Write A Correct llms.txt

It formats. You type the site name, the summary sentence, and the links you want an agent to find, and it writes them out in the shape the llms.txt proposal asks for: an H1, a blockquote, optional notes, H2 sections with one link per line, and an Optional section at the end. It checks that every URL is absolute and starts with https, leaves out the ones that are not, and warns you if the summary is empty. Nothing leaves your browser. Copy the result and save it as /llms.txt at your domain root.

I wrote in Does llms.txt work? that a generator adds nothing a text editor does not. That is still true. The reason this one exists is that people kept asking for it, and the version they were pasting from other tools had the sections in the wrong order, relative URLs, and summaries that said nothing. If you are going to publish the file, publish a correct one.

The blockquote is the only part of the file with a payoff. Write it the way a buyer would describe you to a colleague: what it is, who it is for, and the category name they would type into a search box. "Product analytics for small SaaS teams, priced per seat" is a summary. "The modern platform for data-driven growth" is not, and an agent reading it learns nothing it can repeat. Spend your ten minutes on that sentence and let the tool handle the rest.

It will not get you cited. When I measured what moves AI citations across the four answer engines, the file did not appear in the list: the pages the engines already cite, written so the extraction pulls your answer; the first 200 words of each commercial page as a direct answer; the same positioning in JSON-LD; retrieval crawlers allowed through robots.txt and the CDN; and third-party mentions on the domains the engines trust for your category. Google says it does not read llms.txt. The server-log studies say almost nobody else does either. The evidence is in the article; the short version is that the file is free, harmless, and not a lever.

If the question behind the file is "why does ChatGPT name my competitor and not me", start with the free visibility check on unbranded prompts. If the answer is a number you do not like, the Growth Pilot takes a baseline in week one, fixes the pages that get extracted as an experiment card, and reads the same prompts again on day 28. The llms.txt file gets written on day one and is never mentioned again.

Keep section names short and plain: Docs, Pricing, Compare, Blog. One line per link, and make the description say what is on the page, not why it is great. Put the pages an agent can skip under Optional; the spec uses that name by convention. Absolute https URLs only, because the file is read out of context and a relative path resolves to nothing. If you maintain a long version, add one line to the notes paragraph pointing at /llms-full.txt, the way this site does.

What is llms.txt? A markdown file at the root of a website, proposed by Jeremy Howard in September 2024. The format is small: an H1 with the site name, a blockquote with a one-line summary, optional free text, then H2 sections that list links as '- [title](url): description'. A section named Optional holds links an agent can skip when it needs a shorter context. The spec lives at llmstxt.org.

Does llms.txt help AI search? On the evidence, no. Server-log studies across hundreds of thousands of domains found that almost no llms.txt files are requested, and the requests that do arrive are mostly bots rather than the retrieval crawlers behind ChatGPT or Perplexity. Google says it does not use the file. It is not a ranking factor and it does not move citations. It costs nothing and helps coding agents that need your docs, so keep it, but do not report it as the reason a visibility number changed.

Where do I put llms.txt? At the root of the domain, served as plain text: https://yourdomain.com/llms.txt. It should return 200 with a text/plain or text/markdown content type. If you also publish the long version, put it at /llms-full.txt and mention it in the short file.

What is the difference between llms.txt and llms-full.txt? llms.txt is the curated index: the summary and a short list of links with one line each. llms-full.txt is the whole content of those pages in one file, for an agent that wants everything without following links. Most sites only need the short one. This generator produces the short file.

Should the file list every page? No. List the pages a buyer or an agent would need to answer the questions people ask about you: what it is, who it is for, pricing, docs, comparisons. A file with three hundred links is a sitemap with extra steps. The summary sentence is the part that carries the meaning; the links are there to point at the evidence.