89% of AI search categories have no owner yet

Kevin Indig got six months of ChatGPT answer data from Semrush — 1,094 US categories, 50,000+ brands, 600,000+ citations — and published the breakdown last week. The headline finding, the one that went round LinkedIn, was that topical authority matters in AI search and category leaders are hard to dethrone once they're established.

That's the tidy version. The version worth your attention is a number a few paragraphs deeper: 89.3% of estimated AI search demand sits inside categories with no clear owner. Not one brand showing up in four of five test prompts. Not one brand five points ahead of the runner-up. Just an open field with credible contenders and nobody close to locking the door.

The categories worth the most money to occupy are, right now, the least decided. And most UK businesses I speak to are still deciding whether GEO is a real thing.

What "no owner" actually means

Indig's definitions are worth understanding before the take lands. A category has a "clear owner" when one brand appears in at least four of five prompts and beats second place by five percentage points or more. An "emerging leader" leads three of five prompts without hitting that gap. Everything else — no brand leading even three of five — counts as unsettled.

In June 2026, 15.2% of categories had a clear owner. 31.1% had an emerging leader. 53.7% were open fields. Then Indig split all 1,094 categories by estimated AI search volume into two even halves. The higher-volume half — the categories with real money behind them — had a lower owner rate. 11.3% versus 19% for the lower-volume half.

Stack that up and the big-money categories are the least contested ones. The stuff nobody serious wants to lose is the stuff nobody has bothered to win yet.

This is a window, not a permanent state

The second half of Indig's analysis is where the strategic urgency comes from. He tracked month-over-month leadership changes and found clear owners held first place 90.4% of the time. Once someone locks a category with a wide enough lead, they mostly stay locked.

Ownership, once established, is durable. Which means the window to become the owner is finite.

Ownership, once established, is durable. Which means the window to become the owner is finite.

The switches that did happen clustered in categories where the lead was already thin — median 1.3 points versus 2.9 points for categories where the leader held on. Momentum on its own didn't rescue narrow leaders. A one-point lead is not a lead worth defending. Somewhere between two and three points is where things start to calcify.

Which means we're in a specific phase. Most categories are still contested. But the categories that do get won are getting won for good, or something close to it. The economic value of showing up consistently across a prompt set right now is genuinely enormous, and it declines every month someone else keeps showing up and you don't.

The citations versus mentions thing

One finding buried in Indig's data undercuts almost every GEO tool pitch I've seen this year. Citations and brand mentions have a correlation of -0.229. Slightly negative. The most-cited domain in a category matches the most-mentioned brand only 20.8% of the time.

I've written about this before in the recognition versus mention gap — but Indig's data adds a wrinkle. It's not just that AI can describe brands it doesn't recommend. It's that the brands it recommends often aren't the sites it cites. Which means the entire GEO-tooling industry, which is largely built on citation tracking, is measuring a variable that only weakly relates to the outcome that actually matters.

If you sell software or services, the question isn't whether your domain got cited in a Perplexity answer. The question is whether ChatGPT named your brand when someone asked for a shortlist. Those are different problems with different solutions, and the industry has quietly conflated them.

What this means if you're a UK business owner reading this

The instinct when reading a piece like this is to take it as motivation to "do GEO." That's not quite the right read. Here's the more useful version.

First, if you operate in a category — genuinely a category, not a keyword — where you can name your top three competitors without thinking, ask an honest question. When someone prompts ChatGPT for recommendations in your space, does your brand appear? Not your URL in the citations. Your brand name in the recommendation list. Test five prompts a customer might actually use. If you're in the majority of that 89%, nobody owns your category yet. Not you, not the incumbent you assume dominates it.

Second, the mechanism for becoming the recommended brand isn't a GEO tactics list. It's the boring stuff — being mentioned frequently, in context, by credible third parties, with a brand name distinctive enough that a model doesn't confuse you with a competitor. This is what the measurement collapse in AI search piece was getting at from a different angle. The optimisation surface has moved from documents to reputation.

Third — and this is the one that will annoy people — the categories most likely to still be open are the boring ones. B2B services. Regional trades. Industry-specific software. The categories that got obsessive coverage from consumer-facing SEO blogs are the ones most likely to have an emerging leader already. The categories your competitors have quietly ignored because they never seemed sexy enough to write about are the ones sitting genuinely open.

The honest limits

A few caveats worth naming, because Indig's paper doesn't cover them and neither does most of the coverage.

The dataset is US ChatGPT. The UK skews will be different — different citation patterns, different brand recall, different regional context. There's no equivalent dataset for the UK yet that I'm aware of. So the 89% figure is directional here, not literal.

The definition of "category" is Semrush's. Some of those 1,094 categories are narrower than a UK small business would find useful, and some are broader. Testing your own five prompts across your own competitor set is more informative than trying to map yourself onto their taxonomy.

And ChatGPT is one engine. Gemini, Perplexity, and Copilot behave differently, and the Perplexity citation patterns diverge sharply from ChatGPT. Owning ChatGPT in your category is not the same as owning AI search. It's a start.

The close

The industry story about AI search for the past eighteen months has been "everything is changing, nobody knows anything, wait and see." That was true for a while. Indig's data suggests it's stopping being true.

Categories are being won right now. Most of them are still open. The brands that will own the recommendation layer in 2028 are, in most cases, the brands showing up consistently in 2026 while everyone else is still debating whether GEO is a real acronym.

That's the loop. And it's closing.