TL;DR: One SEO ran a genuinely clever experiment: opt a site out of AI Overviews entirely, then watch what happens to its traffic. The results were strange. Clicks didn’t just stay flat when AI visibility disappeared. They dropped, suggesting AI Overviews might be doing something closer to building brand trust than simply stealing clicks outright.
Cyrus Shepard posted the case study on LinkedIn, and it’s gotten more attention from the SEO crowd than you’d expect for a single-site, self-reported test. Part of that is because almost nobody actually runs this experiment. Opting a site out of Google’s AI features on purpose, just to measure the effect, requires a level of comfort with short-term risk that most site owners don’t have.
What Shepard actually did
Using Google’s search generative AI opt-out controls, Shepard blocked one site from appearing inside AI Overviews and AI Mode while leaving everything else about the site untouched. AI Overviews kept appearing on the results pages for those same queries. His site simply stopped showing up inside them during the test window.
He then compared his site’s overall clicks, impressions, and click-through rate during that blackout period against the stretches before and after it, when the site was back to appearing normally inside AI Overviews. By backing out the known AI-attributed impressions from his overall numbers, Shepard estimated a click-through rate of about 0.9% for the AI-visible impressions specifically. Not a huge number on its own, but not what came next.
The part nobody expected
When AI visibility went dark for his site, regular organic clicks dropped too. Not stayed the same. Dropped. Shepard called this the most surprising result of the whole test, and it’s easy to see why. The conventional wisdom around AI Overviews has mostly assumed a zero-sum relationship: every click an AI summary “steals” is a click a traditional blue link would otherwise have earned. If that were the whole story, pulling a site out of AI Overviews should have left regular organic traffic alone, or maybe even boosted it slightly as users scrolled further down the page looking for more detail.
That’s not what the data showed. AI impressions made up roughly 8% of the site’s total impressions during the test, and overall impressions dropped once those AI appearances stopped happening. Shepard’s working theory has a name now.
“This admittedly small case study supports the concept of AI creating a ‘Search SERP Halo Effect,'” Shepard wrote, arguing that showing up anywhere on a results page, AI summary included, makes a user more likely to click something, anything, on that page.
Greg Sterling, whose research firm Near Media has looked at similar questions, backed up the general shape of that theory in a comment on the post. His team’s own research suggests AI Overviews can create brand lift or awareness that shows up later, in a subsequent search rather than an immediate click. Not consistently, he noted, but often enough to matter.
Reasons to not get too excited about this yet
Barry Schwartz at Search Engine Roundtable, who covered the study, was blunt about its limits. One site. A small sample. Estimated numbers derived by subtraction rather than measured directly. He also pointed out something the test itself couldn’t control for: it never compared clicks on a page with an AI Overview against clicks on the exact same query without one. It only measured what happened when one particular site stopped appearing inside an AI Overview that was still there for everyone else.
Schwartz also made an interesting side observation. He says he hasn’t seen anyone else publish similar data, and he suspects that’s because most site owners are too nervous about the downside to run the experiment themselves. That’s a reasonable instinct. Deliberately disappearing from a feature that now touches more than 80% of all Google searches is not a risk most businesses want to take just to satisfy curiosity.
Still, a result this counterintuitive deserves more attempts at replication, not less. If the halo effect theory holds up across more sites and more industries, it changes how publishers should think about AI visibility entirely. Instead of treating an AI Overview appearance as a straightforward loss, it might be more accurate to treat it as a different kind of exposure with its own, harder-to-measure payoff.
What a proper version of this test would need
A single site opting out for a few weeks is a starting point, not a conclusion. A study that actually settled this question would need a much larger panel of sites across different industries, a controlled comparison against similar sites that stayed visible in AI Overviews the entire time, and a longer measurement window to rule out seasonal noise or one-off algorithm changes that happened to coincide with the test period. None of the big SEO platforms has published anything close to that yet, which is itself a little telling given how much money rides on getting this answer right.
There’s an obvious reason nobody’s rushed to fund that kind of study. Google doesn’t make it easy to isolate the effect of AI Overviews specifically, and any site willing to opt out at scale for research purposes is accepting real, measurable revenue risk in exchange for data that mostly benefits the industry at large rather than that specific site. Shepard took on that risk voluntarily for one property. Getting a dozen companies to do the same, with their actual revenue on the line, is a much harder ask.
There’s also a practical middle ground between fully opting out and doing nothing. A handful of SEOs in the comments on Shepard’s post suggested testing this at the page level rather than the whole-site level, opting out a handful of lower-value pages while leaving the bulk of a site’s content untouched. That approach would take longer to produce a clean signal, but it would spread the revenue risk thin enough that more sites might actually be willing to try it, which is exactly the kind of broader participation this research needs to move past a single anecdote.
The bottom line
Take Shepard’s 0.9% figure with a grain of salt, because it is exactly that small and exactly that uncertain. But don’t dismiss the underlying question his test raises. If pulling out of AI Overviews costs you more than the AI clicks you were missing in the first place, the entire debate about whether to fight Google’s AI features or simply live with them needs a rethink, and right now almost nobody has the data to settle it either way.
The Teck Hustlers Editorial Team covers technology, business, finance and gaming news for readers who want the story behind the headline. Every article is researched, written and edited by our in-house team, checked against company announcements, filings and reputable reporting, and updated when new facts emerge. We focus on what each development means for founders, professionals, investors and everyday users. Have a tip or a correction? Reach us through our contact page.
