ChatGPT cites brands in topic categories outside their core expertise about 50 percent of the time. It names those brands in the generated answer only about 25 percent of the time, according to an analysis of Semrush’s AI Visibility Toolkit data that Kevin Indig published for Search Engine Land on August 5. Only about 9 percent of those distant appearances produce both a citation and a named mention. The gap between citation and recommendation is the story. The mechanism behind it runs counter to standard topical-authority advice.
Indig analyzed US ChatGPT data spanning 1,094 categories and five prompt variants per category from January through June 2026. For the breadth question he worked from 283,215 citations and 76,493 cases where a brand was actually named, pairing every one against what happened the month after. The relatedness question rests on a separate set: 45,578 appearances where brands pushed into new categories, covering 1,458 mapped entities. That sample size is large enough to isolate industry-level differences a smaller dataset would miss.
In categories close to a brand’s established expertise, ChatGPT cites the brand about 74 percent of the time, names it about 44 percent of the time, and delivers both outcomes together about 34 percent of the time. Citation-only presence barely moves between distant and close categories, 41 percent versus 40 percent. ChatGPT will use almost any qualifying page as a source. It reserves the named recommendation for brands it already associates with the topic.
The counterintuitive part of the analysis is what happens when a brand shows up in many categories at once. Indig’s data show no citation penalty for breadth at all. That association rises from plus 0.012 at one of five prompt variants to plus 0.062 at full five-of-five coverage. Brand mentions show a different pattern, but the penalty traces to shallow presence, not to breadth itself. Showing up in just one of five prompt variants across many categories associates with a drop of 0.051 in next-month mention share. That penalty turns slightly positive at full five-of-five depth. Spreading thin does not cost citations. It costs mentions, and only until a brand goes deep enough in each category it enters.
The effect is not uniform across industries. Finance and real estate reward repeat citation breadth most strongly. Finance’s citation-breadth association climbs from plus 0.054 to plus 0.139 between one and five prompt variants, and real estate’s rises from plus 0.021 to plus 0.135. Legal and healthcare show the opposite pattern for brand mentions. Legal’s mention-breadth association stays negative even at full coverage, at minus 0.058, and healthcare follows at minus 0.037. Both fields likely require more than source credibility before ChatGPT treats a brand as the recommended answer.
This publication reported on July 30 that Writesonic found roughly 40 percent of AI citations across seven engines never name the source brand. ChatGPT’s own ghost-citation rate in that study measured 37 percent. Indig’s analysis measures the same citation-to-mention gap from inside a single engine’s category structure. It attaches a lever that the Writesonic study did not have: depth within a category, not just presence in it, converts a citation into a named recommendation.
The underlying data comes from Semrush, a vendor with a commercial interest in AI visibility measurement (the sample size behind it, 283,215 citations and 76,493 mention observations across six months, is large enough to support the pattern described). Indig, who writes The Growth Memo, is an independent analyst rather than a Semrush employee. He notes the associations are correlational after statistical controls, not proof that category expansion causes higher citation share.
Search teams tracking ChatGPT visibility should stop treating category breadth as a liability. The next quarter’s audit belongs on prompt-level depth: brands present in only one of five prompt variants across many categories are losing mention share, not the ones present everywhere.
Search Engine Land published Kevin Indig’s analysis, republished with permission from The Growth Memo, on August 5, 2026.
