What is llms.txt?

Your website probably looks like noise to an AI.

When someone asks ChatGPT or Claude "what does Agent Logic do?" or "who's the AI consultant in Fond du Lac?" — the AI has to figure out the answer from somewhere. Often that somewhere is your website. The AI fetches your homepage, tries to parse the HTML, navigates around the menu chrome and the inline JavaScript and the cookie banners, and produces a best guess.

That guess is often wrong, or generic, or missing the thing you most want said about you. Not because the AI is broken — because your website was built for humans, not for AI assistants. Humans skip the cookie banner. AI tries to read it. Humans skim past the boilerplate. AI weighs it equally with the value prop. The result: an answer that's plausible-sounding but not actually yours.

"robots.txt = what AI can read. llms.txt = what AI should know."

A clean, human-written index of your business.

llms.txt is a markdown file you place at the root of your website (https://yoursite.com/llms.txt). When an AI assistant lands on your site, well-behaved ones check that file first. What it finds there is what they trust most when summarizing you.

The file is short. A typical small-business llms.txt looks like this:

The point isn't completeness. The point is signal-to-noise. Your HTML is full of nav, footer, scripts, and styling. Your llms.txt is just the substance. When an AI weighs both, the clean signal in llms.txt wins.

Which AI assistants actually use this.

llms.txt is a young standard — proposed publicly in 2024, picking up adoption through 2025-2026. As of mid-2026, here's who reads it and who doesn't:

• Reads it (or fetches it during normal crawling, which is functionally the same): ChatGPT, Claude, Perplexity, Google AI Overviews, Anthropic's Applebot-Extended, Meta-ExternalAgent, Cohere's crawler, and most "AI assistant" tools that summarize websites.
• Doesn't read it (yet): Bing's search-side crawler still leans on traditional HTML + sitemap. Some niche AI tools haven't implemented support.

The list of adopters keeps growing. Even where adoption isn't formalized, AI assistants that visit your site read your homepage anyway — and the act of writing the llms.txt forces you to articulate what your business actually is in a way that improves your homepage copy as a side effect. Worth doing even if every crawler hasn't fully implemented the spec yet.

This matters more in 2026 than it did a year ago: Google now answers many searches with AI instead of a list of links, and for local services it's even starting to book jobs for customers. If you want the why-now version, read what Google's 2026 search changes mean for local businesses.

Three ways to do this wrong.

1. Writing it like a press release.

llms.txt is for AI assistants. They don't need superlatives or marketing copy. They need plain, factual statements. "Founded by Alex Jahn, a carpenter in Fond du Lac" beats "Pioneering AI-driven solutions for the modern small business." The first one is parseable. The second one is noise.

2. Making it a sitemap.

llms.txt is not every URL on your site. It's the important ones, with context. Your job is curation. Three to five sections, the top ten pages or so, one-sentence descriptions. If it gets longer than two pages, you've drifted into sitemap territory.

3. Setting it up once and never updating it.

llms.txt is supposed to reflect your business today. If you launched a new service, dropped a product, moved locations, added a team member — update it. The file is small enough that re-editing it should take five minutes.

The honest "you might not need this."

llms.txt is mostly free upside, but there are a couple of cases where the ROI is genuinely low:

• You don't have a website. If you only operate via Instagram, Google Business Profile, or a marketplace platform, llms.txt doesn't apply — your discoverability play is structured data on those platforms, not a file at a root URL you don't control.
• You actively don't want AI traffic. If you're a small operation that already has more leads than you can handle, more visibility from AI search may not be useful. (Most operators don't have this problem. If you do, lucky you.)
• Your business is too new to describe accurately. If you don't yet know what you do or who you serve, writing the llms.txt becomes an exercise in pretending you do. Better to figure out the business first, then write the file.

If none of those apply — and they don't for most small operators — llms.txt is a 30-minute job that materially improves how AI tools talk about you for years.

If you'd rather not write it yourself.

If you liked this, the next pieces.