AI Search Categories Don't Change Hands as Often as You Think

I keep running into a version of the same argument, usually from someone smart, usually delivered with real confidence: don't over-invest in AI search category ownership, because the whole market is too volatile to build a durable strategy on. ChatGPT's share falls from 78% to 56% in six months. Sensor Tower has it under 50% entirely. Perplexity fell off a cliff and Copilot practically vanished. Why chase a leaderboard that reshuffles every quarter? It's a reasonable-sounding argument. It's also measuring the wrong layer, and Kevin Indig, of all people, just published the data that shows why.

The 'why bother, it's all volatile' argument

It's worth stating the volatility case fairly, because it's not made up. Kevin Indig's own H1 2026 Halftime Report tracked ChatGPT's AI-search usage share falling from 78% in July 2025 to 56% by July 2026, with Gemini climbing to 30% and Claude to 10% over the same window. Sensor Tower's separately conducted 'State of AI 2026' report, covered by Fast Company in June, measured a different thing, 'true audience' usage, and found ChatGPT had already dropped below 50%, to 46%, ahead of Gemini at 28% and Claude at 10%. Previsible's referral-traffic research, which we covered at the end of July, found Perplexity down 61% from its peak and Copilot down a brutal 96% from its August 2025 high. Stack those next to each other and the story writes itself: nothing in this market holds still long enough to plan around.

If that were the whole picture, the 'don't over-invest' argument would be hard to argue with. Spending a year building comparison content and third-party corroboration to win visibility inside a specific engine only to have that engine's overall relevance collapse out from under you would be a bad trade. That's a legitimate risk, and it's the strongest version of the skeptical case.

Kevin Indig's other number: AI search category ownership

Here's what the volatility argument skips. Indig didn't just publish the halftime report on engine-level share. Two weeks earlier, on July 20, he published a separate, much larger study asking a completely different question: once a brand becomes the clearly established source for a specific topic inside AI search, does it tend to hold that position, or does it churn just as fast as the engine-level numbers suggest everything churns?

The study covered more than 220,000 domains and 50,000 brands, tracked across 1,094 US categories with five prompts per category, in monthly snapshots from January through June 2026, more than 600,000 citations total. As of June, only 15.2% of categories had a clear, established owner; 53.7% were still open fields with multiple credible contenders and no settled leader. That part supports the 89%-plus-of-demand-is-unowned framing we've written about before. But the number that matters for this argument is the one about what happens once a category does get an owner: that owner retained first place in 90.4% of month-over-month comparisons. Not 60%. Not a coin flip. Nine times out of ten, whoever owns a topic this month still owns it next month.

Two different things, both true at once

This isn't a contradiction in Indig's own data, and it isn't a case of picking whichever number supports the argument you already wanted to make. It's two different measurements of two genuinely different layers of the same system, and conflating them is the actual mistake, not either number on its own.

“The leaderboard of engines reshuffles constantly. The leaderboard of who owns a specific topic, once it settles, barely moves at all.”

Think about why this makes structural sense, not just statistical sense. Engine-level share moves on product decisions, ad rollouts, model releases, and user habit, forces entirely outside any individual brand's control, which is exactly why it's so volatile. Category ownership moves on a much slower, stickier set of signals: accumulated corroboration across the sources engines already trust, a track record of accurate citation, and enough historical presence that an engine's retrieval defaults toward the source it's already found reliable before. Those signals don't reset every time a new model ships. They compound.

The debate this actually settles

There's a real disagreement happening in the industry right now between people who think the volatility numbers mean GEO is a bad long-term bet, and people who think category ownership data means engine churn barely matters. Both camps are working from real numbers and drawing the wrong conclusion from them individually. The volatility camp is right that no one should build a strategy that depends on ChatGPT staying at 78%, or any single engine holding a specific share, because that number has already proven it won't sit still. The durability camp is right that once you're the established answer for a topic, you're not likely to get dislodged by the next model update. The mistake on both sides is treating these as competing claims about the same thing, when they're actually claims about two different layers that happen to live inside the same three letters, GEO.

Put them together and the actual strategy is neither 'don't bother, it's too volatile' nor 'relax, ownership is permanent.' It's: track engine-level share to decide where to point effort this quarter, and treat category ownership, once you've built it, as the durable asset that survives whichever engine happens to be winning the usage war at the time. A brand that owns its topic on ChatGPT specifically doesn't necessarily transfer that ownership automatically if Gemini becomes the dominant engine tomorrow, but the underlying signals, corroboration, accuracy, established presence, are the same signals that win ownership on whichever engine comes next. That's the part of the 90.4% finding worth internalizing: it's not luck, it's a real, buildable moat, even if it isn't automatically portable to a new leader the moment the leaderboard changes.

What actually determines AI search category ownership

This is where I'd push back on treating 90.4% as a free pass to relax, too. That retention rate applies to categories that already have an owner. It says nothing about how hard it is to become the owner in the first place, and Indig's other number, 53.7% of categories still wide open, echoes what we found when 89.3% of AI-search demand turned out to be unowned: it's the more urgent one for anyone not already established. An open category is a genuine opportunity precisely because whoever locks it down first is statistically likely to hold it for a long time afterward. That's not a reason to relax. It's a reason to move now, before someone else becomes the 90.4%.

What to do with this

Stop reading engine-level market share swings as a verdict on whether generative engine optimization investment is worth it. They're real, they're worth tracking, and they should absolutely inform which platforms get the most attention in a given quarter. But they're a different question from whether category ownership, once won, holds up, and the honest answer to that second question is: yes, most of the time, decisively. If your category is one of the 53.7% still up for grabs, the volatility everyone's worried about isn't really your problem yet. Not showing up before someone else claims it, and locks in a spot they're statistically likely to keep, is.

You're not behind because the market share numbers look chaotic this month. You're behind if you're still waiting for the numbers to settle down before you start building the thing that determines whether you own your category once they do. They're not going to settle down. Indig's own data says the volatile part and the durable part are both permanent features of how this works now. Plan around both, and stop treating one as an excuse to ignore the other.

See where you are cited today

A free snapshot audit of your rankings and AI citations before we ever talk.