Kevin Indig just settled a debate that’s been going on halfway since the launch of AI previews. Do thematic authority which rewards brands for going deeper into a topic instead of broadening it, does it really impact AI search? Or is ChatGPT enough of a blank slate for a general brand to come in and grab a category cold?
Indig had access to six months (January-June 2026) of ChatGPT response data from Semrush, covering 1,094 US categories, five prompts per category, over 50,000 brands and over 600,000 quotes. He published the distribution in his Growth Memo Substack on July 20, with the title found on LinkedIn the same day. His answer is that topical authority is important in AI research, and the reason is sustainability. Once a brand gets a significant share of mentions in a category, it tends to stick with it.
This is the neat version. The more complicated and useful version, buried a few paragraphs into his message, is that almost no deal is settled yet.
The categories that matter most are the ones that no one owns
Indig built three buckets. A category has a clear owner when a brand appears in at least four of the five test prompts and beats the runner-up by five percentage points or more. An emerging leader is when a brand leads at least three prompts without reaching this gap. Everything else, that is, no brand leading even three out of five prompts, is considered unstable.
Run these definitions on the full dataset and June 2026 seems wide open. Only 15.2% of categories had a clear owner. Just over half, 53.7%, were open fields with several credible contenders and no brand close to locking the door.
Indig ranked the 1,094 categories based on AI-estimated search volume and divided them into two equal groups. The higher volume half represents 98% of all AI search demand in the sample, and it has the lower ownership rate of the two, 11.3% compared to 19% for the lower volume half. Stack that up and 89.3% of estimated AI research demand is in categories with no clear owner at the moment. The categories that are worth the most money to occupy are, at present, the least decided.
I’ve spent the last three years watching SEOs debate whether topic authority is a real algorithmic force or a convenient story that agencies tell their clients to justify a content calendar. Indig’s data is the first thing I’ve seen that actually settles the question with a real number. Topic authority is real in AI search, and brands that still view category focus as optional offer nine-figure categories to anyone who bothers to appear consistently across all five prompt types.
Owners are hard to dethrone, but only when the lead is wide
The second half of Indig’s analysis is where the strategy becomes more specific. It tracked month-to-month leadership changes in the same categories and found that a clear owner held the top spot in 90.4% of comparisons.
The changes that occurred were almost entirely concentrated in categories where the lead was already slim. Categories in which the leader changed had a median lead of just 1.3 percentage points before the change. Categories in which the leader held firm had a median lead more than double, or 2.9 points. The growth trajectory alone taught Indig almost nothing. It shows three examples in which a leading brand experienced an upward trend month-over-month and was still overtaken, because a competitor simply grew faster or closed a narrow gap. The lesson is not that momentum is meaningless. That’s because a lead of one or two points isn’t worth defending, and a brand in that range should assume it’s still in the fight.
Indig also discovered something that flies in the face of a common assumption about how property is built. Quotes and brand mentions Their relationship was found to be weak, with a slightly negative correlation of -0.229. The most frequently mentioned domain within a category was rarely the same as the most frequently mentioned brand, matching only 20.8% of the time. But the most cited brand was cited at least once in 69.9% of cases. Being cited a lot doesn’t automatically make you the brand ChatGPT looks for by default. This is a significant distinction for anyone whose GEO Strategy currently begins and ends with “get more quotes”.
The owners shared some traits compared to their finalist: higher branded search volume in 55.7% of pairs, higher organic traffic in 48.4%, and higher Semrush authority score in 52.5%. Indig is careful to call these correlations rather than a formula. A brand does not win ChatGPT default response by purchasing branded search volume. It earns it by being the source on which ChatGPT training and retrieval continues to land as the model attempts to answer all five variations of a question, definition, comparison, list of alternatives, use case, and purchasing decision.
How to use it before closing the window
Here’s where I would designate a practitioner’s next quarter of work, adapted from Indig’s framework and my own Quote Share of voice approach to tracking mentions of AI.
- First choose your battlefield, then measure it correctly. List the 10-20 categories in which your brand should actually be the default. For each, follow the same five types of prompts used by Semrush, a definition question, a comparison, a list of alternatives, a use case and a purchase question, not just your brand name in isolation.
- Sort by lead size, not by whether you’re currently “winning.” A category in which you lead by 1.5 points is not a category you won. Treat anything below about 3 points as contested and devote real resources to closing the remaining velocity gaps, because Indig’s data shows that’s the exact range in which turnover occurs.
- Stop chasing quotes as a global strategy. Use quotes to work your way through the answer set, then redirect budget toward the specific content types, comparison pages, proof points, third-party coverage, and clear positioning that determine whether you’re truly the named brand once you’re in the room.
The GEO conversation over the past year has focused almost entirely on number of citations because quotes are the easiest thing to put into a dashboard. Indig’s data indicates that quotes are a door, not a room. Practitioners who continue to optimize for the door while a competitor optimizes to be the name ChatGPT says out loud will lose categories they thought they would win.
The window described by Indig will not remain open. Eighty-nine percent of demand falling into unmet categories is an unusually generous head start, and the brands that get there over the next few quarters, rather than treating this as one more report to bookmark, are the ones that will still be the default answer when someone conducts this same study in 2027.
More resources:
Featured image: Roman Samborskyi/Shutterstock
