llms.txt: do you actually need it?

What this article is about

Some people swear llms.txt is the future of AI search. Others say it is dead on arrival. Both are wrong.

This article explains what llms.txt does, who actually reads it, and whether you should bother creating one. The answer depends on your goals. By the end you will know exactly what to do for your site.

What is llms.txt?

llms.txt is a plain text file you put at the root of your website. Same idea as robots.txt or sitemap.xml. It tells AI assistants what your site is about, where the important pages are, and what context to use when answering questions about your business.

A minimal llms.txt looks like this.

# AcmeWidget

> AcmeWidget is a CRM for small law firms. We have 12,000 users in 47 countries.

## Documentation
- [Getting started](https://acmewidget.com/docs/quickstart): Set up your firm in 10 minutes.
- [API reference](https://acmewidget.com/docs/api): REST endpoints and webhook payloads.

## Resources
- [Pricing](https://acmewidget.com/pricing): Plans from $19 per month.
- [Customer stories](https://acmewidget.com/customers): 12 case studies.

That is it. A title. A short description. Then sections of links with descriptions.

The file is meant to give AI agents a map. Instead of guessing what your site is about, the agent reads llms.txt and knows immediately. The proposal originated with Jeremy Howard at Answer.AI in 2024.

Who actually reads llms.txt right now?

This is where the debate gets interesting.

The honest answer is that llms.txt is read by almost no major AI engine in 2026. The standard's adoption has stalled. Google explicitly says you do not need it. The optimistic case is that Anthropic and a few others might use it more in the future. The pessimistic case is that it goes the way of Microformats.

So why have I spent three paragraphs on it. Because the file still has value, just not the value most articles claim.

When should you bother with llms.txt?

The file is worth creating in three cases.

First, if you are a documentation-heavy site. AI agents that integrate with code editors (Cursor, Windsurf, Continue) actively look for llms.txt and llms-full.txt. They use it to give better autocomplete and answer questions about your API. Anthropic, Stripe, Vercel, and Mintlify all have one for this reason.

Second, if you want to look polished to a small subset of users. Some AI-savvy buyers check for llms.txt the same way they check for security.txt. It signals that your team understands modern web standards. The cost is one hour. The benefit is small but real.

Third, if you have a complex catalog. E-commerce stores with many product categories, software companies with many products, agencies with many service lines. llms.txt forces you to write a clean catalog of your own site. The exercise alone often improves your information architecture.

If none of these apply to you, skip it. There are higher-impact things to do.

What should be in your llms.txt if you make one?

The Mintlify team analyzed real llms.txt files from leading tech companies. The pattern is consistent.

1. Title. Just your brand name. One H1.
2. Description. One sentence. Definition format. "X is Y."
3. Two to four sections. Each section is one H2.
4. Each section contains 5 to 15 links.
5. Each link has a description after the URL. Two short sentences max.
6. No marketing language. No "world-class". No "leading provider".

Total length. Around 50 to 200 lines. The Anthropic one is 8,364 tokens. Vercel's is shorter. There is no fixed length.

A common pattern is to also publish llms-full.txt. This is the complete content of your most important pages, concatenated into one file. AI agents that want full context can fetch it. The Anthropic full file is 481,349 tokens. That is the entire Claude documentation in one place.

Should small businesses make a llms-full.txt?

Probably not. The full version makes sense for documentation. It does not make sense for a five-page marketing site.

If you are a small business, here is the practical guide. Make a 10-line llms.txt with your brand definition, one paragraph of context, and links to your three most important pages. That takes 20 minutes. Skip llms-full.txt entirely. Spend the saved time on actually improving the content of those three pages.

If you are a SaaS company with documentation, make both. Auto-generate them from your docs platform if you can. Mintlify, Docusaurus, and Nextra all support this. The maintenance cost should be near zero once set up.

Where llms.txt actually fits in 2026 GEO strategy

I rank GEO investments by impact per hour. Here is my honest order, based on what I see in our audit data.

llms.txt is in the bottom third. It helps a little. It helps no one if your homepage's first 100 words are a corporate brochure.

Summary

llms.txt is real. It exists. Some AI engines read parts of it. Most do not. It is worth making if you have documentation, complex catalogs, or want to signal modernity. For most small businesses, it is the seventh or eighth thing on the priority list, not the first.

The first thing on the priority list is to find out where your site actually stands. The free audit at AIFreeAudit checks 35 GEO signals including llms.txt, structured data, content patterns, and citation readiness. It takes 30 seconds.

If you want a hand-built llms.txt for your site, email me at paul at aifreeaudit dot com. I have a free template and can usually turn one around in a day for any small business with a clear product.