The window is still open: why 89 percent of AI search demand belongs to nobody yet

There is one number that shows how early the phase is that brands find themselves in when it comes to AI answers: 89.3 percent of estimated AI search demand currently sits in categories where no brand holds a clear lead (Kevin Indig / Search Engine Journal). Put differently: in the vast majority of topics people ask ChatGPT and its peers about, the position of "default recommendation" is still vacant, waiting for someone to take it. We tested that claim against our own real customer data, with a surprisingly clear result. More on that below.

How the study measures who "owns" a category

For his Growth Memo Substack, Kevin Indig analysed six months of ChatGPT answer data from Semrush (January to June 2026): 1,094 US categories, five standard prompts each (definition, comparison, alternatives, use case, purchase intent), more than 50,000 brands and over 600,000 citations (Search Engine Journal).

The definitions are deliberately strict:

• Clear owner: a brand appears in at least 4 of 5 prompts and leads the runner-up by at least 5 percentage points.

• Emerging leader: a brand leads in at least 3 prompts but does not reach that margin.

• Unsettled: everything else, several brands competing without one clearly ahead.

The numbers in detail

• Only 15.2 percent of the categories examined already have a clear owner. 53.7 percent are open fields with several credible candidates.

• The imbalance grows once you weight by volume: the higher-volume half of the categories accounts for 98 percent of all AI search demand, yet has the lowest ownership rate (11.3 percent against 19 percent in the lower-volume half). Exactly where the most is at stake, the least has been decided.

• Whoever takes the lead usually keeps it: 90.4 percent of clear owners defended their position over the observation period. Where the lead did change hands, the previous margin was a median of just 1.3 percentage points; where the leader held, the median margin was 2.9 points. The line between "still open" and "as good as settled" therefore sits at roughly 3 points.

• And topical authority is genuinely measurable. Owners more often had higher branded search volume (55.7 percent of pairs), higher organic traffic (48.4 percent) and a higher Semrush authority score (52.5 percent) than the respective runner-up.

Being cited is not the same as being recommended

Perhaps the single most important finding of the study: citation and brand recommendation barely correlate (-0.229, slightly negative). In only 20.8 percent of cases was the most-cited domain also the most-named brand. Even the most-named brand was cited at all in just 69.9 percent of cases. Indig's conclusion: "Citations are a door, not the room." Being cited gets you into the answer; it does not hand you the recommendation.

The practical test: what our own customer data shows

We did not want to take this gap between citation and recommendation on trust from someone else's study, so we checked it against our own data. An important caveat first: our sample is considerably smaller than Indig's. Five anonymised customer accounts across five sectors (consulting, B2B industrial, e-commerce, SaaS, automotive), 270 individual measurement points (a combination of category, prompt type, date and AI model) totalling 6,592 individual runs across ChatGPT, Claude, Gemini and Perplexity, collected between 13 June and 30 July 2026.

That is a small addition, not a replacement for the larger study. But in our data the direction of the finding is even more pronounced:

• Across all measurement points, the average named rate (the brand is actually recommended) was just 17.5 percent, while the average cited rate (the brand domain is referenced as a source) was 63.3 percent. A gap of almost 46 percentage points between "used as a source" and "actually recommended".

• For 4 of the 5 brands examined, the cited rate was well above the named rate (gaps between 30 and 95 points depending on category). Only one brand showed the reverse, healthier picture.

• Within our data, "gets cited" does correlate slightly positively with "gets recommended" (correlation 0.31): where a brand was cited in an answer, the average recommendation rate was 23.2 percent, against just 7.7 percent without a citation. But even with a citation, a brand still goes unrecommended in three out of four cases.

The finding matches Indig's core statement, even though the exact metric is not directly comparable in method: Indig compares the most-cited domain against the most-named brand across many competitors, while our figure measures, per brand, whether an individual answer containing a citation also contains a recommendation. Both angles arrive at the same conclusion: being cited is the ticket in, not the win.

What to do about it now

Taking stock helps nobody on its own. The study already offers three clear levers that can be applied immediately:

1. Choose the battlefield, then measure it properly. Do not try to occupy every category at once: identify 10 to 20 categories in which your brand has to become the default answer, and track the same five prompt types there consistently (definition, comparison, alternatives, use case, purchase intent). The Klariton Free Check gives you a first reading in 60 seconds of how well your own site even qualifies as a source for those prompt types.

2. Sort by size of lead, not by "currently winning". A lead of 1.5 points is not a victory; anything under roughly 3 points counts as contested and deserves deliberate defence or attack. The gap between "is cited" and "is recommended" is precisely the point of attack: continuous prompt monitoring in the Klariton Studio app shows whether a category is genuinely tipping towards ownership, or whether only the citation rate is climbing while the recommendation rate stands still.

3. Do not optimise for citations alone. Citations open the door, but stopping there loses the category in the end to competitors who also invest in concrete content formats: comparison pages, proof points, third-party editorial mentions, clear positioning. Structured, technically clean product data is the foundation for all of it. Background in the Klariton magazine articles „How shops make their products readable for AI: structured data, step by step" and „Why most AI crawlers cannot read your JavaScript content".

Conclusion

The window is real, but it does not slam shut. It closes category by category, as soon as one brand reaches the roughly 3-point lead that holds in 90 percent of cases according to Indig's data. Whoever starts now to distinguish rigorously between citation and actual recommendation, and to steer where the gap is widest, has a genuine head start in the 89 percent of categories still open. It is shrinking, but it is still there.

Sources: Search Engine Journal, Kevin Indig, "89% Of AI Search Demand Has No Clear Owner" · Our own anonymised analysis of Klariton customer data (5 accounts, 270 measurement points, 6,592 individual runs, 13 June to 30 July 2026)