The format is deliberately minimal. The file lives at the root of your domain (yoursite.com/llms.txt) and is written in plain Markdown: an H1 with your brand name, a one-line blockquote saying what the site does, then sections of links, each with a short annotation explaining what a model will find there. The point is to spare AI crawlers the noise of a rendered HTML page — navigation, cookie banners, scripts — and hand them the ten pages that actually explain your business. Some sites pair it with a fuller companion file that inlines whole documents, but the curated index is the core idea.
Here is the honest part: no major AI provider has publicly confirmed that its crawlers read llms.txt, and there is no proven effect on whether or how you get cited in AI answers. Treat it the way early adopters treated XML sitemaps — a cheap, low-risk courtesy to machines that may pay off as the convention matures. It should never be your main move. Whether an assistant can quote you depends far more on pages that answer real questions directly, clean structured data, and content worth citing — the fundamentals covered in our answer engine optimization guide.
On Brohns, llms.txt shows up as one item on a longer checklist. The GEO auditor agent scans how findable your site is in AI answers like ChatGPT and Perplexity, benchmarks you against competitors, and generates concrete fixes — an llms.txt file is exactly the kind of low-effort improvement it can propose when yours is missing. Like everything an agent produces on the canvas, those fixes arrive as drafts with the agent's reasoning attached; nothing touches your live site until you have read the proposal and signed off. That keeps a speculative convention where it belongs: worth trying, with your eyes on it.