Short answer: llms.txt is a Markdown file at yourdomain.com/llms.txt that tells language models who you are and which pages matter. The format is simple: one # title, one > summary, then ## sections with lists of links. Copy the template below, fill in the brackets and put it in your public/ folder.
When I audited 286 developer portfolios for an earlier study, 85% didn't have one. It's one of the cheapest items on any AI-readiness checklist, so here's how to do it right.
What llms.txt is (and isn't)
Jeremy Howard proposed llms.txt in September 2024. The idea: HTML pages are full of navigation, scripts and layout that language models have to dig through. A short Markdown file at a known address gives them the essentials directly.
Be honest with yourself about what it does:
- It is a curated summary: who you are, what you offer, and links to your best pages.
- It isn't a ranking signal anyone has confirmed. OpenAI, Anthropic and Google haven't said they use it.
-
It doesn't replace
robots.txt(what bots may fetch) orsitemap.xml(every URL you have). -
It doesn't fix a site that's blank without JavaScript. If a bot sees
<div id="root"></div>, pre-render first.
So why bother? It takes five minutes, costs nothing, and forces you to write a clear, two-line description of what you do. That sentence is useful everywhere: your bio, your meta description, your LinkedIn.
The format
The spec is short. In order:
-
One
#heading with the name of your site or project. This is the only required part. -
One
>blockquote with a short summary. Make it the sentence you'd want an AI to repeat about you. - Optional paragraphs or lists with more context, but no extra headings.
-
##sections, each with a list of links in the form- [Title](url): one-line note. -
An optional
## Optionalsection for links a model can skip when it's short on space.
The template
Copy this, replace everything in brackets, and delete the sections you don't need:
# [Your name or business]
> [What you do, for whom, and where, in one or two sentences.
> Example: Frontend developer in Lisbon who builds fast, accessible
> React apps for early-stage startups.]
[Optional: one short paragraph with useful facts. Languages you work in,
availability, years of experience, how quickly you reply.]
## Services
- [Service one](https://[yourdomain.com]/services/one/): [what it is and who it's for]
- [Service two](https://[yourdomain.com]/services/two/): [same]
## Projects
- [Project A](https://[yourdomain.com]/projects/a/): [what you built and one concrete result]
- [Project B](https://[yourdomain.com]/projects/b/): [same]
## Contact
- [Email](mailto:[you@yourdomain.com]): [usual reply time]
- [GitHub](https://github.com/[you])
- [LinkedIn](https://www.linkedin.com/in/[you]/)
## Optional
- [Blog](https://[yourdomain.com]/blog/): [what you write about]
- [About](https://[yourdomain.com]/about/)
Where to put it
The file has to be reachable at the root of your domain, served as plain text:
| Stack | Path in your repo |
|---|---|
| Astro, Vite, Next.js | public/llms.txt |
| SvelteKit | static/llms.txt |
| Hugo | static/llms.txt |
| WordPress | Upload to the site root, or use an SEO plugin that supports it |
| Plain HTML | Next to your index.html
|
After you deploy, open https://yourdomain.com/llms.txt in a browser. If you see your Markdown as text, you're done. If you see your homepage or a 404, your host is rewriting the route. That's common with single-page app setups that send every path to index.html.
Five mistakes that make it useless
- Linking to pages that are blank without JavaScript. The model follows the link and finds an empty shell. Fix the page first.
- Writing a keyword list. "React, Vue, Node, SEO, freelance, cheap" tells a model nothing. Write sentences.
- Leaving it stale. If your llms.txt lists a service you dropped a year ago, it's worse than not having one. Update it when you update your site.
-
Bare URLs with no notes.
- https://site.com/projects/is half the value. The note after the colon is what helps a model decide which link answers a question. -
Blocking the bots that would read it. If your
robots.txtblocks GPTBot or ClaudeBot, they won't fetch llms.txt either. In my study, 17 of the 19 sites blocking AI crawlers had that block written by Cloudflare, not by the owner.
What about llms-full.txt?
Some documentation sites also publish llms-full.txt, with the full text of every page in one file. It isn't part of the original proposal. It makes sense for docs with hundreds of pages. For a portfolio or a small business site, llms.txt is enough.
Check it
Run your site through the free audit. It checks whether your llms.txt exists, whether AI crawlers are blocked in robots.txt, and whether your pages have content without JavaScript, in about ten seconds and with no sign-up.
Originally published on wickz.dev.
Top comments (0)