One AI Visibility Score Hides Two Different Problems

• In the US, Gemini mentioned more unique brands per answer than ChatGPT in every measured month;
• ChatGPT included the most sources per answer, Gemini the fewest;
• In ChatGPT citations in the US, news and informational sites led;
• AI Overviews relied more heavily on streaming services, business software, online education, and medical reference sites;
• AI Mode concentrated citation weight in a narrow circle of leading domains.

AI Overviews stands out the most: its top mentions overlapped with the three other platforms less than any other pair, in every measured month. Its source overlap with ChatGPT also proved minimal. A good result in ChatGPT says almost nothing about how a brand looks in AI Overviews.

The conclusion is simple: if customers use several AI search engines, first find the platform where you're failing — and only then spend money on fixing it.

Who is actually "carrying" the brand?

The old SEO instinct says to start with your own site. In the AI visibility context, this works less well. If the brand appears in the answer but the sources are publications, retailers, communities, and reference sites rather than your own domain, the problem lies outside the site.

In the US, Chevrolet, Disney, Ford, Honda, Marvel, Nintendo, and Toyota all made the top-100 by mentions on all four platforms — and at the same time, none of them cracked the top-500 by citations in any measured month.

Part of the work on AI visibility thus goes beyond SEO. The answer is obvious: when the brand is in the answer but the site is not among the sources, you need to study the third-party pages being cited. And the problem may need to be fixed not by the SEO team, but by content, engineers, or PR. The advice to "create more content" is a weak default if the source gap lies elsewhere.

The category's role is no smaller than the platform's

Even within the US, broad AI visibility benchmarks can mislead. Semrush found that in every measured month, differences between categories in mention-citation overlap exceeded differences between platforms.

In US real estate, the same set of source domains — Apartments.com, Realtor.com, Redfin, and Zillow — recurred across all four platforms every month. In the industrial sector, no such stable core of sources was found. A real estate team can start from an established list, while industrial companies first need to build a specific set of customer and competitor questions — a broad benchmark will tell them little.

A useful question about AI visibility always contains four anchors: the customer question, the set of competitors, the platform, and the category. Start with separate baseline measurements of mentions and citations against a fixed list of customer questions — and only then study the answers and cited pages themselves, before funding tactics. Business results should be kept separate from visibility metrics: the Semrush study does not prove that changes in AI visibility are causally linked to commercial outcomes.

The working principle sounds simple: first make a diagnosis — which gap you actually have — then hand the work to the team that can realistically close it. The full report is available for download after registering on the Semrush website.