In short: a good llms.txt file is one H1 with your site's name, a one-sentence blockquote a model could repeat verbatim, then a handful of H2 sections containing your most important links with one-line descriptions. Below are four complete examples you can adapt, and the mistakes that turn the file into a worse sitemap.
If you want the background first — what the file is, where it came from, and an honest read of whether AI engines use it — start with our guide to what llms.txt is. This page is the practical companion: what a good file looks like when it's finished, for businesses that aren't developer-tool companies.
The format, in 30 seconds
The llms.txt convention was proposed by Jeremy Howard of Answer.AI in September 2024 and is maintained at llmstxt.org.[1] The structure is deliberately minimal:
# Your Site Name
> One sentence explaining what this site is and who it's for.
A short paragraph of extra context if needed — what you do,
where you operate, anything an AI needs to describe you accurately.
## Main content
- [Page title](https://yoursite.co.uk/page): One line on what this page covers
- [Another page](https://yoursite.co.uk/other): And what this one covers
## Optional
- [Less critical page](https://yoursite.co.uk/extra): Secondary materialFive rules that matter:
1. It lives at the root. yoursite.co.uk/llms.txt — not in a subfolder, not on a different subdomain. The file describes the exact host it's served on: a file at www.yoursite.co.uk/llms.txt does not cover yoursite.co.uk. Publish it on the same host your canonical URLs use.
2. One H1, then a blockquote. The H1 is your site name. The blockquote is the single most important line in the file — it's the one-sentence answer to "what is this site?" that a model reads first.
3. Markdown only. No HTML, no JavaScript, no styling. If it wouldn't paste cleanly into a plain text editor, it doesn't belong.
4. Links carry descriptions. - [Pricing](url): Plans from £149/month tells a model what it will find before it fetches. A bare list of URLs is just a worse sitemap.
5. "Optional" means optional. The spec reserves an ## Optional H2 for content an AI can skip when its context is tight. It's the one section name with defined meaning — use it for genuinely secondary material.
Example 1: a small service business
Most llms.txt examples online are for developer documentation, because that's where the standard started. But the format works for any business. Here's a complete file for a fictional regional accountancy firm:
# Fenwick & Hart Accountants
> Chartered accountants in Leeds serving owner-managed businesses
> across Yorkshire — year-end accounts, VAT, payroll and tax planning.
Founded 1998. Regulated by ICAEW. We work with limited companies
and sole traders with £100k–£5m turnover.
## Services
- [Year-end accounts](https://fenwickhart.co.uk/accounts): Statutory accounts preparation and filing for limited companies
- [VAT returns](https://fenwickhart.co.uk/vat): Quarterly VAT preparation, MTD-compliant
- [Payroll](https://fenwickhart.co.uk/payroll): Managed payroll for teams of 1–50
## About
- [Our team](https://fenwickhart.co.uk/team): Partner profiles and qualifications
- [Pricing](https://fenwickhart.co.uk/pricing): Fixed monthly fees from £95/month
## Optional
- [Blog](https://fenwickhart.co.uk/blog): Tax deadlines and guidance for small businessesWhy this works: the blockquote answers the exact question a buyer asks an AI — "accountants in Leeds for small businesses". The context line carries the trust signals — founding date, regulator, client profile — that help a model describe the firm accurately and confidently. And every link description is specific enough to be quoted as written.
Example 2: a SaaS product
# BrightDesk
> Help desk software for UK e-commerce teams — shared inbox,
> live chat and order lookups in one tool, from £29/month.
## Product
- [Features](https://brightdesk.co.uk/features): Shared inbox, automations, Shopify and WooCommerce integrations
- [Pricing](https://brightdesk.co.uk/pricing): Three plans, 14-day free trial, no card required
- [Integrations](https://brightdesk.co.uk/integrations): Works with Shopify, WooCommerce, Klaviyo and Slack
## Compare
- [BrightDesk vs Zendesk](https://brightdesk.co.uk/vs-zendesk): Feature and pricing comparison
- [Alternatives guide](https://brightdesk.co.uk/alternatives): Honest comparison of UK help desk options
## Docs
- [Getting started](https://brightdesk.co.uk/docs/start): Setup in under 15 minutes
- [API reference](https://brightdesk.co.uk/docs/api): REST API documentationThe detail worth stealing is the Compare section. When someone asks ChatGPT "BrightDesk vs Zendesk", a comparison page you wrote is the best source the model can find — and the llms.txt file is you putting your hand up to say it exists. Comparison content is among the most-cited page types in AI answers, and most businesses bury it. If earning citations is the goal, that's a content problem before it's a file problem — our guide to getting cited by AI search covers what actually moves answers.
Example 3: a documentation site
This is the format's home ground — and the one place with evidence of real use. Coding-focused AI tools like Cursor, Claude Code and GitHub Copilot actively fetch llms.txt when working with documentation sites, and platforms like Mintlify generate the files automatically for the docs sites they host.[2] The pattern:
# ProductName Documentation
> Developer documentation for ProductName — API reference,
> SDKs and integration guides.
## Getting started
- [Quickstart](https://docs.example.com/quickstart): First API call in five minutes
- [Authentication](https://docs.example.com/auth): API keys and OAuth setup
## API reference
- [REST API](https://docs.example.com/api): Full endpoint reference
- [Webhooks](https://docs.example.com/webhooks): Event types and payload schemas
## Optional
- [Changelog](https://docs.example.com/changelog): Release notes
- [Community forum](https://community.example.com): Support discussionsDocs sites sometimes also publish a second file, llms-full.txt, containing the complete documentation as one Markdown document rather than links. If your product gets asked about inside coding assistants — "how do I authenticate with X's API?" — that's the audience the extra effort serves. For everyone else, the navigation file alone is enough.
Example 4: an e-commerce or content site
For a site with hundreds of product pages, the mistake is trying to list them all. Don't. List the category and buying-guide pages a model would actually cite:
# Alpine Supply Co
> UK outdoor gear retailer — hiking, camping and climbing equipment
> with free delivery over £50 and expert buying guides.
## Shop
- [Hiking boots](https://alpinesupply.co.uk/boots): Men's and women's boots from Scarpa, Salomon and Meindl
- [Tents](https://alpinesupply.co.uk/tents): Backpacking and family tents, 1–8 person
- [Delivery & returns](https://alpinesupply.co.uk/delivery): Free UK delivery over £50, 60-day returns
## Guides
- [How to choose hiking boots](https://alpinesupply.co.uk/guides/boots): Fit, terrain and budget explained
- [Tent buying guide](https://alpinesupply.co.uk/guides/tents): Weight vs space trade-offsThe Guides section is doing the visibility work. Product listings get summarised; buying guides get cited.
The five mistakes we see most often
We've reviewed a lot of generated files through our free llms.txt generator. The same problems recur:
1. The dumped sitemap. Two hundred URLs, no descriptions, no hierarchy. The whole point of llms.txt is curation — a model reading your file should find your ten best pages, not your entire archive.
2. The marketing blockquote. "The UK's leading provider of innovative solutions" tells a model nothing it can repeat. Compare: "Payroll software for UK care homes, from £2 per payslip." One of these gets repeated in an answer; the other doesn't.
3. Broken scope. File on www., site canonicalised to non-www — or vice versa. The file describes the host it's served from, so check yours resolves on the same host your canonical URLs use.
4. Stale links. An llms.txt full of dead URLs is worse than none, because you've explicitly directed a crawler to broken pages. If you restructure your site, the file is part of the migration checklist.
5. HTML in the file. Copy-pasted markup from a CMS export. Markdown only — if the file doesn't read cleanly as plain text, it fails at the one thing it exists to do.
One check before any of this matters: the file only helps crawlers that can reach your site at all. If your robots.txt is blocking GPTBot or ClaudeBot, llms.txt is a menu pinned to a locked door. Our free AI crawler access checker tells you in seconds.
Does any of this actually work? An honest note
Worth being straight, because plenty of vendors aren't. As we covered in detail in the main llms.txt guide, there's no measurable evidence that publishing the file increases how often AI engines cite you: large-scale studies have found no correlation, and Ahrefs' analysis found 97% of published files received zero requests.[3] No major AI platform has confirmed using it as a citation signal.
One thing has genuinely changed since we published that guide: Google's Lighthouse now includes an llms.txt check in its agentic browsing audits, flagging pages when the file fails to load.[4] That's not a ranking signal — but it is the clearest indication yet that the convention is being folded into how AI agents are expected to navigate sites, and it's why doing the file well is worth fifteen minutes.
Our position is unchanged: a well-made llms.txt is cheap insurance and good hygiene — a machine-readable version of your positioning that costs minutes and has no downside. It is not a substitute for the things with proven weight in generative engine optimisation: being reachable by AI crawlers, publishing content that answers real buyer questions, and being present in the third-party sources engines already trust. Make the file properly, then move on to the work that moves answers.
Make yours in two minutes
Free AI llms.txt Generator
Paste your URL. Get an AI-drafted llms.txt following the structure above — edit, copy, download. No signup to generate.
Generate yours free →Then the install guide covers where to put the file for every common platform — WordPress, Shopify, Squarespace, Wix and plain hosting.