GEO—generative engine optimization—borrows its name from the SEO playbook some brands are importing into AI search. The research says it isn’t working. The bigger risk is what happens to AI search visibility—and the trust buyers have built with LLMs—if it does.

Search engines never asked anybody to trust them. It handed you a page of links and left the judging to you, and after twenty years everybody had a rough sense of which links were bought. An AI engine hands you a judgment instead—ask which platforms you should be looking at and three names come back with a reason attached to each, and buyers are taking that at close to face value in a way nobody ever took a results page. Search was a directory. This is, for all intents and purposes, a referral.

That makes it the most valuable thing to happen to B2B discovery in a generation, and it’s exactly what brands are now trying to pry open, using more or less the playbook they ran with when search was new.

The downside isn’t the same this time. Manipulating search made results worse and people adapted by scrolling further down the page. Gaming makes buyers stop believing the answer, and there’s no adapting to that. They lose a tool they’d started to depend on, and we lose the one channel where a brand can be recommended rather than merely found.

The SEO tactics brands are importing into AI search

Those old SEO tactics have been tested on AI and the results are in. Mint Studios published a test log in February, failures included, which is more than most firms have done. They buried white text inside articles instructing the model which brand to recommend. Nothing. They published an llms.txt file. Nothing. SE Ranking looked at llms.txt across 300,000 domains and couldn’t find a traceable citation benefit, and a peer-reviewed Princeton and Georgia Tech study found keyword stuffing produced no gain at all. Hidden text and keyword stuffing failed in search a long time ago, and most of the people doing search work professionally came through that era and moved past it. The tactics being imported into AI engines are the ones the SEO field already discarded, and they fail for a deeper reason than age—search never had a trust layer. It matched queries to pages. An AI engine checks whether independent sources agree about you before it recommends you, and no amount of hidden text or metadata can fake that agreement.

Honest SEO practice transfers to AI engines—depth instead of keyword targeting, real author credentials, consistent entity data, earning links instead of buying them—precisely because it was building the kind of credibility AI engines now check for. What’s happening now is an attempt to essentially hack LLMs as though the lessons of the early SEO days aren’t still applicable. 

The point isn’t that these attempts were unsuccessful. The point is what happens to the channel if someone figures out how to make them work. Every shortcut that succeeds teaches the next brand to take it, and the trust that makes AI recommendations valuable is the first thing that erodes when they do.

Some of the newer moves are harder to dismiss. Firms are writing articles they don’t put on their own blog, because the pieces have no real argument in them and are developed for bot and not a reader. Sure a  person can read one, but a  human isn’t who it’s for. I can’t draw a clean line between that and a well-structured explainer that answers a real question, because there isn’t one, and somewhere along that road you stop writing for a reader and start writing for a retrieval system with nobody ringing a bell when you cross it.

And then there are paid mentions. I came across a network recently that pays publishers to mention brands and sells it as a route into AI answers. The pitch is accuracy, and there’s something real in it—but an engine citing three-year-old pricing is a genuine problem. An engine quoting a stale roundup is a thing somebody should fix, but not a thing somebody should sell. Anything you pay for stops being earned the moment you pay for it, because that’s what the word means. Pay three publishers and you didn’t buy an agreement. You bought one source and ran it three times. An engine can’t see the invoice, so it may well work, and that’s the part that bothers me, and it should bother you too .

Why B2B brand authority can’t survive a shortcut

Not one person in that chain has to be a bad actor. Every step looks reasonable next to the step before it. That’s exactly the problem, because there’s no obvious place for a sensible marketer to stop.

We’ve seen this before, twice. Search results filled up with sponsored placements until everybody learned to scroll past the top of the page. Feeds filled up with brands selling things until nobody would think to ask a feed for a recommendation. No single company wrecked either one. Everybody optimized at once, every move defensible on its own, and the people who got hurt were the ones who had built an audience the slow way. Emma Chase, our director, made that point on a production call last week and it’s better than anything I had.

The difference this time is that the bet behind all of it—that nobody will notice—falls apart faster in B2B. Somebody signing a seven-figure platform deal earned their way into that chair and doesn’t take one recommendation at face value. They cross-check it. They call somebody. DerivateX found that ChatGPT cites a B2B vendor’s own site only 11.6% of the time when recommending it across 40 categories and 233 recommendations, so nearly everything a buyer sees is somebody else vouching for you. A paid endorsement sits inside that pool, and a B2B buyer will find it. When one buyer finds one, they don’t discount one brand. They discount the channel.

What actually builds AI search visibility

If you want to know where you actually stand, pull the ten questions a buyer would ask an AI engine before shortlisting anybody in your category and look at whether you show up in the comparisons and roundups that come back. If you don’t, there’s no file you can upload that fixes it. Earned coverage has started showing up in AI recommendations for our clients inside about 90 days, and it keeps compounding after that.

Nobody has to be the villain here. Enough of us just have to reach for the shortcut at the same time, and we’ll wreck the one channel we’re all trying to get into.

CTA: The Brand Authority Index diagnoses where your brand stands across the seven signals AI engines actually weigh. [Request an assessment →]