The Open Topic: Why 85% of ChatGPT Has No Winning Brand

Most marketplace sellers still measure their AI search presence one prompt at a time: does ChatGPT mention me when someone asks for my product? New research published this week suggests that is the wrong unit of measurement — and that the mistake is leaving the largest opportunity in AI search sitting open.

A study of 1,094 product and service categories, tracked in ChatGPT every month from January through June 2026, found that only 15.2% of topics had a single brand that clearly owned them. The other 85% — spanning more than 50,000 brands and over 600,000 citations — had no consistent winner across the questions a buyer actually asks in that category.

The practical conclusion is direct: the question in AI search is no longer “do I rank for this prompt?” It is “when the AI answers the full set of questions a buyer asks in my category, does my brand keep showing up?” For sellers, that shift changes where to invest effort — and it reveals that most categories are still winnable.

What “owning a topic” actually means

The study, published by Semrush in partnership with analyst Kevin Indig, defined a topic as a cluster of five prompts that mirror how a buyer researches a purchase: a definition question, a comparison question, an alternatives question, a use-case question, and a buying question. Think of a category like “best pillows for neck pain” or “small business banking,” and the five prompts as the standard set of questions someone asks while evaluating it.

A brand was counted as the category owner only when it met two conditions at once: it appeared in at least four of the five prompts, and it led the runner-up by at least five percentage points. That is deliberately strict. Winning a single prompt — even the buying prompt — does not make a brand the answer ChatGPT associates with the topic. Showing up consistently across the whole research journey does.

Three quarters of the categories did not even come close. Only 15.2% had a clear owner. Another 31.2% had an emerging leader — one brand mentioned in at least three prompts, but without the lead or the breadth to lock the topic down. The remaining 53.7% were fully unsettled, with no brand appearing reliably across the category’s prompts.

The biggest topics are the most wide open

Here is the part that should change a seller’s priority list. You would expect the most-asked topics — the ones buyers bring to ChatGPT most often — to be locked up by established brands, the way high-volume keywords are on Google. The data shows the opposite.

When the researchers split the 1,094 categories by AI search demand, only 11.3% of the top-half (most-asked) topics had a clear owner, compared with 19% of the bottom-half (less-asked) topics. That top half accounts for roughly 98% of the AI search volume in the sample. So almost every question buyers bring to ChatGPT lands in a topic where no brand has consolidated a lead.

The likely reason is not that big brands are absent. It is that the topics buyers ask about most are also the most fragmented — more competing mentions, more sources, and no single brand has yet built coverage consistent enough to win across all five question types. For a marketplace seller, that is an opening, not a wall: the categories with the most buyer demand are precisely the ones where a focused seller can still become the name the AI reaches for.

Strong SEO does not decide who wins a topic

Sellers with modest domain authority sometimes assume they cannot compete in AI search against bigger, more established brands. The data complicates that assumption.

The study compared each topic’s owner against its runner-up on three domain-level metrics — branded search volume, organic traffic, and Authority Score. The owner came out ahead only about half the time: higher branded search volume in 55.7% of pairs, higher organic traffic in 48.4%, and a higher Authority Score in 52.5%. Branded search volume was the only statistically significant differentiator, and even then the edge was modest. In nearly half the topics, the runner-up had at least one of these advantages and still lost.

The takeaway is not that traditional SEO is irrelevant. It is that broad SEO strength is not enough to explain who owns a topic. Once several brands are in the running, what separates the owner from the runner-up is depth of coverage inside the specific topic — a focused library of content, consistent mentions on the sources ChatGPT draws from, and clear entity signals tying the brand to the category. A seller with a smaller domain but deeper, more consistent topical coverage can take a topic a larger competitor has neglected.

What the AI says matters more than what it links to

One finding reorders how sellers should think about “visibility” in AI search. The study tracked two signals separately: which brands ChatGPT mentions in its answer text, and which sources it cites as links underneath. They barely overlap. Only 21% of the most-cited domains in a category were also the most-mentioned brand, and the two signals correlated slightly negatively (–0.229).

This matters because buyers act on what the AI says, not on the citation list. In a separate usability study of 185 documented purchase tasks, 74% of AI Mode users’ final shortlists came directly from the shortlist the model itself built — buyers accepted the AI’s picks rather than constructing their own, the way classic Google search users do.

For a seller, the implication is sharp. Chasing source citations — getting your URL listed under an answer — is worth less than being the brand the AI names in the answer itself. The mention is what the buyer reads, remembers, and acts on.

Owners keep their lead — but only when the lead is wide

Once a brand does own a topic, it tends to hold on to it. Clear category owners stayed in first place in 90.4% of month-over-month comparisons. But that stickiness depends almost entirely on the size of the lead.

Where the leading brand held its position, the median lead was 2.9 percentage points. Where the lead flipped to the runner-up, the median lead had been just 1.3 points. Across the dataset, the top brand switched in 1,950 of 5,470 monthly comparisons — and those switches clustered in categories where the original lead was small.

The study is careful not to over-claim why leaders flip. It could reflect changes in ChatGPT’s sources, changes in how the model generates answers, or genuine shifts in a brand’s relevance. What the data does show is a simpler, actionable pattern: margin matters more than momentum. A seller who leads a topic narrowly is not safe; a seller who consolidates a wider lead becomes very hard to displace. And a challenger entering a topic where the current leader’s lead is thin is entering at exactly the right moment.

How a seller actually wins a topic

The research points to a concrete operating method, and it does not require outspending anyone.

Pick two or three topics, not fifty. Choose categories where you already sell, where you have some content or signal, and where buyers ask the five standard questions. Trying to build authority across every topic dilutes everything; concentrating on a few is what turns a runner-up into an owner.

Cover all five question types. Ownership requires showing up across the definition, comparison, alternatives, use-case, and buying prompts — not just the buying one. For each topic, make sure your listing, product page, and supporting content answer each of those questions explicitly and in self-contained passages the model can extract.

Earn mentions, not just links. Because buyers act on mentions, the work that moves the needle is third-party coverage on the sources ChatGPT draws from — reviews, roundups, Reddit, industry publications, expert commentary. A brand named across those sources in the context of a specific topic is the brand the AI reaches for when a new question in that topic arrives.

Lock the entity signals. Consistent brand naming, structured data, and third-party citations that tie your brand to the topic are what let the model connect your listing to the questions being asked. An anonymized FirstShelf audit of marketplace listings over the past 90 days found structure quality averaging just 33.7 out of 100 and semantic density at 43.1 — the exact dimensions that feed topical authority. Most sellers are not yet giving the model the consistent, structured signal it needs to place them across a topic.

Defend a wide lead; attack a narrow one. If you already lead a topic, keep building coverage until your margin is comfortable. If a competitor leads yours by a point or two, this is the window to out-cover them before the topic consolidates.

How FirstShelf can help

FirstShelf reads your marketplace listing the same way a generative engine does — scoring semantic density, structure quality, entity authority, and platform compliance, the same dimensions that decide whether a brand appears consistently across a topic’s prompts. The audit shows you exactly which entity signals are missing and which claims lack the third-party corroboration a model needs to mention you with confidence.

Use it to pick the two or three topics where you can realistically consolidate coverage, find the structured-signal gaps that are keeping you off the AI’s shortlist, and track your scores as you close them. The same engine that produced the audit figures in this article is the one that scores your listings.