AI and search intent: what two studies found

Two studies, reviewed by Search Engine Journal, converge on one point: the familiar split of queries into informational, navigational and transactional is falling apart.

Who measured what

The first study looks at the impact of AI Overviews on Spanish media (Clara Soteras, MJ Cachón and University of Barcelona researchers Mari Vállez and Carlos Lopezosa, within the CUVICOM project). The second is an eye-tracking experiment by Diego Criado, MJ Cachón and the Laika Team, measuring where people actually look on a results page.

The numbers behind the conclusion:

• 68% of US Google searches end without a visit to a website — against 49% in 2019;
• AI Mode passed a billion monthly active users, its queries are three times longer than usual, and refinements grow 40% month over month;
• four-word queries trigger AI Overviews 48.1% of the time, and three- and four-word queries account for nearly 70% of all AI answers;
• "why" pulls an AIO in 92.3% of cases, "what" in 85.7%, "who" in 68.4%;
• evergreen topics produce 75.2% of AI triggers: AIO appears on 34.6% of evergreen queries but only 1.1% of breaking news.

What the eye-tracking showed

Eye-tracking added an unwelcome detail: the AI Overviews block converts attention into a click 86.4% of the time, classic organic 95.5%, while images and product listings deliver 0% clicks despite capturing attention entirely. The "Explore Further" button is also zero.

Five actions instead of three types

Google already describes intent not as three types but as five actions: explore, decide, learn, create, do. "We no longer optimise for fragmented nouns — content now has to answer complex, timeless human questions directly," the author of the review puts it.

In practice this changes the unit of work. A cluster used to form around a high-volume phrase; now it forms around a question a person asks in full, with its context and follow-ups. Old keyword research does not become useless: it still shows demand, it simply stops being a ready-made page structure. How to build a core and turn it into structure is covered in the guide on keyword research.

The gap between evergreen and news deserves separate attention: 34.6% against 1.1%. A news item almost never makes it into an AI block, while a thorough explainer keeps working for months — and it is the explainer that builds visibility in the new interfaces. For a content plan that is a direct argument against pouring every resource into the news feed.

What to do with this

• take long question queries with "why" and "what" — they trigger AI answers most often;
• write detailed evergreen material for them: AI blocks barely appear on news;
• give a complete answer in the opening paragraphs, not after three screens of preamble;
• do not spend resources on SERP periphery that catches the eye and returns no clicks;
• count AI citations as an authority signal, not as lost traffic.

The real shift is not that clicks became scarcer, but that a query stopped being a set of words. It became a question — and whoever answers it fully wins.

Related reading: SGE and AI Overview: Optimising for AI Search · SEO Website Promotion