Semrush: only 15.2% of ChatGPT categories have a clear brand owner — and the number falls where the demand is

What the study measured

The primary source is Semrush's ChatGPT topic-authority study, published 20 July 2026 by Margarita Loktionova, and written up by Search Engine Land on 21 July. Semrush built five representative prompts per category — definition, comparison, alternatives, use case, and buying question — and recorded, month by month from January to June 2026, which brands ChatGPT mentioned in the answer text and which domains it cited as sources. The disclosure we make on every vendor study applies here too: Semrush sells an AI-visibility product, and so do we. The measurements are the durable part; discount the conclusions.

The classification is strict and worth reading carefully, because the headline number depends on it. A category owner is the most-mentioned brand appearing in at least 4 of the 5 prompts with at least a 5-percentage-point lead over the runner-up. An emerging leader appears in at least 3 prompts but misses those margins. An unsettled category has no brand in 3 prompts at all.

Ownership, once won, is sticky — but the margin decides

Month over month, category owners kept first place 90.4% of the time. The exceptions have a signature: in categories where the leader flipped, the median lead was 1.3 percentage points; where the leader held, it was 2.9. So a narrow lead is not a small version of ownership — it is a different, unstable state. If you are tracking your own position in an AI answer set, the size of the gap is the number that tells you whether it will survive the month, not the rank.

No classic SEO metric predicted it

Semrush correlated ChatGPT category ownership against the usual SEO signals. Branded search volume was the strongest at 55.7%, then Authority Score at 52.5%, then organic traffic at 48.4%. Every one of those leaves roughly half the variance unexplained — meaning your rankings tell you something about whether ChatGPT names you, but nowhere near enough to substitute for measuring it. The sharpest number in the study is the gap between the two things people conflate: only 21% of the most-cited domains in a category were also that category's most-mentioned brand. Being the source an engine reads and being the name it recommends are close to independent outcomes.

Our own data: at the local level, the field is not open

"53.7% unsettled" is a finding about broad national categories. We measure something one level down — buyer-intent questions naming a specific city — and there the picture inverts. In the 15 local markets where we have audited three or more competing businesses (59 businesses, 343 ChatGPT and Perplexity answers, 12 June to 10 July 2026), every single market had a hard core of businesses named in every audit of that market — a mean of 4.7 names, taking 34.6% of all rival mentions, while 62% of the 598 distinct rivals we recorded appeared in only one audit. Locally, ownership is not up for grabs; it is already assigned, and the work is dislodging an incumbent set.

The two findings are consistent, not contradictory. A broad category ("project management software") aggregates thousands of intents and thousands of brands, so no one name dominates 4 of 5 prompts. "Best med spa in Scottsdale, AZ" is a single intent with a shortlist of maybe six credible answers, and the engines converge on it. If you sell to local businesses — or run an agency that does — the national statistic is the wrong prior to bring to the pitch.

What to do with this

• Measure at the level you sell at. A national-category ownership figure says nothing about whether ChatGPT names you in your metro. The only way to know is to ask the buyer's question, with the city in it, and record the answer.
• Track mentions and citations separately. With only 21% overlap between most-cited domain and most-mentioned brand, a tool that reports one and calls it "AI visibility" is reporting half the outcome.
• Watch the margin, not the rank. A 1.3-point lead flipped; a 2.9-point lead held. Report the gap to the runner-up, not the position.
• Don't stop at one engine. In our corpus, ChatGPT and Perplexity answering the identical question overlap on a mean of 10.1% of the businesses they name. Ownership measured on one engine is not ownership.

If you want the local version of this number for a specific business, that is exactly what our free AI visibility check runs: the buyer's question, put to multiple engines, with the names they return recorded verbatim.