There is no published confirmation from OpenAI, Anthropic or Google that they consume llms.txt, so anyone promising ranking lift from it is overclaiming. It is still worth adding, for a different reason: it is cheap, it cannot hurt, some tools do read it, and writing one forces you to state — in machine-readable form — what your site is and which pages matter. Treat it as hygiene and entity clarity, not as a ranking lever.
llms.txt is a proposal by Jeremy Howard: a Markdown file at your site root that gives AI systems a curated map of your most important content. Structurally it sits alongside robots.txt (which controls crawling) and sitemap.xml (which declares URLs) — but unlike those two it is not a standard any major engine has committed to honoring.
The honest state of adoption: the proposal's own site presents no evidence of engine consumption, and no major engine has published support. So the defensible claim is "present and well-formed llms.txt = a low-cost hygiene signal of unknown direct impact", not "llms.txt improves your AI visibility".
The indirect value is real, though. Producing a good llms.txt requires deciding what your site actually is, which pages are canonical entry points, and how to describe them plainly. That clarity tends to improve your on-page structure, internal linking and metadata — all of which DO affect extractability.
What matters far more, if you have limited time: make sure you are not blocking the retrieval bots in robots.txt (that one genuinely makes you invisible in live AI answers), ship correct schema.org structured data, and make your key claims extractable from raw HTML rather than JavaScript.