Someone tried to solve an AI visibility problem with a file. Google's John Mueller told them, in public, that the file does nothing. It is a small exchange, and it is the cleanest example this year of a pattern worth internalizing before you waste a quarter on the next one.
The person had a real problem. Two other people share their name, both more prominent online, so search and AI systems kept confusing them. A reasonable frustration. The chosen solution was the issue.
The File That Does Nothing
The proposed fix was a file called llms-author.txt, meant to hold a job title, location, and practice area so AI systems could tell this person apart from their namesakes. It is not a standard. It is a made-up format hoping to become one by being used.
Mueller's answer was flat. Google does not use llms.txt or llms-author.txt, and he is not aware of other crawlers or language models using them either, beyond a handful of SEO tools talking to each other. He went further on a related idea, the Content-Signal header, calling it something a CDN made up that has no effect whatsoever on any crawler or model, and noting it just adds file bloat.
So two of the tidy technical fixes floating around AI-SEO circles right now do precisely nothing, confirmed by the person at Google whose job is to know. Not underrated. Not early. Nothing.
Why This Keeps Happening
The appeal of a file is obvious. AI visibility feels large and uncontrollable, and a text file you drop at your domain root feels small and controllable. Publish the file, check the box, tell yourself you handled the AI thing. It is the same emotional trade that made every SEO shortcut popular, and it fails for the same reason.
I wrote about the llms.txt file almost nobody reads, and this is the sequel with the same ending. A file is an instruction you leave at the door. It assumes the model is looking for your permission or your self-description. It is not. The model is building a picture of you from what the open web already says, and a file you wrote about yourself is the least trustworthy input in that entire process.
This connects to a harder truth I keep returning to. The manufactured versions of AI visibility do not work, whether the manufacture is a self-serving file or a bought endorsement. I called buying AI citations the new link farm, and the author.txt file is the gentler cousin of the same mistake. Both try to engineer the appearance of a signal the model is supposed to earn by reading reality. Both get ignored or filtered, just on different timelines.
What Actually Solves the Namesake Problem
The useful part is that the real fix was named, and it is not technical. The person's problem was not a missing file. It was a thin web presence next to two people with stronger ones. The model confused them because the open web gave it more and better material on the other two.
The solution follows directly. Build the presence. Interviews, podcasts, published work, mentions on sources the model already reads, a consistent professional identity stated the same way everywhere. Do that and the disambiguation happens on its own, because the model now has enough real signal to tell you apart. No file required, and no file could have substituted for it.
From building GEOflux, this is the pattern in the data, edition after edition. The brands and people represented accurately in AI answers are the ones with genuine, consistent, findable presence across sources the model trusts. The ones that are missing or confused almost always reach for a technical shortcut first, publish it, and wait for a result that never comes. The tool telling them their file does nothing would save them the quarter, if they listened.
The uncomfortable part is that the real fix is slower and less satisfying than dropping a file at your domain root. You cannot ship presence in an afternoon. It accrues from showing up, over months, in the places your audience and the models both pay attention to. That slowness is exactly why the shortcuts stay popular, and exactly why they keep not working. The file is fast and does nothing. The presence is slow and does everything.
So take the free lesson from someone else's exchange with Google. Before you adopt the next AI-SEO file, header, or tag that promises visibility, ask one question. Is any major AI engine actually reading this, or is it a convention SEO tools invented to sell each other a sense of control. If the answer is the second one, and it usually is, skip it. Put the hour into the thing that has worked the whole time, which is being genuinely present and consistent where the model already looks. The shortcuts keep failing in public. The substance keeps working in quiet.