What is llms.txt and does my business need one?

What the file actually contains

It is plain text, structured with headings, written for a machine to read without guessing. A useful one opens with a short description of the business, then lists the concrete facts: legal name, addresses, phone, email, regulator or licensing body, languages, hours, service areas.

After that come your services as a linked list, your prices if you publish them, and any facts that are frequently asked and rarely findable — whether you offer free consultations, how quickly you respond, which areas you cover.

The point is to remove ambiguity. A web page mixes navigation, marketing language and content, and an assistant summarising it has to infer. A well-written llms.txt states things plainly, so what an assistant says about you is what you wrote rather than what it inferred.

The honest case for and against

Against: it is not a standard, no assistant has committed to reading it, and direct AI referral traffic is around one percent of visits today. Anyone telling you this will transform your enquiries is overselling.

For: it takes a few hours, it costs nothing to maintain, and the visitors who do arrive this way convert well — recent retail data has AI-referred traffic converting substantially better than non-AI traffic, having been substantially worse a year earlier. The direct referral figure also understates the effect, because a meaningful share of AI influence arrives later as branded search rather than a trackable click.

The realistic framing is insurance, not growth. It is cheap to do now and awkward to retrofit across hundreds of pages later.

What matters more than the file

Two things, both bigger levers than llms.txt itself.

First, check that you are not blocking the assistant crawlers. Some sites disallow GPTBot, ClaudeBot, PerplexityBot and others in robots.txt, often without the owner knowing, because it was a default in a plugin or a decision made by a previous developer. A file describing your business is useless if the crawler is turned away at the door.

Second, structured data. Schema markup for your organisation, services, locations, hours and frequently asked questions is read by both search engines and assistants, is a genuine published standard, and does considerably more work than llms.txt does.