TL;DR

Kevin Indig's usability studies of Google's AI Overviews (2025, 69 participants) and AI Mode (2025, 37 participants) recorded how real people read and act on AI answers. Most users skimmed, rarely read past the top third of an AI Overview and seldom clicked its citations; in AI Mode, about three-quarters of sessions never left the AI pane. Yet for purchases and high-stakes decisions, people still checked recognised brands, review sites and community sources before deciding. The practical conclusion: aim to be cited near the top of AI answers and to be the trusted source people verify with, and expect fewer, better-qualified clicks.

Who Kevin Indig is and what he studied

Kevin Indig is a growth advisor who writes the Growth Memo newsletter and contributes to Search Engine Journal. In 2025 he published what he described as the first comprehensive UX study of Google's AI Overviews, co-designed with Eric Van Buskirk of Clickstream Solutions, who directed the research.

Most discussion of AI search relies on aggregated click data. Indig's studies are different: they watch individual people search. In the AI Overviews study (published May 2025), 69 English-speaking US adults, recruited through Prolific, each completed eight Google searches, six of which triggered an AI Overview. Sessions were recorded in UXtweak with screen capture, scroll and cursor tracking and think-aloud commentary, and coders logged every element that held a participant's attention. A follow-up AI Mode study (October 2025) used the same approach with 37 participants and seven tasks each.

What people actually did with AI Overviews

  • They skimmed. 86% of participants skimmed quickly. Median scroll depth within an AI Overview was 30%, and about seven in ten searchers never read past the first third.
  • They expanded but did not dig. 88% clicked "show more" to expand a truncated overview, but that did not translate into reading it fully.
  • Citation clicks were rare. Only 19% of mobile users clicked a citation element inside the overview, and 7.4% of desktop users clicked within it.
  • Device mattered. Mobile users scrolled further than desktop users (54% versus 29% scroll depth) and clicked out more. Users aged 25 to 34 on mobile relied on AI Overviews most heavily.
  • Trust decided the next step. Where recognised brands, authority sites or .gov and .edu results appeared, they were chosen first in 58% of cases. Hesitation rose on medical and financial queries when an overview cited unfamiliar brands.
  • Commercial decisions happened elsewhere. For commercial and transactional queries, 81% of final answers on desktop and 78% on mobile came from elements other than the AI Overview.

Indig reanalysed the same recordings in a September 2025 follow-up, Trust Still Lives in Blue Links, and identified four patterns: efficiency-first users who take the quick answer, trust-driven users who scroll more and click out to authority sites, comparative users who open several tabs, and sceptical users who largely ignore AI answers on high-stakes queries. His summary:

"AIO is a preview layer. Blue links still close." — Kevin Indig, Search Engine Journal

AI Mode is stickier still

The AI Mode study showed a more enclosed experience. About three-quarters of sessions never left the AI Mode pane, the median number of external clicks per task was zero, and 77.6% of sessions had no external visits at all. Participants spent roughly 50 to 80 seconds per task, more than twice as long as with AI Overviews, but mostly inside the answer.

Where people did leave, the destination depended on intent. In product tasks, most exits went to brand sites or marketplaces; in a software comparison task, 83% of exits went to review sites. Indig's recommendation is to treat AI Mode as a destination rather than a doorway, and to shift the goal from winning the click to earning the citation.

What this means for content and click expectations

Google itself says, in its guidance on AI features, that clicks from AI Overviews tend to be higher quality. Indig's observations offer a plausible reason: the people who click have often already read a summary and are verifying or ready to act. For a marketing team, that means two adjustments: accept lower volumes of informational traffic, and make sure you are the source people trust when they do check.

  1. Track citations, not just sessions. For your priority queries, record whether you are cited in AI Overviews and AI Mode, how high in the answer, and alongside whom. Most users never scroll far, so position within the answer matters.
  2. Put the answer first. Open key pages with a concise, self-contained answer, then the detail. This suits both skimmers and AI systems that extract passages.
  3. Lead with visible authority. Show named authors with relevant expertise, credentials, certifications and real reviews. Users in the study filtered by trust first and relevance second.
  4. Own the validation channels. People double-checked on Reddit, YouTube and review sites. Make sure your brand is present and accurately represented where your buyers verify.
  5. Make product pages decision-ready. Indig advises merchants to show prices, availability and proof above the fold, because the few people who leave AI Mode arrive ready to decide.
  6. Segment by device and audience. Mobile users behaved differently from desktop users, and younger users relied more on AI answers. Plan content and measurement accordingly.
  7. Re-forecast traffic-dependent revenue. If your model depends on informational sessions, such as advertising or affiliate income, plan for fewer of them.

Limits of the research

Indig is transparent about the constraints. The samples are small (69 and 37 people), US-only and skewed young (58% of AI Overviews participants were under 35). Think-aloud protocols can lengthen dwell time, the AI Overviews fieldwork overlapped a Google core update, and the AI Mode study did not include a direct comparison with non-AI results. Some figures on click loss also differ between sections of the original write-up. Treat the studies as rich qualitative evidence of how people behave, not as precise traffic forecasts for European markets.

Key takeaways
  • In Indig's 2025 study, most users skimmed AI Overviews and about seven in ten never read past the first third.
  • Clicks on citations inside AI Overviews were rare; AI Mode kept about three-quarters of sessions inside the pane.
  • For purchases and high-stakes topics, users still verified with recognised brands, review sites and communities.
  • Optimise to be cited high in the answer and to be the trusted source people check.
  • Expect fewer but better-qualified clicks, and measure visibility as well as traffic.

FAQ

Do AI Overviews kill clicks completely?
No. In Indig's study, most commercial decisions were still made from non-AI results, and users verified with trusted brands and review sites. Informational clicks fall most; clicks from people ready to act are more resilient.
Why does being a recognised brand matter so much?
Participants applied a trust filter before relevance. Recognised brands and authority sites were chosen first in 58% of cases where they appeared, and unfamiliar brands cited in health or finance answers caused hesitation.
Do the findings apply to European users?
The studies only covered US participants, so they are indicative. The underlying behaviours, skimming and verifying with trusted sources, are plausible elsewhere, but test with your own analytics and customer research.

Sources

Plan for fewer, better clicks

Talk to JRK about how AI answers are changing your search traffic and how to become the source your buyers trust.

Book a free growth session →

Want this working for your brand?

Book a free 30-minute growth session. We'll look at your funnel and tell you what we'd do first — honestly.

Book a growth session →