How AI Search Summaries Are Changing Organic Click-Through Rates

What the 2025 Click Studies Found

Pew Research Center’s July 22, 2025 report examined browser activity from 900 US adults during March 2025. Users clicked a traditional search result after 8% of Google searches with an AI summary, compared with 15% of searches without one. Pew also found clicks on links inside the AI summary in approximately 1% of searches where a summary appeared.

This was an observational comparison, not an experiment that randomly assigned AI summaries to otherwise identical queries. The query mix and search context can differ between the two groups. The figures describe behaviour within the study, not a universal CTR reduction for every ranking position or website.

Similarweb reported another historical trend: the proportion of news-related Google searches ending without a click to a news website rose from 56% in May 2024 to nearly 69% in May 2025. The statistic concerns searches related to news, not all websites, and cannot isolate AI Overviews as the sole cause. Similarweb’s publisher research provides the original market context.

These studies do not make search ranking irrelevant. They show why a search impression, an organic click and an engaged website visit need to be reported separately.

What Search Console Now Reports for AI Overviews and AI Mode

On June 3, 2026, Google announced separate Generative AI performance reports for Search and Discover, initially for a subset of websites. Its official documentation states that the reports reached websites worldwide by August 31, 2026.

The Search Generative AI performance report counts eligible impressions from AI Overviews and AI Mode. It allows site owners to review which pages received exposure, where it occurred by country and device, and how impressions changed by date. There is a distinct report for generative AI features in Discover. These are reporting views, not new ranking systems.

Google’s AI features documentation confirms that AI Overview and AI Mode activity remains part of the standard Web Search performance totals. The dedicated report is a different view of exposure, not additional traffic that should be added to those totals. It does not currently provide AI-only clicks, CTR, queries or average position, and Search Labs experiments are excluded.

If the dedicated report does not appear for a property, Google says insufficient eligible AI impressions may be the reason. Absence of the report is not proof that the website received no Google search visits.

Who Needs a Different Interpretation of the Data?

Search result behaviour varies by topic and audience. A simple informational question may be answered without a visit, while a product comparison, a booking request or a detailed investigation may still require the original page. The Pew and Similarweb findings should not be applied to every sector without checking the site’s own queries and outcomes.

Publishers and Information-Heavy Websites

For publishers, compare the pages gaining AI impressions with their total Web clicks and relevant query trends. A rise in AI exposure alongside fewer clicks is a reason to investigate changes in search behaviour, not proof that the AI feature caused a precise number of lost visits. Subscription signups, returning readers and other business outcomes are separate measures.

The related MOCOBIN analysis of AI Overviews and publisher traffic considers the publisher-specific questions in more depth.

Small Businesses and Local Service Providers

For a local provider, the more important outcomes may be qualified calls, bookings, leads and completed sales. Segment AI impressions and general search performance by relevant pages and markets, then inspect first-party conversion records. Correct service details, opening hours, locations and customer-facing answers where those facts genuinely need updating.

A company serving several countries should compare local search intent rather than assume an English query behaves the same way when translated into Korean or Japanese. That is a planning precaution, not a finding established by the two click studies.

A Practical Reporting Sequence After the August Rollout

1. Start with AI exposure. Open Search Console’s Generative AI report for Search and identify the pages, countries, devices and settled dates with measurable AI impressions. Use like-for-like periods and note any preliminary data.
2. Compare ordinary Web performance. In the standard Performance report, inspect relevant queries, pages, clicks and CTR. Remember that AI Overview and AI Mode metrics are already included in Web totals, so do not sum AI impressions into them a second time.
3. Check the business outcome elsewhere. Review GA4 landing-page sessions, meaningful engagement, enquiries or sales. Search Console impressions are not sessions, and an AI impression is not a visit. Referrals from third-party assistants such as ChatGPT also need separate source-level analysis.
4. Diagnose before rewriting. Inspect search demand, SERP changes, indexing, technical releases and user intent before assigning a cause to lower clicks. Update an article when its evidence, comparison or usability is actually weaker.

The August 2026 Search Console AI reporting news explains the rollout and report definitions in more detail. For wider measurement questions, consult MOCOBIN’s AI search attribution analysis.

What the New Report Still Cannot Answer

The new visibility view is useful precisely because it no longer requires every AI exposure estimate to be inferred from overall CTR changes. But it does not yet tell site owners how many clicks an AI Overview or AI Mode result produced separately from the rest of Web search.

That limits causal claims. If a page’s AI impressions rise and overall Web clicks fall, the two observations may reflect changes in query demand, ranking positions, content relevance or search result presentation. No single chart proves which factor drove the change.

Google may expand the report’s metrics, but no additional release date or future click breakdown has been established here. Until then, the defensible approach is to use the dedicated report for exposure, the standard Web report for overall search interactions, and first-party analytics for outcomes. Those distinctions are more useful than applying the 2025 studies as forecasts for individual sites.