Websites are built for people, and most pages carry a lot besides the facts: menus, banners, scripts, pop-ups. A language model that reads a page has to dig the useful part out of that. The idea behind llms.txt is to hand it a clean summary instead, in one small file, so that it does not have to guess which of your pages matter.
What llms.txt is
llms.txt is a proposed file that sits at the root of a website, so it loads at your-domain.com/llms.txt. It is plain text written in markdown. It is meant to be read by language models and by the tools built on them, the same way a sitemap is meant to be read by search engines.
The important word is proposed. It is not an official standard, and no one is obliged to read it. Treat it as a courtesy to any tool that does: a tidy, honest description of your site in your own words. It is related to, but different from, the sitemap and robots.txt you may already have.
What goes inside
The proposal describes a simple layout. These are the parts, in order.
| Part | What it holds |
|---|---|
| A title | The first line, written as a markdown heading: the name of your business or site. |
| A short summary | One or two sentences in a quote block that say what you do, for whom and where. Plain words, no slogans. |
| Optional notes | A few lines of detail that help a reader interpret the links, such as the areas you serve. |
| Sections of links | Headings that group your key pages. Under each heading is a list of links, each with a few words on what the page holds. |
| An optional section | Links that are useful but can be skipped when a tool has little room to read. |
A small example
Here is a made-up file for an imaginary bakery. The business and its pages are invented for illustration.
# Example Bakery > A family bakery in Hyderabad making custom cakes > and daily breads. Orders are taken by phone and online. ## Cakes - [Custom cakes](https://example.com/cakes): sizes, flavours and lead times - [Birthday cakes](https://example.com/birthday): popular designs and how to order ## Visit us - [Location and hours](https://example.com/contact): address, phone and opening hours ## Optional - [Our story](https://example.com/about): how the bakery started
Notice what it does not do: no claims about being the best, no keyword lists. It says what the business is and where to find each kind of fact. If you would rather not write the file by hand, the free llms.txt generator builds one in your browser.
What it cannot do
It cannot make an AI engine read it, and it cannot make an engine name you. It does not block or allow crawlers; that is the job of robots.txt. It does not replace clear pages: if the page it points to is vague, a neat summary will not fix that. And it is not private. Anyone can open the file, so list only pages you are happy to have read.
Whether a given AI product reads the file can change, and Omnishown does not claim to know which ones do today. The safe way to think about it is as low-cost groundwork: it takes under an hour, and it is useful to any tool or person that does look.
Do you need one?
You do not need one. Your visibility in AI answers will depend far more on whether your pages state your facts plainly, whether your business details agree everywhere, and whether crawlers can read your site at all. Do those first.
Once they are in order, an llms.txt file is a reasonable extra, most useful when your site is large or has pages that are hard to tell apart from their titles: a documentation site, a shop with many categories, a business with several locations. For a five-page local business site, it is optional.
How to write a good one
- Write the summary last, but keep it short. One or two sentences: what you do, for whom and where. Use the same words your homepage uses.
- List the pages that answer real questions. Services, prices if public, contact and hours, key guides. Leave out login pages, cart pages and thin pages.
- Describe each link in a few plain words. “Sizes, flavours and lead times” is more useful than “Click here”.
- Use full, working URLs. Open each one before you publish. A link that returns an error teaches a tool to distrust the file.
- Only say what is true. Do not add claims you would not put on your homepage. The file should match the pages it points to.
Keeping it true
The risk with llms.txt is not writing it. It is forgetting it. A year later the site has new pages, old URLs have moved, and the file still describes last year's business. Make it part of your release checklist: when a page it lists changes, the file changes too.
That is the sort of check Omnishown Check is built for. It reviews each push to your repository for SEO, AEO and GEO issues, and Omnishown Fix sends any proposed change as a pull request that a person approves. Omnishown Diff shows what changed after each release. To see whether AI engines name you, read how to check by hand, or look at Omnishown Watch. The AI visibility guide has the wider checklist, and you can join the pilot to try Omnishown on your own site.