Google's AI Overviews cut outbound organic clicks by about 40% when they fire. That number alone has been circulating for months. But the updated finding from researchers Saharsh Agarwal and Ananya Sen is sharper, and more useful for anyone running pipeline off organic: the clicks that disappeared weren't junk.
Bounce rates, time on site, and return-to-SERP behavior showed no statistically significant difference between users who saw AI Overviews and those who didn't. About 4 in 10 same-tab clicks led back to the results page in both groups. Roughly 18% of visits ended within 10 seconds in both conditions. Time on site was statistically indistinguishable.
That matters because Google's own narrative, voiced by VP of Search Liz Reid, has leaned on the idea that AI Overviews reduce "bounce clicks" — low-value visits that don't convert anyway. If that were true, the no-Overview group should have shown worse engagement metrics. It didn't.
What the Data Actually Says (and What It Doesn't)
The study is a working paper on SSRN, not yet peer-reviewed. Worth flagging. But the experimental design is solid: randomized field experiment, treatment-control rotation after two weeks. When the groups switched, behavior reversed. Participants who lost AI Overviews saw clicks per search go up; those who gained them saw clicks drop. Zero-click rates mirrored each other.
AI Overviews triggered on roughly 41% of queries in the study window. With an Overview present, organic click rate sat at about 8%. Without one, 15%. Position-1 CTR dropped somewhere in the 32–58% range depending on the query, and informational searches bore the brunt. Navigational and transactional queries? Barely moved, though those samples were smaller.
The authors are direct about it: their result is "at odds with the view that AIOs primarily eliminate low-engagement website visits."
Why This Breaks Your Reporting (and How to Fix It)
If you're in marketing ops, this study forces a structural change in how you report organic performance. Impressions are up — 49% in some analyses — while click-throughs are down 30%. Those two numbers moving in opposite directions will make your dashboards look broken if you don't separate them deliberately.
Here's the operational problem: most B2B SaaS reporting stacks treat organic sessions as the top-of-funnel signal. Fewer sessions looks like regression. But if the remaining clicks convert at the same rate (which the engagement data suggests they might), your pipeline from organic could hold steady even as traffic drops. You won't see that unless you're tracking conversion rate and downstream engagement independently from session volume.
The zero-click increase (~35%) is structural, not a blip. AI Overviews are expected to appear on more queries over time. The researchers said as much. Plan accordingly.
What to Actually Do This Week
Step 1: Segment your reporting. Split organic performance into two buckets: queries where AI Overviews likely fire (informational, 53% trigger rate in the study) and queries where they don't (navigational, transactional). If your analytics can't do this natively, use Search Console data filtered by query intent categories. Even a rough split is better than blending everything.
Step 2: Stop chasing volume on informational queries. The CTR ceiling on informational searches with AI Overviews is around 8%. That's the new baseline. Redirect effort toward earning citations inside the AI answer itself, or toward content that targets transactional and navigational intent where clicks still flow.
Step 3: Measure what the remaining clicks do. If bounce rates and engagement don't differ (per this study), your conversion rate on organic traffic may actually improve as the low-intent tail gets absorbed by AI answers. Track MQL and SQL rates from organic separately from session counts. If conversion rate rises while sessions fall, your pipeline math might be fine.
The hypothesis (make it falsifiable): If we separate organic reporting by AI Overview likelihood and track conversion independently from sessions, then we'll find pipeline from organic is stable or declining less than session volume suggests, because AI Overviews absorb visits that weren't converting anyway.
Success = pipeline from organic holds within 10% of prior quarter despite session decline. Guardrails = conversion rate from organic doesn't drop below historical baseline. Stop-loss = if pipeline from organic drops more than 20% quarter-over-quarter, revisit content strategy and paid coverage for affected queries.
The Trade-Off Nobody's Naming
There's a real risk in this framing. "The clicks we still get are fine" can become an excuse to ignore a 40% traffic loss. It shouldn't. The study tells you the quality signal is intact; it doesn't tell you the volume loss is acceptable for your pipeline targets. Those are two different questions.
For most B2B SaaS companies, organic was already a long game with attribution problems. AI Overviews make it longer and muddier. The correct response isn't panic, but it isn't complacency either. It's adjusting your measurement to match the new reality, then making channel-mix decisions based on what the adjusted numbers actually show.
Google said the lost clicks were low quality. A randomized experiment says otherwise. The clicks you're losing were just as engaged as the ones you're keeping. That's not comforting. That's a signal to stop reporting on sessions and start reporting on what those sessions produce.
