Somebody is going to sell you an llms.txt file. Don't buy it.

I taught journalism for eleven years, Baylor first, then BU. Every single semester, a student asked me for the trick.

There is no trick. There has never been a trick. But somebody is going to try to sell you one this quarter, and right now it’s called an llms.txt file, a little text file that supposedly tells AI models how to understand your site. Sounds reasonable. Of course you want one.

Except there’s no evidence anything reads it. Otterly ran a 90-day server-log study: the file got 84 hits against more than 62,100 AI bot visits to the rest of the site. One-tenth of one percent. A separate analysis of 300,000 domains found zero correlation between having it and getting cited. No major AI lab has confirmed their systems consume it, and most of them don’t publish one on their own sites.

Schema markup is in similar territory. Ahrefs ran a controlled test on 1,885 pages that added structured data and measured the citation effect between −4.6% and +2.2% depending on platform. Statistically, nothing. I’m not saying skip them. They’re cheap. Do them in an afternoon.

I’m saying don’t pay a retainer for them. And I’ll tell you why we put that in writing on our own blog: a retainer built on files nobody reads is the easiest money in this category, and we would rather have the client for five years. Our audits ship with a list of the things we are specifically not going to charge you for. Nobody has ever left that meeting trusting us less.

When a vendor’s core deliverable is a file nobody reads, that tells you what they’re really selling. I told my students the same thing every year: the shortcut is always somebody else’s business model.