Does Your Business Website Need an llms.txt File?

llms.txt is a guide, not an AI-visibility strategy

The promise is attractive: publish one small file and make the business easier for AI systems to understand. The file can be useful, but that promise jumps over the harder commercial work. An AI assistant still needs accurate public pages, clear evidence, consistent business information and permission to access the content. A map does not improve a destination that is thin, outdated or difficult to trust.

The llms.txt project describes a Markdown file containing a short site summary and selected links to agent-friendly content. Its second version, updated in August 2026, still calls itself a proposal. That distinction matters. Adoption by publishing tools and documentation sites shows practical interest; it does not establish universal support, better rankings or more qualified enquiries.

Google is unusually explicit: Google Search does not use llms.txt, and the file neither helps nor harms visibility in Google Search or its generative AI features. Google points owners back to useful non-commodity content, crawlability, indexability, internal structure and the search experience. OpenAI's current crawler guidance likewise tells site owners to manage ChatGPT search discovery through OAI-SearchBot and robots.txt.

My verdict is therefore narrow: add llms.txt when maintaining it is cheap, the public knowledge base is already strong and a compatible agent could benefit from a curated route through it. Do not fund it before fixing missing service pages, weak evidence, conflicting business details, blocked crawlers or content that does not answer buyer questions.

This is distinct from deciding whether to block AI crawlers. Robots.txt communicates access preferences; llms.txt supplies optional context. It is also narrower than my guide to business discoverability in Google and AI search, which covers the whole evidence and discovery system.

The AI Discovery Priority Gate

Use these six checks before paying for a plugin, consultant or technical project. The order is deliberate. If the first five are weak, the sixth will create activity without solving the visibility problem.

The Gate prevents a cheap technical task from displacing higher-value work. A business with vague positioning and five thin pages does not have an llms.txt problem. A business with hundreds of detailed resources, several audiences and a clear maintenance owner may have a credible navigation use case.

There is also a governance question. The file is public. Its links can draw attention to pages you nominate as authoritative, so every included claim, offer and policy should be current. Never list confidential repositories, customer records, private documentation or content the public should not retrieve. Robots.txt is not a security mechanism either; genuine access control belongs in the application.

The Guide-to-Evidence Loop

If the Priority Gate passes, use a small operating loop. The purpose is to test whether the file helps compatible tools reach better sources—not to manufacture an impressive implementation report.

Keep the file curated. A second dump of every site URL adds little beyond the sitemap and makes maintenance harder. The proposal itself recommends brief context and organised link lists, with secondary information separated so an agent can skip it when less context is needed.

Do not confuse a successful retrieval test with business impact. An agent reaching the intended page shows that the guide can work in a controlled journey. It does not prove wider adoption, citations, search visibility, qualified traffic or revenue. Those outcomes require separate observation and a defensible attribution boundary.

When should a business add llms.txt?

If the real issue is a lack of credible material to reference, evaluate original research as a discovery investment. If you already have the evidence but do not appear in relevant answers, use the broader AI-mention and citation framework. The file belongs after those strategic questions, not before them.

A 30-day llms.txt decision test

The test has a deliberately modest success condition: the guide helps an agent reach the right public evidence with less ambiguity, and the business can keep it current at low cost. Do not set “rank in AI search” as the acceptance criterion because the file does not control that outcome.

Measure wider discovery separately. Track whether relevant questions produce accurate brand presence, whether cited URLs are authoritative, whether qualified visits reach the site and whether those visits progress to meaningful conversations. My AI-search measurement guide separates answer presence from pipeline.

Before implementation, review growth partnership services, AI Growth support, case-study evidence, evidence standards and Thomas's direct operating model. The commercial goal is not to own another file. It is to make trustworthy business evidence easier for buyers and relevant systems to find and use.

Practitioner note: I would add llms.txt to a mature resource library when the implementation and upkeep are genuinely small. I would not let it consume the week that should have been spent answering a high-value customer question or fixing a blocked, unconvincing service page.