Rémi Kerhoas, paid media and data director at the agency Eskimoz, reports in Search Engine Land that search terms reports across 400-plus Google Ads accounts show no growth in longer queries. That contradicts what Google’s claims would predict. Google has said AI Mode prompts average triple the length of classic queries.
The finding matters because advertisers have no native way to see AI search traffic. Google Ads does not split performance between traditional and AI results, so query length is one of the few indirect signals a team can inspect. If that signal is flat, AI search activity may be present but unreadable.
Kerhoas disclosed his own stake: Eskimoz supplied the data. He compared three markets. The U.S. runs ads inside both AI features. The U.K. has the features but no ads yet. France gained the features in late July, also without ads. His hypothesis was that U.S. accounts would show the clearest trace.
They did not. In the U.S., query-length graphs showed no meaningful growth in long queries, and conversions told the same story. Broad match and AI Max terms, the likeliest carriers of AI-surface ads, stayed flat too. Most AI Max search terms held only one to three words. The U.K. and France looked much the same.
Kerhoas offers three explanations and rates each one.
- Buying-intent prompts may simply be briefer (confidence: Low). Google’s figure averages every kind of AI Mode session, while his performance-oriented sample may undercount lengthy informational questions.
- AI placements may make up a minor part of ad delivery (confidence: Moderate). Per Google, AI Mode earns less per query. Showing ads there requires a campaign setup built on Performance Max, AI Max, or broad match keywords.
- The report does not display prompts as typed (confidence: High). Terms with thin query volume get dropped, and for AI features the listed term is Google’s best approximation of intent.
The third explanation is the strongest by his own rating, and it carries the most consequence. It means the report may be answering a different question than advertisers think. Unreported terms hid about 40% of search terms in his data, and that share has barely moved in nearly three years. Kerhoas argues paid search now looks more like paid social, where targeting is automated and term-level detail is thin.
The limits are real. The data covers Search and Performance Max campaigns from Q1 2024 to Q3 2026, though Performance Max search terms exist only from the second quarter of 2025. Campaigns spanning several countries were left out. Unreported impressions were estimated by subtraction. Most clients chase performance goals, and Kerhoas believes upper-funnel queries fill most of the AI Overviews and AI Mode supply. His sample therefore underweights exactly the traffic Google describes.
Google has not commented on the analysis, and this is a single agency’s book of business, not an independent audit. It also arrives as Philipp Schindler, Alphabet’s chief business officer, told the Q3 earnings call that “lower monetization” in AI Mode weighed on Search revenue growth. That fits the moderate-confidence explanation that AI ads are a thin slice of delivery.
For search teams, the practical step is to stop using query length as a proxy for AI exposure. Judge AI search impact through landing-page and conversion trends, and ask account reps directly which campaign types are eligible for AI placements before the next budget cycle.
Search Engine Land, analysis by Rémi Kerhoas of Eskimoz, published November 9, 2026.
