llms.txt: what it is, who actually reads it, and whether to ship one

llms.txt is a proposed convention: a plain-markdown file at the root of your site that tells AI systems what your site is, what it offers, and where the important pages are. Think of it as a sitemap written for language models instead of crawlers. We shipped one at tryzene.com/llms.txt. The short answer, updated with what we learned in 2026: Google says its search ignores the file, and the largest study so far found almost nobody reads it - so ship one only if it costs you twenty minutes, and never instead of the work below.

What the file actually is

The format, proposed in 2024, is deliberately simple: an H1 with your name, a blockquote summary, then sections of links with short descriptions. No registration, no protocol - just a markdown file at a predictable URL. The pitch: HTML pages are noisy (nav, cookie banners, JavaScript), and a model with a limited context window benefits from a clean, curated summary.

What we actually know (October 2026)

Two things changed this year, and both point the same way.

Google said no, in writing. Its guide to optimizing for AI features (last updated July 10, 2026) says: "You don't need to create new machine readable files, AI text files, markup, or Markdown to appear in Google Search," and that Google Search ignores them - while adding that it's "completely fine" to keep an llms.txt if you want one. So for AI Overviews, AI Mode and Gemini's search grounding, the file does nothing.

Almost nobody fetches it. Ahrefs studied the 137,210 domains in its web analytics product that had traffic in May 2026 (published June 15, 2026). About 38,000 of them had a valid llms.txt, and 97% of those files received no requests at all that month. Among the few that were read, the top fetchers were GPTBot and Claude-Code - a training crawler and a coding agent, not the bots that build search answers. And AI bots never asked for an llms.txt that didn't exist: "They never go looking." (Ahrefs sells SEO and AI-visibility tools; the study covers its own customers' sites.)

Neither finding proves the file can never matter - a coding agent reading your docs summary is a real use. But there is now no evidence that llms.txt changes whether ChatGPT, Gemini or Perplexity mention your brand, and direct evidence that Google ignores it. To see whether anything fetches yours, check your own logs: here's how to spot GPTBot, ClaudeBot and the other AI crawlers.

Why we shipped one anyway

Two reasons, neither heroic. It costs almost nothing - one static file you touch when your pages change - and unlike schema spam, a clean llms.txt can't hurt you when the engines ignore it. But the real payoff was the writing itself: forcing our product into two factual sentences a model could quote verbatim exposed how vague our own homepage had been, and that clarity flowed straight back into the pages engines do read.

How to write one in 20 minutes

• H1 with the brand name, then a blockquote that states what you do in two factual sentences. No superlatives - write the sentence you'd want quoted.
• List your key pages (home, features, pricing, docs, blog) with one-line descriptions each.
• Include the facts models get wrong most: category, pricing model, who it's for.
• Keep it current. A stale llms.txt that contradicts your site is worse than none.

# Your brand

> Two factual sentences: what you are, who you're for. No superlatives.

## Key pages

- [Home](https://example.com): what Your brand is and who it's for
- [Pricing](https://example.com/pricing): plans, prices and limits

Want a standalone page you can bookmark or share with a teammate? The same generator lives at tryzene.com/tools/llms-txt-generator.

What to do instead of overinvesting here

The things with actual evidence behind them: pages that answer buying questions directly, schema markup, parseable pricing, and presence on the third-party pages engines already cite. Ship llms.txt in the gaps between those, not instead of them.

Before you optimize how engines read you, it's worth knowing whether they mention you at all. A free audit answers that on ChatGPT and Gemini (all five engines on Pro) - the measurement half of this discipline.