llms.txt is a plain Markdown file at the root of a website (yoursite.com/llms.txt) that tells AI models what the site is and which pages matter, in a form that is cheap for a model to read. Jeremy Howard proposed it in September 2024. It is a convention, not a standard, and no major AI product has said it reads one.
That last sentence matters more than most write-ups admit. Below is what goes in the file, what we found when we checked 220 company websites for one, and our honest view on whether it’s worth your afternoon. (Short version: yes, because it’s cheap, but it is not where AI visibility is won.)
What goes in the file
The proposal at llmstxt.org is short. The file is Markdown, and its parts come in a fixed order:
| Part | Required? | What it holds |
|---|---|---|
H1 (# Name) | Yes, the only required part | The name of the site or project |
Blockquote (> ...) | No | A short summary: what this is, in a sentence or two |
| Plain paragraphs or lists | No | Anything a reader needs to make sense of the links |
H2 sections (## Docs) | No | Lists of links, each [name](url) with an optional note after a colon |
## Optional | No | Links an agent can skip when it is short on context |
A minimal one for a design studio looks like this:
# 4080 Studio
> 4080 Studio is a brand and web design studio in Uttarakhand, India,
> founded in 2023. We design brands and websites for startups.
## Key pages
- [Services](https://4080studio.com/services): what we do and how long it takes
- [Pricing](https://4080studio.com/pricing): what's included in each project
- [Case studies](https://4080studio.com/work): finished client workThe proposal also suggests serving a clean Markdown copy of each important page at the same address with .md on the end. Many sites publish an llms-full.txt as well, with the full text of their key pages in one file. That one is a common habit, not part of the proposal.
What we found on 220 websites
In October 2026 we audited the homepages of 110 SaaS companies and 110 Indian startups, fetching /llms.txt from each the way a crawler would. 214 of the 220 gave a readable answer.
| US SaaS | Indian startups | Both | |
|---|---|---|---|
| Publish an llms.txt | 69% (75 of 109) | 50% (53 of 105) | 60% (128 of 214) |
| File opens with an H1, as the proposal asks | 95% | 94% | 95% |
| File opens by defining the company (“X is a…“) | 75% | 40% | 60% |
| Shopify’s default file, not written by the company | 0 | 7 | 7 |
| Median file size | 12 KB | 12 KB | 12 KB |
Three things stood out.
It is more common than its status suggests. A young, unofficial convention is on six in ten of these sites. Growth-stage SaaS companies led at 83%. For a file that no AI company has committed to reading, that’s a lot of adoption, and it says something about how much fear of missing out is driving AI search work right now.
Some of the files weren’t written by the company. Seven of the Indian llms.txt files were Shopify’s auto-generated agent instructions, on stores like Mamaearth and boAt. The file explains how an AI agent can shop the store. It never says what the brand is. If you’re on Shopify, check what’s at your /llms.txt before you assume you have one.
Sizes are all over the place. The smallest was 435 bytes. Twilio’s was 2.3 MB and Salesforce’s over 1 MB, which defeats the point of a short summary. 17 of the 128 files were over 100 KB. If the idea is to save a model the effort of reading your whole site, a file the size of your whole site doesn’t help.
The gap that matters more
The most useful finding wasn’t about the file at all. Of the 73 SaaS companies where we could read both the homepage and the llms.txt, 42 define themselves clearly in llms.txt (“Acme is a payroll platform for…”) but not in the opening of their homepage. Only 21% of the SaaS homepages opened with a sentence that names the company and says what it does. For the Indian startups it was 3%.
So the sentence answer engines want already exists. It’s just sitting in a text file that may never be fetched, while the page every crawler does read opens with a slogan. If you do one thing after reading this, copy the first line of your llms.txt into the first paragraph of your homepage. We wrote about how to write that line.
Does any AI product read it?
Not that anyone has confirmed. Here is what’s on the record:
- In June 2025, Google’s John Mueller wrote that “no AI system currently uses llms.txt”, and said server logs show chatbots fetching pages but not the file.
- Google’s own guide to AI features in Search says you don’t need special AI text files to appear, that Search doesn’t use them, and that creating one will neither help nor harm your visibility there.
- Chrome’s Lighthouse added an llms.txt check under its new agentic browsing category. It only flags a server error when fetching the file, not a missing one, and frames the file as a way to save browsing agents some effort.
- In our own October check of AI answers, Brave’s AI answer to “who founded 4080 Studio” listed our
/llms.txtamong its sources. That shows a search engine indexed the file like any other page. It doesn’t show anyone treats it as special.
OpenAI, Anthropic, Perplexity and Google have not said their assistants look for llms.txt when answering questions. Treat any article that tells you it “gets you into ChatGPT” with suspicion.
So should you publish one?
Yes, with modest expectations. It takes an hour, costs nothing to host, and does no harm. Coding agents and browsing agents are the likeliest readers, so it’s most useful if you have documentation or an API. For a marketing site, it’s a tidy summary of who you are that may or may not get read.
If you write one:
- Open with the definition. Company name, “is a”, what you do, who for. One sentence.
- Link the pages that answer buyer questions. Services, pricing, case studies, comparisons. Not your cookie policy.
- Keep it short. A few kilobytes. If you want a full-text version, put it in a separate file.
- Generate it from your content. Ours is built from the same data as the pages, so it never drifts out of date. You can see it at /llms.txt.
- Then fix the homepage. The crawlers that definitely read your homepage matter more than one that might read a text file.
Our llms.txt generator writes a first draft from your homepage, and the AI visibility checker tells you whether you already have one, and whether AI crawlers can reach and read the page that matters more. For the short definition, see the glossary entry. For the rest of the work, start with how to get your website cited by ChatGPT.
Quick recall
What is llms.txt? A Markdown file at the root of a site that summarises it for AI models and links its key pages.
Who proposed it? Jeremy Howard, in September 2024. It is not an official standard.
How common is it? 69% of the SaaS sites and 50% of the Indian startup sites we audited publish one.
Does ChatGPT or Google read it? No major AI product has confirmed it does. Google says Search doesn’t use it.
Is it worth doing? Yes, because it’s cheap. But a homepage that says what you do in the first sentence matters more.
Written by 4080 Studio.
