An SEO who flipped off Google's new AI Overviews opt-out toggle for one site didn't get the clean win he expected. Regular organic clicks dropped too, not just the ones coming from AI Overviews and AI Mode. Search Engine Roundtable reported the case study on October 30, and it's the first real before/after number most of us have seen on a toggle Google only finished rolling out globally this year.
What did the AI Overviews opt-out test actually show?#
Cyrus Shepard opted one site out of AI Overviews and AI Mode using Google's search generative AI control in Search Console, then compared clicks before and after. AI impressions, about 8% of the site's total impressions, went to zero the moment he flipped the switch — exactly what the toggle promises. The surprise sat in the rest of the data: regular organic clicks on the same site fell too, in the same window.
This is a different switch than the one we covered when Google rolled out its AI-training opt-out signal in robots.txt. That one blocks your content from training future models. This one blocks it from appearing in today's AI Overviews and AI Mode answers. Different button, different consequence.
The estimated click-through rate on those AI impressions, before the opt-out, worked out to roughly 0.9%. Shepard got there by subtracting known AI impressions from total site clicks and impressions — an estimate, not a direct measurement. He's calling the pattern a "SERP halo effect," the idea that showing up anywhere on a results page, even in an AI Overview snippet most people skim past, can nudge a click toward one of your other listings further down the same page.
Does this change anything if you're not watching this closely?#
If you've never touched Search Console's AI opt-out toggle and have no plans to, this changes nothing for you today. Keep reading anyway. The toggle exists now, and someone on your team, or your agency, will eventually get asked whether to use it.
At WebEpex we build SEO and AEO programs for small and mid-size businesses across the GCC, Europe, USA, Canada and Indian markets, and we fielded client questions about this exact toggle within two days of the Roundtable piece going up.
If you've been eyeing that switch because AI Overviews felt like they were stealing your clicks, this is the data point to sit with before you touch it. We've written before about how fragile any single AI-citation number already is, and a one-site, uncontrolled case study is exactly that kind of fragile number. I wouldn't treat 0.9% as gospel. I would treat "opting out might cost you clicks you weren't expecting to lose" as a real risk worth checking before you act on a hunch.
The honest trade-off here#
The case for taking this seriously: Shepard's chart shows AI impressions hitting zero the instant he opted out, confirming the toggle works exactly as described, and organic clicks dipping in that same window is at minimum a real correlation, not a theory. Greg Sterling of Near Media backed the pattern with his own research, saying AI Overviews create "a kind of brand lift" that shows up in later searches and clicks, "not consistently but many times."
The case for skepticism: it's one site, no control group, and AI Overviews kept showing for everyone else on that results page the entire time. The test removed this one site's citation, not AI Overviews itself. The site's average search position also rose during the window, because of new posts and brand search, which muddies any clean before/after read. Barry Schwartz, who reported it, called the study "super limited" himself. I agree with him, and I'd still rather have one small honest data point than none.
How we're handling it at WebEpex#
We pulled Search Console's AI opt-out report into the SEO reviews we run for retainer clients the week this story broke, specifically so nobody flips that toggle off a feeling instead of their own numbers. We also track AI referral traffic a layer deeper than Search Console shows by default, because a report that drops to zero doesn't tell you where those clicks went instead. That's the part Shepard's case study hints at and nobody has fully mapped yet.
We got this wrong once ourselves, early on, by reading an AI-visibility dip as a ranking problem when it was actually a measurement gap. We check both now before telling a client anything moved.
What I'd tell a client asking about this#
Don't flip the opt-out toggle this week over one LinkedIn post, mine included. Pull your own Search Console performance report, filter for AI Overviews impressions, and look at what share of your total clicks they represent before you decide anything. If AI impressions are under 5% of your traffic, this is background noise right now. If they're higher, test on one page type first, not your whole site, and watch total clicks, not just AI-sourced ones, for at least two full weeks.
If you want a second pair of eyes on what your Search Console AI data is actually telling you, send me what you're seeing and I'll give you a straight read, no pitch attached. Two minutes, that's it. cal.com/webepex/growth-review
