Kevin Indig conscionable settled an statement that’s been moving successful half-finished shape since AI Overviews launched. Does topical authority that rewards brands for going heavy connected a taxable alternatively of wide, really transportation complete into AI search? Or is ChatGPT capable of a blank slate that a generalist marque tin locomotion successful and drawback a class cold?
Indig sewage entree to six months (January-June 2026) of ChatGPT answers information from Semrush, spanning 1,094 U.S. categories, 5 prompts per category, much than 50,000 brands, and complete 600,000 citations. He published the breakdown successful his Growth Memo Substack connected July 20, pinch the header uncovering connected LinkedIn the aforesaid day. His reply is that topical authority does matter successful AI search, and the logic is durability. Once a marque earns an outsized stock of mentions successful a category, it tends to clasp onto it.
Screenshot from LinkedIn, July 2026That’s the tidy version. The messier and much useful version, buried a fewer paragraphs into his post, is that almost nary of the marketplace has settled yet.
The Categories That Matter Most Are The Ones Nobody Owns
Indig built 3 buckets. A class has a clear proprietor erstwhile 1 marque shows up successful astatine slightest 4 of 5 trial prompts and thumps the runner-up by 5 percent points aliases more. It’s an emerging leader erstwhile a marque leads astatine slightest 3 prompts without hitting that gap. Everything else, meaning nary marque leads moreover 3 of 5 prompts, counts arsenic unsettled.
Run those definitions crossed the afloat dataset, and June 2026 looks wide open. Only 15.2% of categories had a clear owner. Just complete half, 53.7%, were unfastened fields pinch aggregate reliable contenders and nary marque adjacent to locking the door.
Indig classed each 1,094 categories by estimated AI hunt measurement and divided them into 2 moreover groups. The higher-volume half accounts for 98% of each AI hunt request successful the sample, and it has the little proprietor complaint of the two, 11.3% versus 19% for the lower-volume half. Stack that up and 89.3% of estimated AI hunt request sits wrong categories pinch nary clear proprietor yet. The categories worthy the astir money to inhabit are, correct now, the slightest decided.
I’ve spent the past 3 years watching SEOs statement whether topical authority is simply a existent algorithmic unit aliases a convenient communicative agencies show clients to warrant a contented calendar. Indig’s information is the first point I’ve seen that really resolves the mobility pinch a existent number. Topical authority is existent successful AI search, and the brands still treating class attraction arsenic optional are giving distant nine-figure categories to whoever bothers to show up consistently crossed each 5 punctual types.
Owners Are Hard To Dethrone, But Only When The Lead Is Wide
The 2nd half of Indig’s study is wherever the strategy gets sharper. He tracked month-over-month activity changes crossed the aforesaid categories and recovered that a clear proprietor held first spot successful 90.4% of comparisons.
The switches that did hap were concentrated almost wholly successful categories wherever the lead was already thin. Categories wherever the leader changed had a median lead of conscionable 1.3 percent points going into the switch. Categories wherever the leader held connected had a median lead much than double that, 2.9 points. Growth trajectory told Indig almost thing connected its own. He shows 3 examples wherever a starring marque trended upward period complete period and still sewage overtaken, because a competitor simply grew faster aliases closed a constrictive gap. The instruction isn’t that momentum is meaningless. It’s that a one- aliases two-point lead is not a lead worthy defending yet, and a marque sitting successful that scope should presume it’s still successful a fight.
Indig besides recovered thing that undercuts a communal presumption astir really ownership gets built. Citations and brand mentions turned retired to beryllium only weakly related, pinch a somewhat antagonistic relationship of -0.229. The astir often cited domain successful a class was seldom the aforesaid arsenic the astir often mentioned brand, matching up only 20.8% of the time. But the astir mentioned marque was cited astatine slightest erstwhile 69.9% of the time. Getting cited a batch doesn’t automatically make you the marque ChatGPT reaches for by default. That’s a meaningful favoritism for anyone whose GEO strategy presently starts and ends pinch “get much citations.”
Owners did stock immoderate traits versus their runner-up: higher branded hunt measurement successful 55.7% of pairs, higher integrated postulation successful 48.4%, and a higher Semrush Authority Score successful 52.5%. Indig is observant to telephone these correlations alternatively than a formula. A brand doesn’t gain ChatGPT’s default answer by buying branded hunt volume. It earns it by being the root ChatGPT’s training and retrieval support landing connected erstwhile the exemplary tries to reply each 5 variations of a question, the definition, the comparison, the alternatives list, the usage case, and the buying decision.
How To Use This Before The Window Closes
Here’s wherever I’d constituent a practitioner’s adjacent 4th of work, adapted from Indig’s model and my ain Citation Share of Voice attack to search AI mentions.
- Pick your battlefield first, past measurement it properly. List the 10 to 20 categories wherever your marque really needs to beryllium the default answer. For each one, way the aforesaid 5 punctual types Semrush used, a meaning question, a comparison, an alternatives list, a usage case, and a buying question, not conscionable your marque sanction successful isolation.
- Sort by lead size, not by whether you’re presently “winning.” A class wherever you lead by 1.5 points is not a class you’ve won. Treat thing nether astir 3 points arsenic contested and put existent resourcing down closing the remaining punctual gaps, since Indig’s information shows that’s the nonstop scope wherever turnover happens.
- Stop chasing citations arsenic the full strategy. Use citations to gain your measurement into the reply set, past displacement fund toward the circumstantial contented types, comparison pages, impervious points, third-party coverage, and clear positioning, that find whether you’re the marque really named erstwhile you’re successful the room.
The GEO speech complete the past twelvemonth has fixated almost wholly connected citation counts because citations are the easiest point to driblet into a dashboard. Indig’s information says citations are a door, not the room. Practitioners who support optimizing for the doorway while a competitor optimizes for being the sanction ChatGPT really says retired large are going to suffer categories they thought they were winning.
The model Indig describes won’t enactment open. Eighty-nine percent of request sitting successful unsettled categories is an unusually generous caput start, and the brands that move connected it successful the adjacent fewer quarters, alternatively than treating this arsenic 1 much study to bookmark, are the ones that will still beryllium the default reply erstwhile personification runs this aforesaid study successful 2027.
More Resources:
- Free AI Citations Won’t Last. I’ve Watched Google Fence Off A Wide Open Field Before
- 62% Of AI Brand Recommendations Vanish After One Buyer Question – New Data
- GPT-5.5 Update Changes How ChatGPT Cites Sources
Featured Image: Roman Samborskyi/Shutterstock
English (US) · 
Indonesian (ID) ·