# llmtxt.info, curated full reference > A current, English-language companion to https://llmtxt.info/llms.txt. It gives a compatible client or reader a compact, sourced orientation to the proposed llms.txt convention, the site’s tools, measurements and limits. Last reviewed 2026-09-07. This is a curated reference, not a concatenation of every page. ## Scope and source policy llmtxt.info documents the community proposal known as llms.txt. It is not a W3C or IETF standard, and a published file is not proof that a model, crawler or search engine uses it. The site separates four kinds of statements: - Provider documented: a behavior stated in the publisher’s own documentation. - Observed by llmtxt.info: a narrow, dated public response or measurement. - User-triggered workflow: a user or developer can supply a file or URL in a documented flow. - Unknown: no claim when support is neither documented nor reproducibly observed. The source hierarchy is a specification, provider documentation, a served public response, then reproducible llmtxt.info data. Secondary reporting is context, not a substitute for a primary source. The detailed policy is https://llmtxt.info/editorial-policy/ and the measurement method is https://llmtxt.info/methodology/. ## What llms.txt is llms.txt is a Markdown convention proposed by Jeremy Howard at Answer.AI in September 2024. A typical file has one H1 identifying the site, an optional blockquote summary, and H2 sections containing Markdown links. It gives a reader or compatible client a curated map of a site’s high-value resources. It is distinct from other web files: - robots.txt expresses crawler instructions for compliant clients. It does not enforce access. - sitemap.xml is a discovery list for search engines. - llms.txt is a curated resource map. - llms-full.txt is an optional full-text companion, intended for bulk reading rather than link selection. The current v2 proposal, modified on 2026-08-10, describes scoped files, Markdown page alternates and a describedby relation. The site exposes these relations and generated Markdown versions of its pages. Those features can be published and checked. Their presence does not prove that every client retrieves them automatically. Canonical proposal: https://llmstxt.org/ Comparison: https://llmtxt.info/comparison/ Format reference: https://llmtxt.info/llms-txt-format/ ## What llms.txt does not establish Publishing a file does not by itself establish any of the following: - Google Search ranking, AI Overviews ranking or inclusion in Google AI Mode. - Automatic discovery by ChatGPT, Claude, an agent, a crawler or a coding assistant. - Traffic, citations, safety, a commercial partnership or vendor endorsement. - Access control, privacy protection or a replacement for authentication. Google’s public guidance says no special machine-readable file, including llms.txt, is required for Google Search’s generative features. For the source and practical implications, see https://llmtxt.info/does-google-use-llms-txt/. ## Publishing a maintainable file A useful file is concise, uses absolute canonical URLs and is maintained as the underlying site changes. Start with pages that answer a real task: a product or service explanation, key documentation, pricing or policy information, API reference, support, changelog, security documentation and current status. A long indiscriminate list is less useful than a short map with accurate notes. Serve the file as UTF-8 plain text or Markdown at a public HTTP URL. Root publication at /llms.txt is common. A path-scoped file can be appropriate for a separate documentation area. Recheck redirects, content type, broken links and freshness after a release. Guide: https://llmtxt.info/how-to-create/ Best practices: https://llmtxt.info/best-practices/ Stack guides: https://llmtxt.info/llms-txt-astro/, https://llmtxt.info/llms-txt-cloudflare/, https://llmtxt.info/llms-txt-nextjs/, https://llmtxt.info/llms-txt-wordpress/ ## Tools The validator at https://llmtxt.info/validator/ parses pasted content in the browser. Its public URL mode retrieves the file through a Cloudflare function before parsing the response in the browser. Hosting logs may exist; see https://llmtxt.info/privacy/. It reports syntax diagnostics such as a missing H1, malformed list items and relative URLs. A valid syntax result does not make a quality or usage claim. The generator at https://llmtxt.info/generator/ creates a draft from a sitemap or form input. It is a starting point and requires editorial review before publication. The public verifier at https://llmtxt.info/verify/ checks one public HTTP(S) response live. Its receipt reports the resolved file URL, HTTP status, content type, structural counts, time and validator version. The corresponding badge is a live status display, not a certification of AI visibility, security or partnership. ## Observatory The Observatory records a fixed, curated panel of public hosts. Current published snapshot: fixed-panel-2026-10-05-f768192b685a, generated on 2026-10-05. - Panel size: 219 hosts. - Tested: 217; unreachable: 2. - Hosts serving a qualifying file: 125 (57.6% of tested hosts). - Median parsed file size: 14313 bytes. - Discovery observations: 330 linked pages checked, 40 Markdown alternates, 9 describedby relations, and 23 hosts with qualifying subpath files. These numbers come from the published release manifest. The panel is developer-heavy and does not estimate the whole web. Publication and discovery metadata do not prove use by an AI system. The examples gallery displays only files retrieved successfully with no structural errors detected by our parser: 43 files rechecked on 2026-09-06. The current receipts are at https://llmtxt.info/snapshots/2026-09-06/public-examples-verified/manifest.json. The original 100-URL archive of 2026-09-05 retains unsuccessful responses and diagnostics. Both are distinct from the fixed-panel measurements and must not be combined with them. The gallery is an editorial selection open to visitor contributions, not a record of visitor submissions. Its display starts with llmtxt.info, followed by well-known brands in editorial order. Visitors can suggest a public llms.txt URL and a short factual description from the Examples page, using Gmail, Outlook, their email application or the copy-address option. The interface opens a composer; it does not send automatically. Suggestions are reviewed before possible inclusion; submission does not guarantee publication or a deadline. Observatory: https://llmtxt.info/observatory/ Methodology: https://llmtxt.info/methodology/ Release manifest: https://llmtxt.info/observatory.json CSV: https://llmtxt.info/observatory.csv Panel: https://llmtxt.info/adoption-panel.json ## Governance and corrections The site uses visible update dates, source links and evidence labels. Material factual corrections are recorded publicly, while changes to data, tools and method are listed separately. Editorial policy: https://llmtxt.info/editorial-policy/ Corrections ledger: https://llmtxt.info/corrections/ Changelog: https://llmtxt.info/changelog/ Citation guidance: https://llmtxt.info/cite/ Press and research information: https://llmtxt.info/press/ Questions or correction reports: mailto:info@llmtxt.info ## Language variants The site publishes English pages and translated sections. Coverage differs by language; each locale index lists only the routes available in that language. Technical identifiers, code and observed source data retain their original form. - Français: https://llmtxt.info/fr/llms.txt - Deutsch: https://llmtxt.info/de/llms.txt - Español: https://llmtxt.info/es/llms.txt - Italiano: https://llmtxt.info/it/llms.txt - Русский: https://llmtxt.info/ru/llms.txt - العربية: https://llmtxt.info/ar/llms.txt - 简体中文: https://llmtxt.info/zh/llms.txt - 日本語: https://llmtxt.info/ja/llms.txt The root https://llmtxt.info/llms.txt is the curated multilingual entry point. For the complete inventory of generated Markdown page versions, use https://llmtxt.info/markdown-pages.json. ## Reading Google AI performance data The article https://llmtxt.info/blog/search-console-ai-report/ explains the scope of the Google Search Console generative AI report and provides a blank tracking CSV. Google announces worldwide availability from 2026-08-31 in its report help. Unavailable values shown as ~ or - are exported as zeros; preserve availability separately before interpreting changes. The teaching cases are synthetic, not site traffic observations. Report help: https://support.google.com/webmasters/answer/16984139?hl=en Blank tracking template: https://llmtxt.info/downloads/search-console-ai-tracking.csv Synthetic teaching cases: https://llmtxt.info/snapshots/2026-09-07/search-console-ai-reading.json Page impressions, file retrieval and demonstrated use by an agent are separate observations. The article does not claim an effect of llms.txt on Google visibility.