What is llms.txt, and does your site need one?

llms.txt is a Markdown text file served at /llms.txt at the root of a site. It introduces the site in a few lines and links its key pages, so an AI agent can read it instead of parsing every page.

Jeremy Howard proposed it on 3 September 2024; the proposal lives at llmstxt.org and was last revised on 10 August 2026. It is an open proposal, not a standard any body enforces, and its author says the file has mostly been read by models at answer time rather than used for training.

The file opens with a heading naming the site, the only required part, then a short summary, then sections of links, each link with a one-line note on what the page holds.

This is the order llmstxt.org proposes:

• The title: the name of the site or project, on one H1 line.
• The summary: one or two sentences on what the site is and who it is for, as a blockquote.
• Sections of links: H2 headings, each over a Markdown list of links with a note on each.
• An Optional section: secondary links an agent can skip when it needs less context.

Google does not rely on it. Its guide to optimising for generative AI features says you do not need AI text files to be shown in Google Search, and that Google Search ignores them.

The same guide, last updated on 10 July 2026, adds that such files neither harm nor help your site in Google Search, and that a page must be indexed and eligible to be shown with a snippet, on a site included in Search generative AI features in Google Search Console.

As for the chat engines, we found nothing in their documentation saying they read it. OpenAI's and Perplexity's crawler pages, as we read them on 11 October 2026, mention llms.txt only as a link to their own documentation's file. So we do not count it as a route to a citation.

Tasadr (تصدَّر) keeps a real one at tasadr.ai/llms.txt, which we read on 11 October 2026. Its title is our name in Arabic and English, followed by a summary with our definition in both languages, then these sections:

We keep it as documentation, not as a lever. It is generated from the same files our pages are built from, so it cannot state a price or a number the page does not, and anyone who reads it, person or agent, finds our facts in one place.

• Facts: short lines covering the plans and their prices in US dollars, the engines, the markets and the free trial.
• Questions customers ask in Arabic: each with the answer of the page that answers it, as the page states it, and its link.
• Start here, the guides and the other pages: each page at both its addresses, Arabic and English.
• Optional: the terms, privacy and refund pages, for whoever needs them.

Your site does not need one to be cited. Start with what an engine actually reads: pages that answer your customers' questions in their first sentence and that crawlers can reach. Add the file once you have something worth summing up.

It is worth your time in two cases: a site with documentation or an API that developers read with the help of AI agents, and a large site whose fixed facts, such as prices and branches, you want gathered in one place that is easy to keep current.

Otherwise, an hour spent writing a clear answer paragraph to a question your customers ask does more for your site than an hour spent on the file.

Nothing on the page itself. An engine cites a page it read, not a summary of it, and what raises your chances is the page's own text: a direct answer, dated facts and clear links.

If you write the file, keep it true and current, because a file quoting an old price contradicts your page. Link the pages that answer your customers' questions, and describe each link in the words your customers ask with.

Before the file, make sure engines can reach your pages at all, as in can AI engines read your site?, and that your pages answer, as in the steps to lead AI answers.