What Is llms.txt? The Honest Answer From Someone Who Ships One

TL;DR: A Sitemap for Robots That Read Prose

Half the AEO industry is currently selling llms.txt as a magic file that makes ChatGPT cite you. Let's fix that in one post.

I'm in an unusual position to write this one: I've had an llms.txt live on this site for months, it's auto-generated by the same Rails app that publishes these posts, and I run a weekly citation monitor that tells me exactly which of my pages AI engines cite and for what. Which means I can tell you what the file is, how to ship one in 15 minutes — and, more honestly than the people selling AEO packages, what it does and doesn't do.

What llms.txt Actually Is

llms.txt is a convention proposed by Jeremy Howard of Answer.AI in September 2024. The idea: HTML pages are a noisy format for language models — navigation, scripts, cookie banners, three newsletter popups. So you serve a clean markdown file at yoursite.com/llms.txt that says, in plain prose: here's who this site belongs to, here's what it covers, here are the pages that matter, one line each.

The format is deliberately simple. Mine looks like this (live at /llms.txt):

# Justin McKelvey
> Software consultant helping founders build and ship products.

## Blog Posts
- [Post Title](/blog/slug): Short description of what the post answers

That's the whole trick: an H1, a blockquote of context, and curated link lists with descriptions. A companion file, llms-full.txt, goes further — the full content of the site flattened into one markdown document, so a model can ingest everything without crawling.

llms.txt vs sitemap.xml vs robots.txt

The meaningful difference is curation. A sitemap lists everything with equal weight; llms.txt is you editorializing — "these are the pages worth your context window." That's also why a stale llms.txt is worse than none: you're actively pointing readers at the wrong doors. Mine regenerates automatically every time a post publishes, which is the only maintenance policy that survives contact with reality.

The Honest Part: Does Anything Actually Read It?

Here's where I break ranks with the AEO sales decks. As of 2026, no major AI provider has publicly confirmed that it uses llms.txt for retrieval, ranking, or training. Google's John Mueller has compared it to the keywords meta tag — the classic example of a signal sites volunteer and engines learned to ignore. AI crawlers do fetch the file (server logs show it), but a fetch is not a citation strategy.

And my own data agrees. My site gets cited by AI engines on dozens of tracked queries — that's the whole citation dataset — and every page that earns citations earned them the boring way: an answer in the first 100 words, question-phrased headings, specific numbers, FAQ schema, topical depth. Not one citation pattern changed when the llms.txt file went live. The file didn't hurt. It just wasn't the lever.

So Why Ship One at All?

Because the trade is asymmetric: ~15 minutes of cost, zero downside, and a real option on the future. The standard exists because model providers have an actual problem (HTML is noisy), which means there's a plausible world where adoption arrives — and in that world, sites with clean, current llms.txt files are grandfathered in. Plus, if an agent shopping on a user's behalf does read your file today, you'd rather it found your curated pitch than your cookie banner.

Today's micro-action: ship the minimum viable version. One markdown file — H1, one-line blockquote about what you do, an H2 with your 10 most important pages linked and described. Put it at your web root. If your site is code, make it generate from your content so it never goes stale. Then stop thinking about it, because the real work is elsewhere.

Where llms.txt Fits in an Actual AI Visibility Strategy

File this under "table stakes, not strategy" — alongside schema markup and clean HTML. The layers that demonstrably move citations, in the order I'd spend hours on them: answer-shaped content on questions your buyers actually ask (the playbook: how to get cited by ChatGPT), engine-specific trust signals (ChatGPT SEO is a different game than Perplexity), and measurement so you know what's working (track your brand mentions in AI). The full discipline is answer engine optimization; llms.txt is one small tile in it.

Want to know whether AI engines mention your business today — before you spend a dollar on AEO anything? I run a free AI visibility report: I check the queries your buyers ask, tell you who's getting cited in your space, and what would actually move it. No magic files involved.

Related guides: answer engine optimization, LLM SEO, AI search optimization, how to get cited by ChatGPT, who do AI engines cite.