Your llms.txt is doing about half the job you think it is, and the missing half is not in the file. I checked ours before I wrote this. Same problem.
The direct answer: llms.txt works, but not for the thing most people bought it for. It does nothing for your Google ranking, because Google Search Central states that you do not need to create new machine readable files, AI text files or markup to appear in AI Overviews or AI Mode. Where it does work is with agents that fetch it on demand while answering somebody, and there it only works properly if you shipped the second half of the specification, which almost nobody did.
I have published one of these files for imisofts, it is part of why an AI sent me a client, and I have built answer engine optimization for clients on top of it. That is exactly why the gap in mine annoyed me enough to write this.
Everybody printed the menu. Almost nobody opened the kitchen the menu points at.
The mistake: filing it under SEO
The file gets sold as a ranking asset. Generators, validators, checkers, a WordPress plugin, an audit line item. Then somebody publishes one, watches Search Console for a month, sees nothing, and concludes the whole idea was hype.
Both halves of that are wrong. Google Search Central says plainly that no new machine readable file is required for its AI features, so there was never a ranking to win. And the author of the specification is just as plain in the other direction: the expectation was that llms.txt would be useful for inference rather than training, and that is how it has been used. Inference means an agent reading it live, mid answer, on behalf of somebody who asked a question about you. Different instrument, different scoreboard.
"We published llms.txt. Did rankings move?"
Google says no file is needed for AI Overviews or AI Mode. There was no ranking on offer.
An agent fetches it while answering a buyer, follows a link, and needs clean text at the other end.
"Does a link in it lead anywhere an agent can read?"
The spec is two proposals. You shipped one.
Read llmstxt.org properly, in the version 2 revision published on August 10, 2026. It proposes the llms.txt file. It then proposes something separate: that pages an agent might need also publish a clean markdown version at the same URL, either with .md appended or with the extension replaced. The document calls these two proposals, and it says the links inside your llms.txt should point at that LLM friendly content.
That is the part everyone skipped. Open your own file and look at where the links go. If they resolve to ordinary HTML, an agent that follows one lands in navigation, a cookie banner, a footer and a pile of JavaScript, and then spends context stripping all of it back out to find your pricing. Which is the exact cost the file exists to avoid. The menu is clean. The kitchen is still locked.
I ran the count on ours before publishing this. Fifty seven links. Zero pointing at a markdown version of anything. We built the useful half and shipped the half that looks finished, and we are fixing it this week.
Lighthouse cannot fail you for not having one
Chrome shipped an llms.txt audit inside Lighthouse, under agentic browsing. Half the industry read that as Google now checking for the file, and started selling the check.
Read what the audit actually does. Chrome documents it flagging a page when a server error occurs while retrieving llms.txt. If the file is simply not there and the server returns a 404, the audit is marked Not Applicable, because providing the file is optional. So the audit cannot fail you for not having one. It can only fail you if your server breaks while trying to serve it.
Which leaves two Google teams saying compatible things that get quoted as a contradiction. Search says the file is not an input. Chrome says the file is optional and only checks that it is not erroring. Neither one is asking you for it. The people asking for it are agents, and they are not scored by either.
Version 2 also quietly moved two things nobody has implemented yet. The file no longer has to sit at your domain root: it can live at any path, it covers everything under that path, and where more than one applies an agent should use the most specific. And it defines two link relations, one pointing at the markdown version of a page and one pointing at the llms.txt that describes it, both deliverable as an HTTP Link response header from your CDN without editing a single page.
Four moves on your llms.txt this week
The bottom line
Does llms.txt work? For agents, yes, and only as well as what sits behind its links. For Google rankings, no, and the people who told you otherwise were selling a generator.
The tell is who publishes one properly. OpenAI, Anthropic and Google all serve an llms.txt for their own developer documentation, and documentation platforms generate the markdown pages alongside it as a matter of course. The AI labs did not ship the file and stop. They shipped the thing the file points at.
None of this is expensive. A markdown build step, a CDN header, and a file split by path. What it costs is admitting that the version you already published, and quietly counted as done, was the easy half. Mine was.
Book a 30 minute call: cal.com/zeeshanwaheed/30min or email [email protected]. I answer the first one myself, and you can see who you are talking to before you book.