GEO vs. SEO: How AI Search Engines Discover and Evaluate Brands

Ben cohen

Picture a buyer who needs a new project management tool for her team. On Tuesday morning she opens an AI assistant and describes her team, her budget, and the two tools her boss already mentioned. Ten minutes later she has a shortlist of three, a comparison, and a rough idea of pricing. She never opened a single website. And if your company isn't one of the three, you never even knew she was looking.

Ido

And the uncomfortable part is that nothing in your analytics tells you it happened. No visit, no bounce, no lost session. You just quietly weren't in the room. So the question for this episode is easy to say and hard to answer. How does a brand stay visible when the thing standing between the customer and the web is an AI assistant?

Ben cohen

Welcome to Be Visible. I'm Ben Cohen.

Ido

And I'm Ido Navarro. This show is about AI search, generative engines, agents, and what all of it means for websites and brands. I come at it from the product and website building side. Ben comes at it from growth and customer acquisition.

Ben cohen

Which means Ido says something technical and I ask what it costs. Let me start with why I think this is urgent. Forrester found that 94% of business buyers now use AI search when they're buying, and they rated it above vendor websites, product experts, and sales reps. On the consumer side, NielsenIQ puts it at 42% of consumers using AI to research products.

Ido

Those are survey numbers, so I'd treat them as a direction, not a precise measurement. But the direction is hard to argue with. A lot of the evaluation is happening before anyone visits you. It might happen even before a human is involved at all.

Ben cohen

So walk me through what actually changed. Because for twenty years the game was, rank on Google, get the click, win the customer.

Ido

Right. Visibility used to mean being found. You were a blue link, and a person decided whether to click. Now there's a whole chain between you and the customer. An AI system has to discover you. Then it has to understand you correctly. Then it has to consider you credible. Then it has to quote you, or mention you, or recommend you when someone is deciding. And more and more, an agent may need to actually do something on your site on the customer's behalf. Every link in that chain is a place you can drop out.

Ben cohen

Okay, but here's the question I know a lot of listeners are asking. Is this really different from SEO? Because half of what I read says GEO is just SEO with a new name.

Ido

I think that's half right, which is the annoying answer. The foundations carry over. If a crawler can't reach your pages, or your content is thin, or nobody trusts you, nothing else matters. But the thing being produced is different. Search gave you a list and the user did the synthesis. A generative engine does the synthesis itself, and your page is raw material. So the questions change. Not just, does my page rank, but, can my page be pulled apart, summarized, compared, and quoted without getting something wrong?

Ben cohen

Let's ground that in data, because Ahrefs published a study this month that I found really useful. They looked at 66.6 million ChatGPT responses, from the free logged out version, running GPT 5.6 Luna, plus 15 million of what's called fan out queries. Ido, explain fan out for people who haven't met the term.

Ido

Sure. When you ask a question that needs fresh information, the assistant doesn't just run your exact words through a search engine. It breaks your question into smaller searches, maybe a search for pricing, another for reviews, another for the official site. That breaking apart is the fan out. And those smaller searches are basically the real queries your page has to match.

Ben cohen

And here's what surprised me. Only 14.4% of those responses contained fan out queries at all, and the ones that did averaged about 1.6 queries. So most answers in that sample didn't search the web.

Ido

With a big caveat. That's free, logged out sessions, one model, August and September. Other studies in the same article's comparison found far more fan outs per prompt, but they used paid models or commercial prompts like best tool for X. So it's a snapshot, and OpenAI changes this constantly. What I take from it is simpler. When retrieval does happen, you might have one or two shots. If your page doesn't match one of those searches, the model never sees it.

Ben cohen

And where do those searches go? This is the part that matters for business. About 6.8% of the fan outs used the site: operator, meaning search inside one specific website. And among the most common of those, 75.9% pointed at brand sites. Apple, Ikea, Nike, that kind of thing.

Ido

That makes intuitive sense. If you want a price or a spec, you go to the source. And the same study tested about 20,000 US prompts. For factual questions about pricing, features, or company details, 84.4% cited at least one official source, and 63.4% of all citations were to official sites.

Ben cohen

But then comparisons flip. For things like A versus B, or which is cheaper, that drops. Only 46.4% of citations went to official sites, so most went elsewhere. Third party publishers, review sites, forums.

Ido

So you get two different jobs. Your own site has to be the place where the facts are unmistakable. Price, specs, who you serve, what you do. And then the rest of the web decides how you look when you're placed next to a competitor. You can't fully control the second part, but you can influence it.

Ben cohen

One more number, because it tells you what models are hunting for. The words price and review are showing up more in fan outs. Price went from 6.04% to 8.88% of fan outs in a month, which they describe as a relative rise of 47%. It's from a small base, so I wouldn't overread it, but it points to people asking AI to help them choose, not just to learn.

Ido

And the same study saw citations shift toward original sources and away from sites that mostly summarize other people's work. Again, a snapshot, and it could change with the next model update. But it fits the theme. Pages that only repeat what someone else published first are the easiest to replace.

Ben cohen

Let me push back on one thing, though. We keep saying brands have to be mentioned elsewhere. A Search Engine Land piece talked about GEO experiments where the sources that mention you may matter more than the content you control. Isn't that a little depressing for someone who just wants to fix their own website?

Ido

It's one article's reading of two experiments, so I'd hold it loosely. But I wouldn't find it depressing. It's the old idea of reputation, just with a new reader. Your site says who you are. Trusted places repeating it is what makes a model believe it. You need both. A site that's perfectly clear but unknown is weak, and a well known brand with a confusing site gets described wrong.

Ben cohen

So are we optimizing for the model, the crawler, or the user?

Ido

All three, and the trick is that the user still comes first. If you make the human experience worse to please a machine, you've lost the plot. But you can add a layer for the machine without taking anything away from the human. That's what the rest of this conversation is about.