Short version: it costs an hour, it is harmless if accurate, it is actively damaging if it contains claims you cannot support, and it is not a strategy. If someone...
Short version: it costs an hour, it is harmless if accurate, it is actively damaging if it contains claims you cannot support, and it is not a strategy. If someone is selling it as the centrepiece of an AI visibility programme, that tells you more about the seller than about the file.
A proposed convention: a markdown file at your root that summarises what your site is and points to your most useful pages, intended to be read by language models rather than by people. Think of it as a curated index written for a machine reader — a cousin of robots.txt in placement, but descriptive rather than restrictive.
If a model or an agent retrieves it, a good file gives a compact, unambiguous statement of your entity, your offering and your best sources, without making the reader infer those from a navigation menu. That is a real if modest benefit, and it is the same benefit you get from a well-structured About page — clarity for a parser.
The file is a marketing surface, and marketing surfaces attract unverifiable claims. We regularly find files asserting precise performance figures — a citation rate, a success percentage, a client count — with no method and no source behind them.
That is worse than an empty file, for two reasons. It is a durable, machine-readable, timestamped record of a claim you may not be able to defend. And if a model does ingest it and reproduce the claim, you have laundered an unsupported number into an apparently neutral answer. If you would not put a number in a contract, do not put it in a file designed to be read by machines and repeated.
We audited our own file for exactly this and removed a performance figure from it, because our measured rate did not support the number that was there.
| Include | Leave out |
|---|---|
| What the organisation does, in one sentence a stranger would understand | Performance percentages without a stated method |
| Markets and languages served | Client names you do not have written permission to use |
| Links to your genuinely best reference pages, annotated | Every URL on the site — this is a curated index, not a sitemap |
| Pricing ranges, if you publish them elsewhere | Pricing that contradicts your own pricing page |
| Contact route and a last-updated date | Superlatives — “leading”, “best”, “number one” |
The ordering is the entire argument. The file is fine. It is simply sixth.
Not if it is accurate. It is a static text file, and it does not affect conventional crawling or ranking.
No. They do unrelated jobs: robots.txt controls access, llms.txt describes content. Keep both.
No. The value is curation. A file with thirty annotated links to your best material is more useful than one with three hundred bare URLs.
When your offering or pricing changes. Put a last-updated date in it so a stale file is visibly stale.
Want your own numbers instead of ours?
Send us your domain. We run the baseline on your category prompts and send back the raw answers, not just a score.
Đội ngũ chuyên gia Vidco Group sẵn sàng đồng hành cùng bạn
Bước 1 / 4
Chúng tôi sẽ liên hệ trong vòng 2 giờ làm việc.