"Add an llms.txt file and you're SEO-optimised for AI." There's no ranking signal involved. What it's for is narrower and more specific.
An llms.txt file isn't a ranking mechanism, and treating it like a new kind of SEO misreads what it's for. What it can genuinely help with is narrower and more useful: telling an AI agent what a business can do, not making its content rank better anywhere.
5 min read
What it isn't
A bed and breakfast owner adds an llms.txt file to the site, the way she once added a sitemap, and describes it to a colleague as "SEO for AI." Nobody corrects her, because the file lives in the same spot a sitemap does and gets talked about in the same breath. It isn't doing the same job at all.
Nothing about an llms.txt file makes a business rank higher in an AI-generated answer the way better SEO used to improve search rankings. There's no algorithm being fed a signal here. A file at the root of a domain, describing a site in plain language, doesn't compete for attention the way a search result does. Calling it "SEO for AI" borrows a framework that doesn't apply.
The job it does instead
Where it genuinely helps is narrower: pointing an agent toward what a business can do, not describing what the business says about itself. A well-written one tells an agent this is a bed and breakfast, here's where the booking information lives, here's what an agent can check or do here, more like a signpost toward capability than a description written to be read and enjoyed.
Its own real limit
Even used well, it's still a file sitting at a domain's root, following a convention, hoping to be found and read the same way robots.txt or a sitemap always has been. That's a real improvement over nothing. It's still the same shape as a chat widget or extended page markup: a text file layered onto the existing website-and-crawler model. A business's own facts, made directly and continuously reachable, is a different thing entirely.
Where this shows up for a business
A business's real facts stay continuously and directly reachable through Selfe, rather than relying on a static file at a domain root to be found, read and correctly interpreted by whatever's looking.
Should we bother adding one at all?
It's a reasonable, low-effort thing to have. Just don't expect it to do the job structured, connected facts do.
What should it say?
What an agent can do here, availability, booking, terms, in plain language, rather than a persuasive description of the business.
A lot of what you've heard about agentic commerce isn't quite right.
Every myth in this hub comes from the same mistake: judging the future by how well it fits today's tools, a page, a widget, a chat window, a markup file, instead of asking what would make sense if you were starting clean. The actual destination underneath all of it is simpler than any of the myths: AI agents talking directly to data, not reading pages built for people.
"AI can just read the page, you don't need structured data." We tested that claim. It doesn't hold up.
The usual fix people reach for is extending the markup already on a page, more tags, richer fields, describing more of the business in the same place. That's still a page-shaped answer. The facts an AI agent needs don't have to live on a page at all, marked up or not.
"WebMCP is the real answer for autonomous agents." Maybe eventually. Worth asking why it's needed at all.
WebMCP is genuinely early, worth flagging as unsettled rather than treated as decided either way. Even taken on its own terms, it's answering a narrower question than it sounds: how an AI agent copes with a web built for people, not what agentic commerce looks like if nobody had to cope with that at all.