New research shows being cited by AI and being recommended by AI are almost two different games, and most businesses are only winning one of them.
Ask ChatGPT what your business does and there's a decent chance it gets the basics right. Ask it who it would actually recommend for the thing you do, and there's a much better chance your name doesn't come up at all. Those feel like the same question. New research says they're not, and the gap between them is where most businesses are quietly losing.
Semrush and Growth Memo just published a study that tracked 1,094 categories inside ChatGPT for six months, from January to June 2026, using five real buyer questions per category, things like "best payroll software" or "small business banking." They logged over 283,000 instances of a domain being cited as a source, and over 76,000 instances of a brand actually being named and recommended. Then they checked how much those two things overlap.
Barely at all. Only 21 percent of the domains that got cited the most in a category were also the brand ChatGPT recommended most in that same category. The two numbers correlate slightly negatively. Being the source ChatGPT quietly pulls a fact from, and being the name it puts in front of a buyer, are statistically two different achievements.
The study found only 15.2 percent of the 1,094 categories had a clear owner, meaning a brand that consistently gets named as the recommendation rather than just referenced as a footnote. In every other category, no single business had built enough repeat authority to be the default answer. There's a citation for almost everyone. There's a recommendation for almost no one.
That distinction matters more once you look at what a citation actually is inside these tools. It's a source link, buried in a response most people never expand. A recommendation is the brand name in the sentence the user actually reads. If you've ever checked whether your business shows up in AI search and felt reassured because your website appeared somewhere in the sources, this research is the reason that reassurance might be premature. Appearing in the reading list and being the name someone remembers are not the same outcome, and only one of them turns into a phone call.
The researchers also found citation rates barely move based on how relevant a category is to a brand's actual expertise. A business gets cited at close to the same rate in a topic it's genuinely known for as in a topic it's only tangentially connected to. Named, remembered recommendations don't work that way. Those cluster tightly around the handful of categories where a business has built real, repeated depth, not the long list of things it could technically also do.
This research isn't happening in isolation. It comes the same fortnight that fresh click data confirmed what a lot of business owners already suspected: when Google shows an AI Overview instead of a normal results page, clicks to the top organic result drop by more than half, in some measurements as much as 58 to 61 percent. Fewer people are clicking through to compare five businesses themselves. More of them are taking whatever name the AI puts in front of them and stopping there.
Put those two findings together and the maths gets uncomfortable fast. Fewer clicks are happening at all, and the ones that remain are going overwhelmingly to whichever brand the AI decided to actually recommend, not whichever brand it happened to cite. Being technically visible inside an AI answer no longer guarantees you get considered. It just means you were in the room.
A citation proves the AI has read about you. A recommendation proves it trusts you. Most businesses have only earned the first one.
Every week another business owner tells me they want to "show up in ChatGPT" or "get cited by AI search," as if that's a single switch to flip. Usually what follows is a scramble to get mentioned anywhere possible: a press release here, a directory listing there, a blog post touching on ten different services because more topics surely means more chances of appearing. This research says that instinct is exactly backwards. Spreading a business thin across many topics is how you end up in the roughly 85 percent of categories with no clear owner, cited occasionally, recommended never.
I'm not here to tell business owners AI visibility doesn't matter. It clearly does, and it's only going to matter more as fewer people click past the answer they're given. What I am here to say is this: the businesses winning this aren't the ones scattered across the widest number of topics. They're the ones who picked two or three categories they genuinely dominate and built enough repeated, specific, evidenced proof in exactly those categories that an AI model has no honest alternative but to keep naming them.
Start by being honest about which two or three categories your business could plausibly own, not the twenty things you're technically capable of doing. Then look at whether your website, case studies, reviews, and outside mentions actually build a repeated, specific case for exactly those categories, or whether they're spread evenly across everything you offer. If a competitor with a narrower focus has more depth in the one category that actually drives your revenue, that's the gap closing your pipeline right now, quietly, one AI answer at a time.
This isn't a content volume problem. It's a focus problem. AI scales what's already there, and right now, for most businesses, what's there is too broad to be recommended and too thin to be remembered.
Nothing changes if nothing changes.