Searches for llms.txt have grown roughly twentyfold in the past year, and most of what you'll find is either breathless ("the new SEO!") or dismissive ("nobody reads it"). Both takes are lazy. We ship an llms.txt on this site, we have opinions about what it earned us, and this is the guide we wish had existed when we wrote ours: what the file is, what it demonstrably does not do, and a template that takes about an hour to fill in.
Key Takeaways
- llms.txt is a markdown file at your site root that gives AI systems a curated, token-efficient summary of who you are and where your important content lives.
- It is an offer, not a control. Unlike robots.txt, nothing forces any AI to fetch it, and no major provider has confirmed automatic ingestion.
- It is not an SEO play. There is no evidence it affects Google rankings, and you should distrust anyone selling it as one.
- It still earns its keep: AI agents doing research fetch it when pointed, and writing it forces you to state the facts you want machines to quote.
- The llms-full.txt variant inlines your full content for single-fetch reading. Ship both; the marginal cost is nearly zero.
What llms.txt Actually Is
llms.txt is a proposed standard from Jeremy Howard of Answer.AI, published in September 2024. The idea is simple: HTML pages are terrible for language models. Navigation, scripts, cookie banners, and layout markup burn through an AI's context window before it reaches your actual content. So you publish a plain markdown file at yoursite.com/llms.txt that says who you are and links to what matters, in a format models parse natively.
The spec (maintained at llmstxt.org) asks for a specific shape:
- An H1 with your site or project name. This is the only required element.
- A blockquote with a one-paragraph summary of what you do.
- Any markdown you want: key facts, positioning, plain-language descriptions.
- H2 sections containing markdown link lists to your important pages, each with a short description.
- An optional section literally titled Optional, marking links an AI can skip when context is tight.
Its companion, llms-full.txt, goes further: instead of linking to your pages, it inlines their full content into one large markdown file, so a model can read everything you publish in a single fetch.
What It Doesn't Do (Read This Part)
Here is the section the "new SEO" crowd leaves out.
No major AI provider has confirmed automatic ingestion. As of mid-2026, neither OpenAI, Anthropic, nor Google has stated that their assistants automatically fetch llms.txt when browsing or training. Server logs across the industry show AI crawlers mostly hitting regular pages. If someone tells you ChatGPT reads llms.txt on every visit, ask them for their evidence, because the public record does not show it.
It is not a ranking factor. Google's search representatives have compared llms.txt to the keywords meta tag, which is search-engine speak for "we ignore it." If your goal is ranking in Google, your energy belongs in content and technical SEO, not this file.
It is not robots.txt for AI. The comparison gets made constantly and it is backwards. robots.txt restricts crawlers and is broadly honored. llms.txt restricts nothing. It is a standing offer: here is the clean version of us, use it if you want it.
Why We Ship One Anyway
Three reasons, in descending order of hard-headedness.
Agents read it when pointed. The growing case is not passive crawling; it is agentic research. When someone asks an AI assistant with browsing to research your company, agents frequently check for llms.txt the way developers check for a README, and coding tools let users feed llms-full.txt directly as context. Documentation platforms like Mintlify now generate these files automatically for thousands of sites. That is where adoption is real today.
Writing it sharpens your facts. The exercise forces you to write down, in quotable plain language, what you do, for whom, since when, and with what results. Those exact sentences are what an LLM lifts when it describes you. If you have never audited what AI assistants currently say about your business, do that first; the results usually fund the rest of the project. It is the core of what we do in AI search optimization.
The cost is an hour. A text file, a deploy, and a line item on your publishing checklist. Cheap forward bets on plausible standards are how you end up early instead of late. This one costs less than your monthly coffee budget and might matter. Might is enough at this price.
The Template
Copy this, replace the placeholders, save it as llms.txt at your site root. Ours is live at gravitas.digital/llms.txt if you want a working example to steal from.
# {Company Name}
> {One paragraph: what you do, for whom, from where. Write it like
> you are answering "who are these people?" for someone smart
> who has never heard of you.}
## Key Facts
- Founded: {year}
- Location: {city, state}
- Founder: {name, title}
- {2-4 more facts you want quoted: clients served, specialties,
credentials. Numbers beat adjectives.}
## Services
- {Service}: {one-line description} - {full URL}
- {Service}: {one-line description} - {full URL}
## Proof
- {Case study or result}: {full URL}
- {Case study or result}: {full URL}
## Content
- {Your best article}: {full URL}
- {Your best article}: {full URL}
## Contact
- Website: {full URL}
- Email: {address you actually answer}
## Optional
- {Deeper pages an AI can skip when context is tight}Rules that make the difference between a useful file and noise: full absolute URLs on every link, one line of description per link, numbers instead of adjectives, and no marketing language a model would have to decode. If a sentence would embarrass you when read back verbatim by a chatbot, rewrite it.
Keeping It Alive
A stale llms.txt is worse than none, because the whole point is being the trustworthy version of your facts. Two habits cover it:
- Add "update llms.txt and llms-full.txt" to your publishing checklist. Every new page or post either goes in the file or consciously doesn't. Ours gets updated in the same commit as the content.
- Add the discovery hint to your pages' head markup: <link rel="llms" href="/llms.txt" type="text/plain">. Not part of the core spec, but it costs one line and makes the file findable from any page.
How to Test It
Open any AI assistant with browsing and ask: "Read {yoursite}/llms.txt and tell me what this company does, who it serves, and why someone would hire them." If the answer comes back accurate and specific, your file works. If the assistant misses things you care about, the file is missing them too. This test takes ninety seconds and is more honest feedback than most marketing audits.
The Bottom Line
llms.txt will not rank you on Google, and anyone selling it as SEO is selling you a keywords meta tag with a 2026 haircut. What it is: the cheapest possible way to control the machine-readable version of your story, at the exact moment more buying research is happening inside AI assistants. We shipped ours in an afternoon and it has been quietly doing its job since.
If you want to know what AI assistants currently say about your business, and how to change it, that conversation is what we do all day.