--- language: en title: "llms.txt: What It Is, How to Implement It, and AI Visibility" description: "Learn about the current llms.txt proposal, how it differs from robots.txt and sitemaps, its uses in documentation, and the limits of evidence on AI citations." canonical: "https://tropk.ai/en/blog/llms-txt-what-it-is-how-to-implement-it-and-what-it-changes-about-ai-visibility" last_modified: "2026-10-12T16:05:28.796Z" --- # llms.txt: What It Is, How to Implement It, and AI Visibility Learn about the current llms.txt proposal, how it differs from robots.txt and sitemaps, its uses in documentation, and the limits of evidence on AI citations. [Luiz Gardelin]() Published on 2026-07-01T02:34:14.000Z Categorias: GEO/AEO [llms.txt](/en/resources/glossary/llms-txt) is a Markdown file that introduces a website and organizes links to useful content for AI agents. It can make documentation easier to consult, but it does not control access to the site or guarantee citations in [ChatGPT](/en/resources/glossary/chatgpt), [Perplexity](/en/resources/glossary/perplexity), or Google. The useful question is what problem the file solves. For a software library, it can help an agent find the right reference. For a blog seeking traffic, the file does not replace trustworthy content, [indexing](/en/resources/glossary/indexing), [internal links](/en/resources/glossary/internal-links), or measuring results. This guide was updated with sources consulted on **October 6, 2026**. The proposal has evolved, and some claims in the original article—including the absence of llms.txt from OpenAI documentation—no longer describe the current state. ## Key points - llms.txt organizes context and links in Markdown for agents that consult this format. - The current proposal allows files in the root directory or subdirectories, as well as Markdown versions of pages. - OpenAI Developers publishes llms.txt; this does not prove it is a citation factor in ChatGPT Search. - Google Search does not use llms.txt as a special file for ranking or generative features. - Implement it with verified links and regular maintenance; assess its usefulness to agents separately from organic visibility. ## What llms.txt is and why it was created Jeremy Howard introduced the proposal in September 2024 to make it easier for a language model to consult pages designed for people. An [HTML](/en/resources/glossary/html-the-web-s-markup-language) document can contain navigation, notices, scripts, and repeated elements. A short index accompanied by clean Markdown versions offers a more direct path to information. The [proposal maintained at llmstxt.org](https://llmstxt.org/) describes a file that can be read by people and programs. The file introduces the project, provides context, and points to more detailed documents. The goal is to serve as an entry point for consultation; the full content can remain on the site's pages. The “[sitemap](/en/resources/glossary/sitemap) for AI” metaphor is a helpful introduction, but it obscures a difference: an XML sitemap is a list of URLs for discovery by search engines; llms.txt is a curated collection with descriptions and context. There is no reason to replace one with the other. ## What changed in the current proposal Version 2 of the proposal allows for a file at the root or at a specific path. This means `/llms.txt` can introduce the entire site, while `/docs/llms.txt` presents only the documentation. When more than one file covers a page, the proposal recommends using the most specific one. The proposal also recommends Markdown versions of pages, with URLs such as `pagina.html.md` or `pagina.md`. For directory URLs, it describes alternatives such as `index.html.md` and `index.md`. These paths are publishing conventions: they need to exist on the server; adding `.md` to a URL does not automatically create a readable version. Another recommendation is to announce related documents using link relations: `alternate`, paired with `type="text/markdown"`, for the Markdown version; and `describedby` for the llms.txt that introduces that section. This can be declared in the HTML or in the HTTP header `Link`. These recommendations expand discovery by compatible agents. They do not mean that all search engines have started using llms.txt or that the file has become a ranking factor. ## Who publishes llms.txt today The [OpenAI developer documentation](https://developers.openai.com/llms.txt) publishes a file with indexes by product. This corrects an outdated claim in the previous article: the file's existence can be observed, although that does not demonstrate that ChatGPT Search uses third-party llms.txt files as a [citation](/en/resources/glossary/ai-citation) criterion. The [Chrome Lighthouse has an llms.txt check for agent navigation](https://developer.chrome.com/docs/lighthouse/agentic-browsing/llms-txt). This audit checks a technical condition of the page; its presence is not equivalent to [SEO](/en/resources/glossary/seo) approval or a test showing an increase in citations. The distinction applies to any documentation platform: generating a file and providing Markdown pages demonstrates a publishing capability. Claiming that a specific AI consults it requires evidence of that AI's behavior. ## Does llms.txt make a site appear more often in AI answers? There is no basis for treating llms.txt as a guarantee of visibility. [SE Ranking analyzed approximately 300,000 domains](https://seranking.com/blog/llms-txt/) and found no positive association between the presence of the file and citation frequency in the set studied. Adoption in that set was 10.13%. This result is observational and limited to the study's sample, systems, and period. It does not prove that the file will never be useful to any agent. It also does not justify promising that creating the file will increase citations. The [Google's official guidance on AI optimization](https://developers.google.com/search/docs/fundamentals/ai-optimization-guide#mythbusting) is explicit: Google Search, including its generative features, does not use these files as a special mechanism for discovering or ranking content. The file can be maintained for other systems that use it. For OpenAI and Perplexity, official access guidance continues to focus on crawlers, [robots.txt](/en/resources/glossary/robots-txt), and the infrastructure that allows requests. Publishing a Markdown index does not remove a server block or guarantee selection as a source. ## llms.txt, robots.txt, and sitemap.xml: different functions | File | Main function | What it does not do | | --- | --- | --- | | `robots.txt` | Declare crawling preferences for compatible bots | Does not protect private content or guarantee removal from indexes | | `sitemap.xml` | Provide URLs and, when accurate, update dates | Does not guarantee indexing or a position in results | | `llms.txt` | Present context and curated links for compatible agents | Does not enforce permissions, control training, or guarantee citations | An llms.txt file should not point to private documents or bypass access policies. Confidential content should be protected with authentication and authorization. To choose which crawlers can access public pages, use the [guide to robots.txt for AI](/en/blog/robots-txt-for-ai-which-bots-to-allow-and-how-to-configure-it). ## When implementation makes sense ### Product and developer documentation The most straightforward case is documentation with references, tutorials, versions, and code examples. An agent can consult a small index and follow only the links relevant to its task, rather than loading the entire site. Describe which version is documented, distinguish reference pages from tutorials, and briefly identify essential materials. If the agent needs to compare two versions of an API, an index without that distinction could lead to the wrong answer. ### Corporate websites and help centers A help center can organize questions about setup, limits, and policies. A corporate website can gather official information about the organization. The expected benefit is to make it easier to consult sources published by the company itself, provided the consumer consults the file. Choose current, readable documents. An index with dozens of duplicate links, empty pages, or promotional material without useful information shifts the navigation problem into the file. ### Blogs and visibility strategies For a blog, llms.txt can be an editorial maintenance aid. The decision depends on the cost of creating, reviewing, and maintaining the index, not on a promise of organic growth. If the site has inaccessible pages, unidentified authors, unsourced content, or broken links, address those gaps first. These problems affect readers and search systems regardless of whether llms.txt is adopted. ## How to create an llms.txt The proposal requires an H1 with the project or site's name. A blockquote summary and sections with links are recommended components, not a full list of required fields. Lists use Markdown link format and can add a description after a colon. The example below is illustrative. Replace the links with documents that exist and check that the descriptions match the published content. ```markdown # Exemplo de Empresa > Informações oficiais e documentação pública sobre a empresa e seus produtos. Este índice reúne conteúdo atualizado. Para dúvidas sobre uma versão de produto, consulte a versão indicada na página de documentação. ## Documentação - [Visão geral](https://example.com/docs/overview.md): Conceitos e limites do produto. - [Primeiros passos](https://example.com/docs/quickstart.md): Procedimento de configuração. - [Referência](https://example.com/docs/reference.md): Parâmetros e exemplos de uso. ## Informações institucionais - [Sobre a empresa](https://example.com/sobre): Identidade e canais oficiais. - [Políticas](https://example.com/politicas): Regras públicas de uso e atendimento. ## Optional - [Histórico de versões](https://example.com/docs/changelog.md): Mudanças anteriores. ``` The `Optional` section is a convention for secondary material that can be omitted when the consumer needs less context. It is not a permission directive. ### How to announce Markdown versions in HTML For a page that actually offers a Markdown version and an index at `/docs/llms.txt`, the links can be declared as follows: ```html ``` The same relationship can be provided in an HTTP header. This is a header line, not an HTML snippet: ```text Link: ; rel="alternate"; type="text/markdown", ; rel="describedby" ``` Before publishing, confirm that the Markdown version URL returns the correct document. Example descriptions and paths should not remain in the final file. ## Is llms-full.txt the same thing? Some platforms offer `llms-full.txt`, which combines the content of many documents into a single file. This differs from the short index in llms.txt. Its convenience depends on the consumer: loading everything can be useful for a broad query, but it can also increase the amount of text, include irrelevant material, and exceed the available context. The current proposal favors an index with links that can be followed as needed. You do not need to produce a huge, complete file to meet the basic format. If the platform generates both, check that the versions stay synchronized and do not expose pages that should require a login. ## How to validate without confusing access with results 1. Open the llms.txt URL and confirm that it returns a public response containing text, without redirecting to a login. 2. Review the H1, descriptions, and every link in the index. 3. Confirm that Markdown versions match the human-readable pages and preserve code, tables, sources, and dates. 4. Ask an agent questions, providing the file as an entry point, and record the paths consulted and any errors encountered. 5. Compare navigation effort and answer accuracy before and after, using the same tasks. These tests assess usefulness for a consumer that received or discovered the file. A test in which you give llms.txt to the agent does not demonstrate that ChatGPT Search will start finding it on its own. To study visibility, keep a separate baseline: the same questions, platforms, language, and period. Record citations, mentions, and pages used. Simultaneous changes to content, links, and [crawling](/en/resources/glossary/crawling) make it difficult to attribute a change exclusively to llms.txt. ## Mistakes to avoid - Treating the file as a training block: that is not what llms.txt does. - Saying that all AIs follow the file: support and behavior depend on the system. - Publishing links to nonexistent pages or old versions. - Copying the entire sitemap without curation or descriptions. - Keeping Markdown versions with information that differs from the human-readable page. - Interpreting an access in the logs as proof of reading, citation, or ranking gain. - Leaving the index outdated after changing URLs or removing documents. ## The place of llms.txt in the strategy llms.txt can be a useful consultation interface for documentation and compatible agents. It is useful when it helps with a specific task: finding a procedure, distinguishing versions, or retrieving an official source. The [GEO](/en/blog/what-is-geo-generative-engine-optimization-complete-2026-guide) strategy should continue to be based on verifiable content, technical access, demonstrated authority, and platform monitoring. Implement the file when there is a clear use case and a maintenance routine; assess citations using your own data, without turning a publishing convention into a performance promise. ## Sources for further reading - [llms.txt v2 proposal]() - [OpenAI Developers llms.txt]() - [Google: AI optimization guide]() - [Lighthouse: llms.txt audit]() - [SE Ranking: study of approximately 300,000 domains]() ## Frequently asked questions ### What is llms.txt, and what is its purpose on a website? llms.txt is a Markdown file that introduces a website and organizes links to documents useful to AI agents. Its purpose is to make content easier to consult for users of this format. The file does not block bots, control training, or guarantee citations in AI responses. ### Does creating llms.txt increase a website’s citations in ChatGPT or Google? Creating llms.txt does not guarantee an increase in citations. A SE Ranking study of approximately 300,000 domains found no positive association in the analyzed sample. Google states that Search and its generative features do not use llms.txt as a special ranking file; OpenAI documentation does not present it as a citation factor in ChatGPT Search. ### What is the difference between llms.txt, robots.txt, and sitemap.xml? llms.txt provides context and curated links to compatible agents. robots.txt states crawling preferences for bots that follow the protocol. sitemap.xml lists URLs that a search engine can discover. The three files have different functions, and none guarantees indexing, ranking, or citations. ### Where should an llms.txt file be published, and which sections are required? According to the current proposal, llms.txt can be published in the website root, as /llms.txt, or in a subdirectory, such as /docs/llms.txt. The file covers pages under that path. An H1 with the project or website name is the only required section; a summary, context, and link lists are recommended elements. ### How can you test whether llms.txt is useful to AI agents? To test whether llms.txt is useful, confirm that the file and its links are accessible, provide the index as an entry point to an agent, and compare the accuracy of its answers on the same tasks before and after. Record the sources consulted and any errors. This test evaluates an agent that used the file; it does not demonstrate a spontaneous increase in citations in search systems. ### What distinguishes llms-full.txt from llms.txt? llms.txt typically serves as a short index with context and links. llms-full.txt, offered by some platforms, combines the text of many documents into a larger file. The full version can be useful for broad queries, but requires care with context volume, updates, and private content; it is not required to publish the llms.txt index. ## About the author [Luiz Gardelin]() Luiz Gardelin is the founder of Tropk.ai and works in GEO/AEO, product vision, strategy, and growth, helping brands understand and improve their presence in AI responses. --- [All pages in Markdown]()