Right now, 89% of AI search demand sits in categories nobody owns. That is a fact with a shelf life.
Yes, topical authority still matters in AI search, but it works differently to the version we all learned for Google. New Semrush data suggests it operates at the level of the category rather than the page: brands earn an outsized share of mentions in a topic, and once they have it, they usually keep it. The question for a marketing team is no longer how much you publish. It is which handful of questions you are willing to be the answer to.
The research comes from Kevin Indig, working with a dataset from the Semrush AI Visibility Toolkit. Semrush published the study on 20 July 2026 and Indig's analysis followed on Search Engine Land on 22 July. The scope: 1,094 US categories, five prompts per category, monthly snapshots from January to June 2026, ChatGPT only. It covers more than 220,000 domains, 50,000 brands and 600,000 citations.
Two things are worth naming before we go further. Search Engine Land, where the analysis was published, is owned by Semrush, and the data is Semrush's own. And the study measures one narrow thing: whether a brand is named in an answer. It does not measure sentiment, recommendation quality, trust or whether anyone bought anything. A mention can be positive, neutral or actively unflattering, and the export cannot tell the difference.
That is a real limitation. It is also, we think, still the most useful public dataset on this question so far, because the alternative most teams are working from is anecdote.
The study defines an owner as the brand with the highest share of brand mentions, named in at least four of a category's five prompts, and at least five percentage points clear of the runner-up. On that definition, only 15.2% of categories had a clear owner in June 2026. Another 53.7% were open fields with several contenders and no settled leader.
The interesting number is what happens next. A clear owner held first place in 90.4% of month-to-month comparisons. That is the finding that makes topical focus worth the constraint. In classic search, a good page can be outranked next quarter by a better page. In this dataset, category ownership behaves less like a ranking and more like a position: slow to earn, slow to lose.
If that holds, the compounding logic of topic focus survives the shift to AI answers intact. What changes is the unit. You are not building authority for a page. You are building it for a question.
In the categories you would assume were already sewn up. This is the counterintuitive part.
Splitting all 1,094 categories into two halves by AI search volume, the top half holds 98% of total demand in the sample. It also has the lowest owner rate: 11.3%, against 19.0% in the smaller-volume half. Stack those together and 89.3% of estimated AI search demand sits in categories with no clear owner.
Demand has concentrated. Ownership has not. For a B2B SaaS company that has spent three years being told the big terms are unwinnable, that is a genuinely different picture, and it is the single most actionable thing in the study. The categories where buyers are actually asking questions are, on this evidence, the ones least likely to have a brand already installed as the answer.
The obvious caveat: this is a June 2026 snapshot of a market roughly eighteen months old. Open fields close.
Because they are not the same thing, and the gap between them is wider than most reporting suggests. In the study, the relationship between citations and brand mentions was slightly negative, at -0.229. The most-cited domain in a category was also the most-mentioned brand only 20.8% of the time. The most-mentioned brand had at least one citation 69.9% of the time.
So citations are close to a prerequisite and a long way from a proxy. You can be the source ChatGPT reads and not the brand it names, which is the least satisfying outcome in marketing.
Why it matters commercially: earlier Growth Memo consumer research found that in AI Mode, 64% of participants clicked nothing at all, and 74% chose whichever option the AI named first. Being named is the win. Being footnoted is not.
We have written separately about the mechanics of earning those citations in How to Get Cited by ChatGPT: A B2B Marketer's Playbook for AI Search Visibility. This piece is about choosing where to point that work.
Sometimes, and the study is unusually specific about when. Leaders changed in 1,950 of 5,470 month-to-month comparisons, so movement is common. But those switches clustered in categories where the lead was thin. Brands that lost first place had held a median lead of 1.3 percentage points. Brands that held on had 2.9.
The practical read: a narrow lead is not a position, it is a coin toss with extra steps. Wide, sustained ownership is much harder to shift.
That cuts both ways for a challenger brand. Attacking a category where one competitor appears in every single prompt is expensive and slow. Attacking one where the leader is barely ahead, or where nobody has settled, is a completely different proposition on the same budget. The study cannot tell us why individual switches happened, and the authors are careful to say so, but the condition for disruption is clear enough to plan around.
Pick five categories. Not fifteen, and not every topic that is semi-adjacent to what you sell.
Start by writing down 10 to 20 categories that genuinely matter commercially, and for each one, five prompts a real buyer would type: a definition, a comparison, an alternatives question, a use case, and a buying question. Run them. Note where you appear and where a competitor does instead.
Then sort them. Categories with no clear leader, or a leader you are within a couple of points of, are where the money goes. Categories you already lead get defended, which mostly means keeping the material current rather than adding to it. Categories where one brand appears in every prompt get skipped, or attacked much later with something genuinely differentiated.
The hard part is not the method. It is the deleting. Five categories means saying no to the twelve other things someone in your business will want a blog post about this quarter, and the discipline of that refusal is most of what separates a brand that owns a category from one that appears in it occasionally. For the wider picture of what AI is changing across B2B marketing, see AI in Marketing: A B2B SaaS Marketer's Field Guide.
Worth holding lightly. This is one vendor's data, on one model, in one country, over six months, measuring mentions rather than outcomes. Ownership might turn out to be a feature of an immature market rather than a durable property. The 90.4% retention figure could look very different after a model update that changes how answers are assembled.
The study is also honest about what it cannot show: comparing owners to runners-up, owners had higher branded search volume in 55.7% of pairs, higher organic traffic in 48.4% and higher authority scores in 52.5%. Those are clues, not causes. Nobody has demonstrated what actually creates ownership.
What we would say is that the direction of the bet has not changed much. Concentrating on a defined set of questions was the right call when it was Google deciding, and it looks like the right call now. The cost of getting your five categories wrong is simply lower this year than it will be next year.
Does topical authority still matter in AI search?
Yes, but at the level of the category rather than the page. In the Semrush and Growth Memo study of 1,094 ChatGPT categories, brands with a clear lead in share of mentions across a topic kept first place in 90.4% of month-to-month comparisons. Authority now attaches to a set of related buyer questions, not to a single ranking URL.
What counts as owning a category in ChatGPT?
The study defines an owner as the brand with the highest share of brand mentions, named in at least four of a category's five prompts, and at least five percentage points clear of the runner-up. Only 15.2% of categories met that bar in June 2026, and 53.7% had no settled leader at all.
Why is 89% of AI search demand described as unowned?
Because demand and ownership have not moved together. The top half of categories by AI search volume held 98% of demand in the sample but had the lowest owner rate, at 11.3%. Combined, 89.3% of estimated AI search demand sat in categories with no clear owner.
Are AI citations and brand mentions the same thing?
No. The correlation between the two was slightly negative at -0.229, and the most-cited domain in a category was also the most-mentioned brand only 20.8% of the time. Citations are close to a prerequisite for being named, but being cited does not mean the AI names your brand in its answer.
Can a smaller brand take a category from an established one?
Sometimes, and mainly where the lead is thin. Brands that lost first place had held a median lead of 1.3 percentage points; those that kept it had 2.9. A category where one brand appears in every prompt is expensive to attack, while an unsettled category or a narrow leader is winnable on the same budget.
How should a B2B SaaS team act on this?
Pick five categories. List 10 to 20 commercially important topics, write five buyer prompts for each (definition, comparison, alternatives, use case, buying question), run them, and note where you and your competitors appear. Invest in categories with no clear leader, defend the ones you already lead, and skip those where one brand appears in every prompt.
SaaStrix is where B2B marketers become agentic marketing leaders, and you can try it free for five days.
.svg){kind=icon}