Social Rocket
Brands are increasingly treating AI visibility as a dedicated marketing KPI. New Comscore data shows at the same time why a single AEO or AI visibility score is becoming less and less meaningful: usage is spreading more widely across multiple assistants, and even within the same journey, sourcing, citation and referral differ.
A recent Digiday report based on Comscore data illustrates this fragmentation clearly. Between January and June 2026, ChatGPT's share of the observed prompt volume fell from around 70 to 50 percent, while Gemini rose from 17 to 30 percent and Claude from 2 to 11 percent. This does not mean ChatGPT is being used less: the overall market is growing strongly at the same time.
AI discovery is spread across multiple assistants
According to Digiday, the Comscore analyses are based on an opt-in panel of around 500.000 to one million users. In June 2026, ChatGPT recorded around 168 million desktop conversations; its cross-platform audience rose at the same time from 59 million the previous year to 99 million.
The overall reach of AI assistants is also growing. In June 2026, according to Comscore, 35 percent of desktop users and 29 percent of mobile users visited an AI assistant. In January 2025, the figures were 25 and 12 percent respectively.
For marketers, it is therefore relevant not only whether a brand appears in AI search, but where. A prompt set that only tracks ChatGPT can now miss an increasingly large part of the discovery journey.
Sourcing is not the same as citation
Even more important is the distinction between the content a model uses to form an answer and the source that users actually see.
In the Digiday report, Comscore cites an example from the travel sector: Tripadvisor reportedly accounted for 61 percent of the travel sourcing signals used by AI platforms, yet appeared in only 21 percent of visible citations. This is a category example, not a general benchmark, but it shows why "being cited" and "being used for the answer" are not the same thing.
For marketing teams, this creates at least four separate measurement questions: Is the brand mentioned? Is the company's own domain visibly cited? Is content used as a source in the background? Does this generate referral traffic or revenue?
A brand can be visible in ChatGPT and invisible in Gemini
Earlier data already showed large differences between models. A Digiday report on AI visibility referred, among other things, to Tinuiti data according to which Amazon reached very different citation shares in July: around 4 percent on Microsoft Copilot, 0,1 percent on ChatGPT and no measurable share on Gemini.
This aligns with the large AI Visibility Index study by Semrush, which analysed 126 million US prompts. Semrush likewise explicitly distinguishes between brand mentions and citations.
Anyone measuring AEO should therefore track visibility separately by platform, prompt cluster, mentions, sources and competitors.
The IAB wants to turn rough benchmarks into reliable measurement
The Interactive Advertising Bureau has published its own measurement framework, "Measuring Visibility in the AI Era". According to the IAB, more than 20 companies now offer AI visibility measurement, often using different methods.
Among other things, the framework distinguishes between directional and decision-grade measurement and calls for more transparency about which prompts, models, time periods and samples feed into a score.
This matters because generative systems are not deterministic. The same prompt can return different answers across several runs. A ranking as in classic Google search cannot simply be transferred one-to-one.
Google itself is broadening AI discovery
Classic search is also increasingly shaped by AI answers. According to Comscore data cited by Digiday, in June 2026 around 39,4 percent of the Google desktop searches analysed contained an AI Overview; in July 2025 it was 25,8 percent.
This shifts the customer journey further. How brands are already adapting their media planning to this is shown in our article on AI Search and upper-funnel budgets. In creator marketing, AI citations are now even appearing as a pitch argument – more on this in our analysis of creator AEO.
Google is also starting to make AI visibility directly visible in commerce products, for example with new Merchant Center insights for AI Search.
What marketing teams can sensibly measure now
A reliable AEO dashboard should therefore not end with a single large score. A more useful approach is a matrix of assistant × prompt cluster × mention × citation × source × referral × conversion.
Reproducibility is also important: teams should document which prompts, models and points in time produced a value, and should not treat results from different systems as directly interchangeable.
The second challenge remains attribution. When bots and AI agents evaluate content and users only convert later, classic traffic signals blur. We have looked at exactly this problem in our piece on AI agents and bot traffic.