AI Overviews are the generated summaries that sit above results for a growing share of queries. They changed the economics of ranking: you can hold position 1 and still lose a third of your clicks. This page tracks what the data shows, and because that data moves fast, every claim carries a date. Last full review: July 2026.
How they work, briefly
The summary is generated from search results, grounded in indexed pages, with citations linking out. That means the pipeline covered in how Google Search works still gates everything: uncrawled, unindexed or unranked pages can’t be cited. AI Overviews changed who gets the click, and left largely untouched who gets considered.
What the click data shows (as of mid-2026)
- Every credible study finds organic CTR falls when an Overview is present. The honest answer is a range: roughly 15% to 60%+ depending on methodology, query mix and position, per the 12-study roundup.
- Ahrefs’ 300,000-SERP study put the drop for organic results at around 34.5% when an Overview tops the page, with a follow-up measuring 58% for position 1 by December 2025.
- Clicks from Overview citations themselves bottomed at 1.3% in December 2025 and climbed to 2.4% by February 2026, per Seer Interactive’s tracking. Small numbers, but the direction reversed, which few predicted.
- The clicks that survive appear to be worth more: Seer’s data shows cited-page visitors converting meaningfully better, arriving pre-informed by the summary.
- Zero-click searches keep climbing: around 65% of searches in 2026 end without any click, per aggregated tracking. Methodologies vary; treat the trend as solid and the decimal as decoration.
Who loses, who holds
Informational queries take the damage: definitions, how-tos, comparisons that a paragraph can answer. Transactional and navigational queries hold up far better, because a summary can’t complete a purchase or be a brand. If your model is ad-funded informational traffic, this is an existential trend line. If you sell things or capture demand, it’s been closer to a rounding error so far.
Getting cited, with the evidence honestly labelled
Correlation studies agree on a few things: cited pages tend to already rank well for the query or its cluster, they answer the question early and extractably, and they carry the trust furniture covered in EEAT. All of that is correlation. Controlled tests are thin, and anyone selling a guaranteed citation method is ahead of the evidence. My working approach: win retrieval the boring way (rank for the cluster), state the answer plainly near the top, and keep passages self-contained so one can stand alone as a source.
Measuring your own exposure
Search Console doesn’t separate Overview impressions or clicks; citations blend into the standard position data. So measurement is inference: segment informational queries, watch CTR at stable rankings, and compare against SERPs you’ve checked by hand for Overview presence. Crude, and currently the only honest method. Track it quarterly; this page gets re-reviewed on that cycle too.
Sources
Linked inline: Ahrefs, Seer via Search Engine Land, the multi-study roundup, and zero-click aggregates. Study findings are theirs; the strategic read is mine.