Kevin Indig simply settled an argument that’s been operating in half-finished kind since AI Overviews launched. Does topical authority that rewards manufacturers for going deep on a topic as a substitute of extensive, truly carry over into AI search? Or is ChatGPT sufficient of a clean slate {that a} generalist model can stroll in and seize a class chilly?
Indig received entry to 6 months (January-June 2026) of ChatGPT solutions knowledge from Semrush, spanning 1,094 U.S. classes, 5 prompts per class, greater than 50,000 manufacturers, and over 600,000 citations. He revealed the breakdown in his Development Memo Substack on July 20, with the headline discovering on LinkedIn the identical day. His reply is that topical authority does matter in AI search, and the reason being sturdiness. As soon as a model earns an outsized share of mentions in a class, it tends to carry onto it.
That’s the tidy model. The messier and extra helpful model, buried a couple of paragraphs into his submit, is that just about not one of the market has settled but.
The Classes That Matter Most Are The Ones No one Owns
Indig constructed three buckets. A class has a transparent proprietor when one model reveals up in at the least 4 of 5 take a look at prompts and beats the runner-up by 5 share factors or extra. It’s an rising chief when a model leads at the least three prompts with out hitting that hole. The whole lot else, which means no model leads even three of 5 prompts, counts as unsettled.
Run these definitions throughout the complete dataset, and June 2026 seems to be extensive open. Solely 15.2% of classes had a transparent proprietor. Simply over half, 53.7%, have been open fields with a number of credible contenders and no model near locking the door.
Indig ranked all 1,094 classes by estimated AI search quantity and break up them into two even teams. The upper-volume half accounts for 98% of all AI search demand within the pattern, and it has the decrease proprietor fee of the 2, 11.3% versus 19% for the lower-volume half. Stack that up and 89.3% of estimated AI search demand sits inside classes with no clear proprietor but. The classes value essentially the most cash to occupy are, proper now, the least determined.
I’ve spent the previous three years watching SEOs debate whether or not topical authority is an actual algorithmic drive or a handy story businesses inform shoppers to justify a content material calendar. Indig’s knowledge is the very first thing I’ve seen that really resolves the query with an actual quantity. Topical authority is actual in AI search, and the manufacturers nonetheless treating class focus as non-obligatory are freely giving nine-figure classes to whoever bothers to indicate up persistently throughout all 5 immediate sorts.
House owners Are Onerous To Dethrone, However Solely When The Lead Is Large
The second half of Indig’s evaluation is the place the technique will get sharper. He tracked month-over-month management modifications throughout the identical classes and located {that a} clear proprietor held first place in 90.4% of comparisons.
The switches that did occur have been concentrated nearly solely in classes the place the lead was already skinny. Classes the place the chief modified had a median lead of simply 1.3 share factors going into the swap. Classes the place the chief held on had a median lead greater than double that, 2.9 factors. Development trajectory advised Indig nearly nothing by itself. He reveals three examples the place a number one model trended upward month over month and nonetheless received overtaken, as a result of a competitor merely grew sooner or closed a slim hole. The lesson isn’t that momentum is meaningless. It’s {that a} one- or two-point lead shouldn’t be a lead value defending but, and a model sitting in that vary ought to assume it’s nonetheless in a battle.
Indig additionally discovered one thing that undercuts a standard assumption about how possession will get constructed. Citations and model mentions turned out to be solely weakly associated, with a barely unfavorable correlation of -0.229. Essentially the most steadily cited area in a class was not often the identical as essentially the most steadily talked about model, matching up solely 20.8% of the time. However essentially the most talked about model was cited at the least as soon as 69.9% of the time. Getting cited rather a lot doesn’t mechanically make you the model ChatGPT reaches for by default. That’s a significant distinction for anybody whose GEO technique at the moment begins and ends with “get extra citations.”
House owners did share some traits versus their runner-up: increased branded search quantity in 55.7% of pairs, increased natural site visitors in 48.4%, and the next Semrush Authority Rating in 52.5%. Indig is cautious to name these correlations slightly than a formulation. A model doesn’t earn ChatGPT’s default reply by shopping for branded search quantity. It earns it by being the supply ChatGPT’s coaching and retrieval preserve touchdown on when the mannequin tries to reply all 5 variations of a query, the definition, the comparability, the alternate options listing, the use case, and the shopping for choice.
How To Use This Earlier than The Window Closes
Right here’s the place I’d level a practitioner’s subsequent quarter of labor, tailored from Indig’s framework and my very own Quotation Share of Voice strategy to monitoring AI mentions.
- Choose your battlefield first, then measure it correctly. Checklist the ten to twenty classes the place your model truly must be the default reply. For each, monitor the identical 5 immediate sorts Semrush used, a definition query, a comparability, an alternate options listing, a use case, and a shopping for query, not simply your model identify in isolation.
- Type by lead measurement, not by whether or not you’re at the moment “successful.” A class the place you lead by 1.5 factors shouldn’t be a class you’ve gained. Deal with something underneath roughly 3 factors as contested and put actual resourcing behind closing the remaining immediate gaps, since Indig’s knowledge reveals that’s the precise vary the place turnover occurs.
- Cease chasing citations as the entire technique. Use citations to earn your means into the reply set, then shift funds towards the particular content material sorts, comparability pages, proof factors, third-party protection, and clear positioning, that decide whether or not you’re the model truly named when you’re within the room.
The GEO dialog over the previous 12 months has fixated nearly solely on quotation counts as a result of citations are the best factor to drop right into a dashboard. Indig’s knowledge says citations are a door, not the room. Practitioners who preserve optimizing for the door whereas a competitor optimizes for being the identify ChatGPT truly says out loud are going to lose classes they thought they have been successful.
The window Indig describes gained’t keep open. Eighty-nine % of demand sitting in unsettled classes is an unusually beneficiant head begin, and the manufacturers that transfer on it within the subsequent few quarters, slightly than treating this as another report back to bookmark, are those that may nonetheless be the default reply when any individual runs this identical research in 2027.
Extra Assets:
Featured Picture: Roman Samborskyi/Shutterstock
