What the llms.txt Builder does
The llms.txt Builder drafts an llms.txt file - a short Markdown map of your most useful pages for language-model tools - from a name, a summary and the sections and links you choose, formatted exactly as the llmstxt.org proposal describes. It previews the file as you type, checks it against the proposal's structure line by line, and can pick pages from a public sitemap.
Be clear about what the file is: a convention proposed in 2024 and used by some documentation sites and developer tools, not a web standard. Google says no special AI files are needed for its AI features, and no major search engine documents reading llms.txt. Build one because it is useful to tools that look for it, not in the expectation of rankings or citations.
How to use it
- Enter the site or project name (it becomes the H1, the only required part) and a one-paragraph summary (the blockquote).
- Optionally add notes - short paragraphs or bullet points, but no headings, which the proposal does not allow before the first section.
- List sections: a line starting ## names each one, followed by one link per line as URL | name | notes. Put secondary links in a section named Optional.
- To start from your sitemap, open Pick pages from a public sitemap, load it, tick the pages worth listing and add them. They arrive grouped by their first path segment.
- Fix anything the format checks flag, copy or download the file, publish it at the root of your site as /llms.txt, then verify the live file with the llms.txt Checker.
Reading the results
The format checks compare the generated file with the proposal: one H1 first; an optional blockquote summary; any Markdown except headings before the first H2; H2 sections whose lines are all "- [name](url)" with optional ": notes"; absolute URLs; no duplicates; and Optional last. A clean result means the file is well-formed under the proposal, nothing more.
Input lines that are neither a ## heading nor a line starting with a full URL are skipped and listed, so a pasted list with stray text does not silently lose links.
Brackets in link names are escaped and spaces or parentheses in URLs are percent-encoded, so every link parses back to exactly what you typed.
Worked example: documentation for a shipping API
The documentation example describes a fictional parcel API. The name becomes "# Parcelhub API", the summary a blockquote, and two notes (the v1 retirement date and the sandbox key prefix) stay as a bullet list before the sections, which the proposal allows.
Three sections hold seven links: Docs (quickstart, authentication, errors), API reference (labels, tracking) and Optional (changelog, v1 migration guide). The links point at .md versions of the pages, as the proposal suggests for pages that have them. The checks report one clean result: an H1, a summary, three sections and seven links, and the file is about 1 KB.
What goes in, and what to leave out
The point of llms.txt is brevity: a reader with a limited context window gets the handful of pages that explain your product, not your whole sitemap. List the pages that answer the questions people actually ask - what it is, how to start, the reference, the policies - and put nice-to-have material under Optional.
The proposal also suggests offering clean Markdown versions of pages at the same URL with .md appended, and some sites publish an expanded llms-full.txt. Both are optional conventions; this builder writes the core file.
Limitations: what the result does not prove
- It builds and checks the file's structure. It does not fetch or test the links you list - the llms.txt Checker does that for the published file.
- The sitemap reader handles standard urlset and sitemapindex files up to 2 MB. Compressed .xml.gz sitemaps are not read.
- Following the proposal does not mean any particular AI product reads the file; adoption varies and is not documented by the major search engines.
- Descriptions are yours to write. The tool never invents summaries or notes for your pages.
Privacy: where your data goes
The analysis runs in your browser. If you choose to load a public URL instead of pasting, a2z.tools fetches that address read-only and returns the text to your tab; private and internal addresses are refused and nothing is stored. Session recording and tag-manager scripts are switched off on this page.
Standards and sources
- llms.txt proposal (llmstxt.org) - checked 19 Sep 2026
- Google Search Central - AI features and your website
- Sitemaps XML format (sitemaps.org)
Frequently asked questions
What is llms.txt?
A Markdown file at /llms.txt, proposed at llmstxt.org, that gives language-model tools a concise, curated map of a site: its name, a summary, some context and lists of links to the most useful pages. It is to LLM tools roughly what a hand-written table of contents is to a reader.
Is llms.txt an official web standard?
No. It is a published proposal, not an IETF or W3C standard, and major search engines have not documented using it. Google's guidance on AI features says no special AI text files are needed. Treat it as an optional convention.
Does llms.txt replace robots.txt or a sitemap?
No. robots.txt (RFC 9309) tells crawlers what they may fetch; a sitemap lists URLs for indexing. llms.txt is a short, human-curated guide to your key content and has no effect on crawling permissions.
What belongs in the Optional section?
Links a tool can skip when it needs a shorter context: changelogs, archives, background reading, detailed references that most questions will not need. The proposal defines Optional for exactly this, and the builder always writes it last.
Why can I not use headings in the notes?
The proposal allows any Markdown between the summary and the first H2 except headings, because H2 headings are what delimit the link sections. The builder escapes any heading you type there so the structure stays valid.
Should my links point to .md files?
If you publish Markdown versions of pages, yes: the proposal suggests them because they are cleaner to read than HTML. If you do not, link the normal pages - the format accepts any URL.
Last reviewed by the A2Z.Tools team against the sources listed above.
