is maintained by its own author at llmstxt.org.
• Google says it does not use the file and that creating one neither helps nor hurts visibility in Search.
• An SE Ranking study of about 300,000 domains found no correlation between having llms.txt and being cited by LLMs.
• The most defensible use is technical documentation, where coding agents need to navigate references and tutorials.
• If you create one, keep it small, current and honest. A file that contradicts the site is worse than none.
• Before any new file, get the basics right: a correct robots.txt, indexable pages and content that answers well.

How to read this: if you run a business and just want to know whether you need this, read the summary and the scenario table. If you work in SEO or development, the sections on the specification, the evidence and log verification give you what you need to decide.

What is llms.txt, and what does it promise?

llms.txt is a plain text file in Markdown, published at /llms.txt, that gives an AI agent a curated view of the site: a summary and a list of links to the most important pages. The proposal starts from the idea that HTML pages have menus, ads and scripts that make them hard for machines to read.

According to the specification published at llmstxt.org, the file can sit at the root or in any subpath, and it covers the URLs under that path. The accepted structure, in order:

1. An H1 title with the name of the site or project (the only required part).
2. A blockquote with a short summary.
3. Free-form text sections with more context.
4. H2 sections with lists of links, each with an optional description.

The proposal also suggests publishing Markdown versions of important pages, with the .md extension on the same URL, and using the rel="alternate" and rel="describedby" relationships to point to them. It describes itself as a complement to robots.txt and the sitemap: the sitemap lists every page, and llms.txt offers an annotated selection.

In plain English: think of a map at the reception desk of an office building. It only helps people who walk in and decide to look at it. If visitors (the AI bots) never stop to read it, the map just sits there with no effect, no matter how well drawn it is.

What does Google say about llms.txt?

Google says Search does not use the file and that it neither helps nor hurts visibility. In its guide to optimizing for generative AI features, the myths list states that you do not need to create machine-readable files, AI text files, markup or Markdown to appear in Google Search, since Search itself does not use them. The text adds that creating the file for other services is acceptable.

This has been said for a while. In 2025, Google's John Mueller compared llms.txt to the keywords meta tag: according to Search Engine Journal in April 2025, he said that, as far as he knew, no AI service had stated that it uses the file, and that server logs indicated they did not even request it. In June 2026, he returned to the subject. According to Search Engine Journal, the argument was that a file written by the site itself cannot be the basis for an LLM to tell one site from another, since every site describes itself favorably. Mueller conceded one narrow use: when the agent is already inside your site, a file may help it navigate.

For SEO professionals: the point about trust is the central one. A self-declared file suffers from the problem Mueller associates with the keywords meta tag: the site says what it is, and the reader has to verify it on the site anyway. Google's statement, however, applies to Google Search. Other AI products may make different choices, which is why the relevant question is "who reads the file?" and not "does Google accept it?".

Is there evidence that llms.txt increases AI citations?

We found no solid public evidence that it does. The best data point comes from an SE Ranking study, and it points the other way. There are also log records showing almost no reads of the file by the major AI crawlers, although those records are small.

Source	What it measured	Result
SE Ranking, November 2025	About 300,000 domains, comparing having llms.txt with citation frequency in LLM answers	Only 10.13% of domains had the file. No correlation with citations. The statistical model became more accurate when this variable was removed
Dejan, July to August 2026	Logs from a single site over 20 days	Crawlers from Google, Anthropic and OpenAI fetched robots.txt 1,150 times and llms.txt 4 times

In plain English: it is like a résumé. People who write their own résumé say they are excellent, and the recruiter has to check elsewhere. A file in which the site describes itself has the same limit as a criterion for choosing sources.

Both results have limits. The SE Ranking study shows correlation at the domain level and does not prove the file is useless in specific cases, such as developer documentation. The Dejan record comes from one site, and requests for the file do not say whether the content was used in answers. In short: the absence of favorable evidence is strong, but it is not proof that no agent uses the file.

The proposal itself, for its part, states that thousands of sites publish the file, that documentation platforms generate it automatically and that Chrome's Lighthouse tool checks for it in agentic browsing audits. It also says OpenAI, Anthropic and Gemini publish llms.txt for their developer documentation. These are claims by the proposal's author, and we did not independently verify whether agents read those files. They show that real use exists in documentation, which is different from showing an effect on citations.

In which scenarios is it worth creating?

It is worth it when your audience includes agents or coding assistants that need to navigate your site, and when the maintenance cost is low. For blogs and service sites that want to be cited in AI answers, the expected gain is small.

Scenario	Worth creating?	Reason
API, SDK or technical product documentation	Yes, worth testing	It is the proposal's original use, and the documentation is already structured
Site with few important pages and stable content	Optional	Low cost, unknown benefit
SEO or marketing blog that wants AI citations	Not a priority	No evidence of an effect. Time is better spent on content and sources
E-commerce with thousands of products	Rarely	A short file does not represent the catalog. Prioritize feed, schema and sitemap
Site that changes every week, with no team to maintain it	Avoid	An outdated file confuses more than it helps

If you use WordPress, llmstxt.org's list of integrations mentions plugins such as Yoast SEO and AIOSEO, which generate the file automatically. Automation lowers the upfront cost but does not replace review: check what was published.

How do you create an llms.txt in five steps?

Create a short file with a few dozen links that reflects what you would insist an agent read first. The work takes hours, not days, and the biggest cost is maintenance. Follow these steps.

1. Choose the pages. List 10 to 30 of your highest-value URLs: guides, product pages, documentation, policies and contact.
2. Write the summary. In two or three sentences, say what the site is, who it is for and what it covers. Use facts you can back up on the page itself.
3. Group into sections. Use an H2 per topic and describe each link in one line.
4. Publish at /llms.txt. Make sure the file returns status 200 and that no CDN or robots.txt rule blocks it.
5. Review every quarter. Remove broken links, update descriptions and check in the logs whether anyone actually requests it (see the next section).

Hypothetical example: the fictional company Faturaí, which sells an API for issuing electronic invoices, would publish something like this.

# Faturaí

> Brazilian API for issuing and querying electronic invoices. This page gathers the technical documentation, usage limits and support channels.

## Documentation
- [Quick start](https://exemplo.com.br/docs/inicio-rapido.md): first invoice issued in five steps
- [API reference](https://exemplo.com.br/docs/api.md): endpoints, parameters and error codes
- [Limits and pricing](https://exemplo.com.br/docs/limites.md): quotas per plan

## Support
- [Contact and status](https://exemplo.com.br/suporte.md): channels and incident history

## Optional
- [Changelog](https://exemplo.com.br/docs/changelog.md): recent changes

The example follows the specification's structure, and the Optional section is used, by convention, for secondary information that an agent can skip when it needs a smaller context.

How do you know whether anyone is reading your llms.txt?

Look for the /llms.txt path in your access logs and see which user agents request it. If after a few weeks only your own hits show up, the practical answer is that no relevant agent is using it. This test is cheap and gives you data from your own site, instead of relying on third-party studies.

grep "/llms.txt" access.log | awk -F'"' '{print $6}' | sort | uniq -c | sort -rn | head

The command assumes logs in the Nginx or Apache combined format, where the user agent is the sixth quoted field.

For SEO professionals: filter by verified IP and by period. SEO tools, curious visitors and scanners also request /llms.txt, and a one-off hit from a coding agent on your documentation does not equal use of the file in search answers. Record the date you published it and compare before and after.

To identify who the AI bots are and validate their IPs, follow the step-by-step in robots.txt and AI crawlers.

What should you do instead of llms.txt, or before it?

First invest in what Google and the studies point to as decisive: crawlable, indexed pages with content that answers well. Google's guide states that unique, useful content tends to influence presence in generative search more than any other suggestion in the guide.

1. Confirm that search bots can access the site, with a reviewed robots.txt.
2. Structure pages so each passage makes sense on its own, as shown in content structure that LLMs can cite.
3. Understand the path a question takes to an answer in how LLMs search.
4. Avoid the easy promises listed in GEO and AEO myths and mistakes.
5. See the full picture in SEO for AI search.

Contentor generates articles with a template-validated structure, internal links and WordPress publishing, which tackles item 2 without an extra file. Get to know Contentor.

Frequently asked questions

Does llms.txt improve my Google ranking?

No. Google states that Search ignores the file and that creating it neither helps nor hurts visibility or ranking. If you create it, it will be for other services that decide to use it, not for Google Search.

Do ChatGPT, Claude or Perplexity read llms.txt?

There is no official documentation from these companies saying they use the file to choose sources. Public log records show requests for the file that are very rare compared with robots.txt. Since it is unknown whether any specific agent uses it, measuring in your own logs is the best answer.

Does llms.txt replace robots.txt or the sitemap?

No. They are files with different functions. robots.txt says what bots may access, the sitemap lists pages for search engines, and llms.txt, in the proposal, offers an annotated selection for agents. The proposal says it coexists with the other standards and does not replace them.

Can creating an llms.txt hurt my site?

According to Google, it does not hurt visibility in Search. The practical risk is maintenance: an outdated file, with broken links or one that describes the site differently from what the page shows, can lead agents astray. Review it together with site updates.

Is it worth it for a small blog?

Generally, it is not a priority. With no evidence of an effect on citations, time is better spent on useful content, clear structure and reliable sources. If you have 20 minutes and a plugin that generates the file, you can create it and measure in the logs. If nobody reads it after a few weeks, forget it.

Next step

If your goal is to be cited, the most direct path runs through content: read how to structure content that LLMs can cite. To control who can crawl your site, start with robots.txt and AI crawlers.

Read also:

• GEO and AEO myths and mistakes
• SEO for AI search
• How LLMs search: RAG and grounding