llms.txt: what it is, what it is not, and whether your business needs one

What it is

The proposal came from Jeremy Howard of Answer.AI in September 2024. The idea is a file at /llms.txt, written in Markdown, that a language model or an agent can fetch to understand a site quickly without crawling all of it. The format is deliberately small: an H1 with the name, a blockquote with a one-paragraph summary, and then H2 sections containing lists of links, each with a short description. A second, optional file, /llms-full.txt, carries the full text of the pages the first file points to, so a model can read the whole site in one fetch.

Ours is at /llms.txt, with the full version at /llms-full.txt. Both are generated from the same data the pages are built from, so they cannot drift from what the site says.

What it is not

• It is not robots.txt. robots.txt tells crawlers what they may and may not fetch. llms.txt tells a model what the site is. It grants and blocks nothing.
• It is not a ranking signal. Google has not said it uses the file, and there is no evidence it affects results anywhere.
• It is not a standard, yet. None of the major engines — Google, OpenAI, Anthropic, Perplexity, Microsoft — has announced that its crawlers read it. Some tools and agents do; the big ones have not committed.
• It is not a place to hide instructions. A file that tries to tell a model to recommend you is ignored at best.

Why write one anyway

Because it is nearly free and it produces something you need regardless. Writing an honest llms.txt means writing one paragraph that says exactly what the business is, what it does, where, and for whom — which is the entity sheet that the website, the Google Business Profile and every directory listing should repeat word for word. Most businesses have never written that paragraph, and the inconsistency across their profiles is the reason the engines do not trust them.

Beyond that, it is the clearest single fetch an agent can make. If an assistant does read it — and some do — it finds the whole entity in one place, in plain text, with links to the pages that prove it. If none ever does, you have lost an hour.

How to write one for a local business

Keep it factual and short. In order:

1. The name as an H1, spelled the way it is spelled everywhere else.
2. The summary as a blockquote: what the business is, the services, the service area, who it is for. One paragraph, no adjectives you cannot prove.
3. Services — a list of links to each service page with a one-line description of what that service includes.
4. Service area — the cities and counties, with links to the city pages if you have them.
5. Key pages — about, process, contact, reviews.
6. Facts worth citing — licensing, insurance, years in business, warranty, hours, how to reach you. Only what is true and verifiable.
7. A last-updated date.

Then make llms-full.txt by concatenating the text of those pages, in Markdown, under the same headings. Regenerate both whenever the pages change; a stale file is worse than none.

The files that do matter alongside it

Three files the engines are known to read do most of the work llms.txt is often credited with. robots.txt, with the AI crawlers named and allowed — GPTBot, OAI-SearchBot, ClaudeBot, PerplexityBot, Google-Extended and the rest — and the private paths disallowed in every group. sitemap.xml, current, with every public page. And the structured data on each page: Organization or LocalBusiness, Service, FAQPage, BreadcrumbList, with a Person node for who is accountable. Get those right first; add llms.txt in the same afternoon.

The free scan checks all four — robots.txt access for the AI agents, the sitemap, the schema and the presence of an llms.txt file.