AI Overview data splits sharply on commercial queries

Tracking AI Overviews is becoming a Rorschach test. Search Engine Journal reports that the picture analysts paint of Google's AI answers depends heavily on which prompts, query types and markets sit in the sample, with commercial queries behaving very differently from informational ones. The implication for anyone briefing a CMO on "AI Overview exposure" is uncomfortable: most dashboards are measuring a slice and calling it the whole.

The mechanics are straightforward. AI Overviews trigger at different rates across query intents, languages and verticals. Studies that lean on broad informational keywords tend to overstate AIO prevalence for brands whose buyers ask transactional questions. Studies that lean on commercial keywords tend to understate the role AIO plays earlier in the funnel, when prospects are still framing the problem. The same brand can therefore appear either besieged by AI answers or comfortably untouched, depending on the analyst's keyword list.

This is not an academic quibble. It changes the budget conversation.

The commercial query exception

Commercial intent queries (those with buying signals, comparison language, vendor names) consistently show lower AIO trigger rates than informational ones. Google has been visibly cautious about injecting generative answers where money is about to change hands, and where hallucinations carry liability. For a bank or an insurer, that means the bottom-of-funnel terms most closely tied to revenue may still resolve to the familiar blue-link auction, while the upstream terms that shape consideration ("how does parametric insurance work", "what is concessional finance") are where AI Overviews are quietly reshaping the answer.

The reverse pattern holds for industrial groups and policy institutions. A cement manufacturer's customers do not type "buy cement online"; they ask about specifications, emissions disclosures, regulatory thresholds. Those are precisely the queries where AIO presence is high and where a single citation can stand in for a brand's entire position on a topic.

Why aggregate numbers mislead

A headline number such as "AIO appears on 47% of queries" is worse than useless for a CMO. It conflates query types whose economics differ by an order of magnitude. The right diagnostic is segmented: AIO presence on the queries that actually matter to the business, in the markets where the business operates, weighted by commercial value.

For multilaterals and policy institutions, the segmentation problem is sharper still. Their queries are overwhelmingly informational, often multilingual, and often poorly represented in the English-language commercial datasets that dominate the tracking industry. UNDRR, CGAP or ISO looking at a vendor's AIO benchmark drawn from US e-commerce keywords will learn almost nothing about their own exposure. They are in the part of the query distribution where AI answers are most aggressive, and where the visible tracking is thinnest.

What to actually measure

Three disciplines separate useful measurement from theatre.

First, build the keyword set from the brand's real demand, not from a vendor's panel. Pull the queries that drive qualified traffic, the ones sales hears on calls, the ones policy teams field from member states. Then check AIO behaviour against that list, not a generic one.

Second, segment by intent before reporting any aggregate. Informational, navigational, commercial and transactional queries should each carry their own AIO presence rate and citation share. A single blended number hides the only finding that matters: where in the funnel the brand is being mediated by a model.

Third, track citation share inside AIO, not just AIO presence. Being absent from an AI answer that appears on a high-value query is a different problem from competing inside one. The fix differs too. Absence is usually a content gap; presence without citation is usually an authority gap.

The brand implication

The brands that will look good in 2026 AI-visibility reports are the ones whose marketing teams stopped accepting vendor benchmarks at face value this year. For financial services firms, that means measuring AIO behaviour on regulated-product queries separately from educational ones, because the trigger rates and the legal exposure diverge. For industrial groups, it means treating technical and standards queries as the priority surface, since that is where AIO is most active and where buyer engineers are doing their homework. For multilaterals, it means commissioning measurement in the languages and topics their stakeholders actually use, rather than inheriting an English commercial dataset by default.

The measurement problem is not a reason to wait. It is a reason to instrument properly now, while competitors are still quoting whichever benchmark flatters them. The brands that know precisely which of their queries trigger AIO, and who gets cited inside it, will be the ones briefing their boards with numbers rather than anecdotes.