Cyrus Shepard pulled one site out of Google's AI Overviews and AI Mode, watched the AI citations disappear, then watched something he didn't expect: his regular organic clicks dropped alongside them. That single-site experiment, reported by Search Engine Roundtable in October 2026, is the closest thing the industry has to a controlled opt-out test. The finding is uncomfortable enough to warrant your own version before you flip any switches.

The measurement problem nobody's solved

Google's Search Console rolls AI Overview traffic into the standard Web performance report with no filter separating AI Overview clicks from classic blue-link clicks. Google introduced a generative AI performance report in Search Console in June 2026, but it shows impressions and pages appearing in AI features, not a dedicated click or CTR metric. Any "AI Overview CTR" number from third parties is an estimate built on assumptions, not a native metric.

Shepard's workaround: compare total site clicks and impressions before, during, and after the opt-out, then back out the known AI impressions. He estimated an AI-feature CTR of about 0.9%, with AI impressions making up roughly 8% of total impressions before the test. During the opt-out, AI impressions fell to zero. The AI Overviews themselves still appeared on the SERP; his site just wasn't cited in them.

Why the "SERP halo" hypothesis matters for pipeline

Shepard characterized the most surprising outcome as a possible "Search SERP Halo Effect": being visible inside an AI answer may increase the probability that a searcher clicks your regular listing elsewhere on the page. Remove the citation, and you lose not just the direct click but some fraction of the organic clicks you thought were independent.

This is a single-site observation, not proof of causality. But it aligns with broader data. Ahrefs estimated a 58% lower position-one CTR when AI Overviews are present, modeled from 300,000 desktop keywords. Seer Interactive reported that across 53 brands and 2.43 billion organic impressions, CTR on queries with AI Overviews was 2.4% versus 3.82% on queries without them. Amsive found an average 15.49% CTR decline for keywords triggering AI Overviews, rising to 19.98% for non-branded keywords across 700,000 keywords and five industries including SaaS. The non-brand gap should concern B2B SaaS teams most: those informational queries are your top-of-funnel acquisition lever, and they're disproportionately the ones AI Overviews summarize.

How to run a small, reversible opt-out test

Setup (week 0): Pick 10-20 pages with meaningful AI impressions using the generative AI performance report in Search Console. Prioritize non-brand, informational pages tied to assisted conversions. Record four weeks of baseline data: clicks, impressions, CTR, average position, and landing-page conversion events (demo requests, trial starts, content downloads). Designate a holdout group of comparable pages you leave included in AI Overviews, matched by traffic volume, intent cluster, and conversion rate.

Launch (weeks 1-4): Use the Search generative AI control to exclude the test group. A URL-prefix property can override its parent's setting, so you can limit the change to one section without touching the whole domain. Changes generally take effect within a few days. Don't change anything else on those pages during the test. One variable.

Readout (week 5): Compare the test group against the holdout on organic clicks, CTR, impressions, and qualified leads (not just form fills). Segment branded versus non-branded queries. Success = qualified leads from the test group hold steady or increase relative to the holdout. Guardrails = total organic clicks on the test group don't drop more than 15% relative to the holdout. Stop-loss = if organic clicks drop more than 25% in weeks 1-2, restore the setting and read what happened.

Recovery (weeks 5-8): Restore the AI inclusion setting and track how long clicks and impressions take to return to baseline. Shepard's test showed recovery, but your timeline may differ.

What not to over-interpret

Seer Interactive cautioned against reading short-term CTR changes (even a two-month rise) as a durable trend. A four-week test gives you a directional signal, not a conclusive answer. If the signal is strong enough to act on, run it again on a different page set before expanding.

CTR declines also don't automatically equal lead declines. A page can lose 20% of its clicks and still produce the same number of qualified leads if the remaining visitors are higher-intent. That's why the experiment measures conversions downstream, not just Search Console metrics. If the halo effect is real for your site, you'll see it in the organic click data within two to three weeks.

One thing worth remembering: opting out doesn't make the AI Overview disappear. Google still shows the answer, built from other sources. If your category's AI answers are full of competitor citations, stepping out hands them the visible real estate.

Shepard's 0.9% CTR looked like a rounding error until the organic clicks fell with it. The number to watch isn't the AI citation click. It's everything else on the page that might be propped up by the citation's presence. Test on a small slice, measure leads (not just clicks), and let the data close the argument before you commit the whole domain.