Emerging Digital PartnersGet in touch

How-to guide · technical

How to write an llms.txt file (and what it does and does not do)

Google says you can ignore llms.txt. Ahrefs found 97 percent of the files it studied were never requested. No major AI lab has committed to reading it. It is still a 30-minute job with no downside, as long as you know exactly what you are and are not getting. Here is how to do it properly.

By Richard Daniel3 min read

We publish an llms.txt on this site. We also want to be straight about why. As of mid-2026: Google's own guidance says you do not need one for AI Overviews or AI Mode; John Mueller compared it to the keywords meta tag; Ahrefs found that 97 percent of the files across 137,000 domains were never requested; and OpenAI, Anthropic, Google and Perplexity have not committed to using it in production. A well-written llms.txt will not get you cited.

What it does do is give AI agents and coding assistants a clean map of your site, and force you to write a canonical description of what you do. It takes half an hour. Do it after you have fixed crawler access, rendering and your About page, not instead of them.

Step 1: Decide whether it is worth your half hour

Yes if: you have documentation, a product with an API, a site you expect AI agents to act on (booking, buying), or you simply want a maintained summary for machines. Also yes if a client or platform you rely on asks for one. Otherwise it is optional, and nothing on this site depends on it.

Step 2: Learn the format

The specification at llmstxt.org is short. A file served at /llms.txt as plain text or markdown containing: one H1 with the site or project name; a blockquote with a one-paragraph summary; optional paragraphs of key facts; then H2 sections each containing a list of links in the form - [Title](url): one-line description. An optional section titled "Optional" lists lower-priority links an agent can skip.

Step 3: Write it from your canonical facts

Use the same 25-word description as your About page and Organization schema. Then the facts an agent would need: what you offer, who for, where you are, how to contact you, key policies. Then links to the ten to twenty pages that matter (services, pricing, about, contact, guides), each with a description that states what the page answers. Here is a compact example:

# Example Digital

> Example Digital is a web design and SEO agency in Eastbourne, East Sussex, working with small and mid-sized UK businesses since 2019.

We build websites, run SEO and improve AI search visibility. Contact: hello@example.co.uk, 01323 000000.

## Services
- [Web design](https://example.co.uk/services/web-design): what we build, timelines and prices from £3,000.
- [SEO](https://example.co.uk/services/seo): monthly retainers, what is included, typical results.

## Company
- [About](https://example.co.uk/about): founders, history, address, registration number.
- [Contact](https://example.co.uk/contact): phone, email, opening hours.

## Optional
- [Blog](https://example.co.uk/blog): articles on web design and search.

Step 4: Serve it correctly

Place it at the root: https://yourdomain.com/llms.txt, returning HTTP 200 with a text/plain or text/markdown content type. On WordPress, Yoast and Rank Math can generate it; on Next.js drop it in public/; on Shopify use a redirect to a hosted file if you cannot serve root files. Check it in a browser.

Step 5: Decide on llms-full.txt

The companion llms-full.txt contains the full text of your key pages in one markdown file, for tools that want to ingest everything at once. Useful for documentation sites and products with long reference material. Unnecessary for most businesses. If you create it, do not also publish separate markdown copies of every page as indexable URLs; that is duplicate content and can affect the pages you actually want to rank.

Step 6: Audit what it claims

Models that do read the file will treat it as your statement of fact. Check every claim, price and link. An out-of-date llms.txt confidently telling an agent the wrong price is worse than no file.

Step 7: Keep it in sync

Add "update llms.txt" to the checklist for any change to services, prices, contact details or site structure. A stale file is the most common state we find.

How to know it worked

Honestly: you will see very little. Check your logs occasionally for requests to /llms.txt; a handful from GPTBot or ClaudeBot is typical. The real test of your AI visibility is the prompt tracking in our measurement guide, not this file.

Common mistakes

  • Publishing llms.txt and reporting AEO as done.
  • Generating indexable .md copies of every page.
  • A description in the file that differs from the About page and schema.
  • Skipping the audit and shipping wrong facts.

Frequently asked questions

Does llms.txt improve AI search visibility?

+

There is no evidence that it does. Google's May 2026 guide lists it among things to ignore for Google's AI features. Ahrefs analysed 137,000 domains and found 97 percent of llms.txt files received no requests in May 2026. Correlation studies across hundreds of thousands of domains found no citation lift. Treat it as documentation for AI agents, not as an optimisation.

Is there any harm in publishing one?

+

The file itself is harmless. The harm comes from two habits around it: creating indexable markdown copies of every page (duplicate content), and treating the file as your AEO strategy while the real problems (blocked crawlers, JavaScript-rendered content, vague copy) go unfixed.

Who does use llms.txt?

+

Mainly AI coding assistants and agents reading developer documentation, which is the use case the format was designed for. Some AI crawlers occasionally fetch it. If your customers are developers or you expect agents to act on your site, it has a real job. Otherwise it is a tidy summary that costs nothing.

Sources

  1. Ahrefs: We analyzed 137K sites, 97% of llms.txt files never get read · ahrefs.com
  2. Google Search Central: Optimizing for generative AI features (mythbusting) · developers.google.com
  3. llms.txt specification · llmstxt.org

Richard Daniel

Automation and Delivery Lead, Emerging Group

Richard leads automation and delivery across the Emerging group, working with EDP on client websites and with ETT on enterprise AI and process automation. He is the person who turns an audit finding into a working fix: crawler access, rendering, tracking, structured data and the plumbing that most marketing teams never see. He writes the technical guides on this site.

Guide15 Sept 2026 · 3 min read

How to get indexed by Bing (because ChatGPT search leans on it)

OpenAI names Bing among the search providers behind ChatGPT search, and independent analyses in 2026 found most of ChatGPT's cited pages also rank in Bing's top results. Copilot is Bing. If your Bing coverage is thin, so is your ChatGPT visibility. Here is the fix.

Richard Daniel
technicalai-searchmeasurement

← All guides