llms.txt: an hour of work, honestly assessed
What it is
robots.txt tells crawlers what they may not read. llms.txt does the opposite: it tells language models what is most worth reading, in the format they parse best, plain markdown. Site name, one-paragraph description, then curated links with a line of context each.
The honest status
It is a community proposal, not an IETF standard. Some AI crawlers read it, some ignore it, and the companies behind the big engines have made no binding commitments. Anyone selling llms.txt as a magic ranking lever is selling you weather. The case for it is asymmetric instead: near-zero cost, no downside, and it doubles as an exercise that forces you to state, in one page, what your site is actually about. Most companies fail that exercise on the first try, which is itself worth knowing.
How to write one
- Start with an H1 of your site name and a blockquote describing what you do in one paragraph a stranger would understand.
- Group your most important pages under H2 sections, each link with a single line of context.
- Include prices and facts you want quoted correctly. Engines repeat what they can parse.
- Keep it current. A stale llms.txt teaches engines stale facts about you.
A working example
# Kadima > Fixed-price SEO and AEO tasks, bought online like software. > Senior team, adversarial QA on every deliverable, prices public. ## Tasks - [Technical SEO Audit](https://heykadima.com/tasks/technical-seo-audit): $295, 5 business days - [AEO Visibility Audit](https://heykadima.com/tasks/aeo-visibility-audit): $149, how AI engines see you
The real thing is live at heykadima.com/llms.txt. Copy the structure, not the prices.
Where it fits
llms.txt is one small piece of AEO. The heavier levers are entity clarity, quotable page structure, crawler access and schema, which is the work our AEO Optimization task ships, llms.txt included.
What we saw after shipping ours
We publish our own at heykadima.com/llms.txt and watch server logs. AI crawlers do fetch it, GPTBot and PerplexityBot among them, and it contributed to our score on the Faro AI Search Readiness Scan moving from 59 to 95. What it did not do is transform citations overnight, and anyone claiming theirs did is selling something. It is an hour of work that makes every crawler visit more productive: a fair trade, honestly priced.
Mistakes that make it useless
Three appear constantly. Marketing copy instead of plain description, which models ignore. Listing every URL on the site, which buries the pages that matter. Writing it once and never updating, so it describes products you no longer sell, which is worse than absence because it feeds engines confident misinformation about you.
Sources
Read the proposal itself at llmstxt.org, then note what is absent from the engines' own documentation: neither Google's crawler documentation nor OpenAI's crawler documentation commits to reading the file. That absence is the honest state of adoption.
Common questions
Does Google use llms.txt?
Google has not committed to it; the file is a convention, not a standard. Gemini's answers still depend on classic crawling and structured data. Publish it for the engines that do read it, and expect nothing from the ones that do not.
Is there a required format?
Convention only: markdown, a short description of the site, then grouped links with one-line annotations. Plain, accurate and current beats clever in every observed case.
How often should it be updated?
Whenever products, pricing or positioning change, and quarterly as a floor. A stale file actively misinforms engines, which is the one outcome worse than not having one.
Want the whole AEO layer done properly?
AEO Optimization ships entity fixes, schema, content restructuring, crawler access and your llms.txt, in 7 business days.