The llms.txt file and the AI-readable site
There's a quiet new file showing up at the root of well-run sites. It's a short note, written for machines, that says: here's what we are, and here's where the truth lives.
You know robots.txt — the file that tells crawlers where not to go. There's a newer cousin worth knowing: llms.txt. It does the opposite job. Instead of fencing machines out, it hands them a clean, curated map of what your brand is and where the authoritative answers sit.
A guided tour for the machine
Think of it as a concierge note left at the root of your site. A short, plain-language file that names the company, points to the pages that actually matter, and explains in a sentence or two what each one covers. No clever markup, no maze. Just the truth, laid out in the order you'd want it read.
Why bother? Because LLMs cite only two to seven domains in a typical answer. The model isn't reading your whole site charitably — it's looking for the cleanest, most extractable source. A good llms.txt makes you that source.
The file is only half of it
A map is useless if the territory is a swamp. The file points; the site still has to be genuinely readable underneath. That means a clean architecture, content shaped like answers, headings that mean something, and pages fast enough to be crawled fully.
The most AI-readable site isn't the one with the most words. It's the one a machine can understand without effort.
For a property studio client, we paired an llms.txt with restructured listing pages — each residence its own clear entity, each fact where a machine expects it. The file invited the reader in; the structure rewarded it.
llms.txt is young, and not yet universal. But the instinct behind it is exactly right: stop hoping machines understand you, and start making yourself easy to read.