Small Google AI Opt-Out Case Study Estimates 0.9% CTR

By Nathan Brooks

A small LinkedIn case study estimated a 0.9% click-through rate for Google AI impressions, which represented about 8% of total impressions.

Cyrus Shepard ran the test through Google Search’s generative-AI opt-out controls. He blocked an unnamed site from appearing in AI Overviews and AI Mode, then compared impressions, clicks and click-through rate before, during and after exclusion. The result is a single-site case observation, not an official Google benchmark or broad ranking study.

What changed when the site opted out?

Shepard reported five changes during the exclusion period:

• AI impressions fell immediately to zero.
• Overall impressions also declined.
• Regular organic clicks declined as well.
• The site appeared in AI Overviews before and after, but not during the test.
• Individual URLs with the most AI visibility dropped during exclusion and rose after restoration.

He also reported that sitewide average position increased. He attributed that increase to a couple of new posts and brand search. The average-position movement therefore did not simply mirror the visibility change.

The post gave no test dates, exact impression count, traffic geography, device split, industry or full sample size. Shepard described it as a “small case study.” The material contains no Google statement, third-party tool analysis or independent replication confirming the 0.9% figure.

How did Shepard calculate 0.9%?

Shepard said: “AI impressions had an estimated ~0.9% CTR based on analyzing overall before/after click data (not too shabby).” He shared a before-and-after chart and later clarified that the estimate covered AI impressions before and after the exclusion. He calculated it from overall site CTR and impressions, then subtracted known AI impressions during the experiment.

That method makes 0.9% a modeled estimate from aggregate data. The post did not present a dated sample, click distribution, query set, confidence interval or comparison site. It also did not break out the 0.9% rate by AI Overviews, AI Mode placement, landing-page type or search intent. The figure should not become a universal benchmark.

Does the test prove an AI halo effect?

Shepard’s explanation was a “Search SERP Halo Effect.” His argument was straightforward: more places for a brand to appear increase the chance that users will click one of those places. He said an AI citation, link or mention may influence clicks on ordinary organic results farther down the results page.

His most surprising observation was that regular organic clicks dropped when AI visibility was disabled. That pattern is consistent with AI exposure supporting later organic clicks. It does not establish that mechanism by itself.

AI Overviews remained visible before, during and after the experiment, while only the tested site disappeared from them during the exclusion. The test therefore does not compare a SERP with AI Overviews against the same SERP without them.

This distinction limits what publishers can claim. The account supports a possible halo effect for one site’s AI-visible pages, but it does not isolate why organic clicks fell. New content, brand demand and the continued presence of AI answers could influence the outcome. The disclosed increase from new posts and brand search also complicates any sitewide ranking comparison.

Which sites and verticals could see this effect?

The post did not identify the site’s industry, so it provides no vertical-specific effect for publishers, ecommerce businesses, local companies or B2B sites. It also does not separate brand-query traffic from non-brand discovery. Any claim that the 0.9% rate applies to a whole category would exceed the evidence.

For the tested site, the clearest signal sat with individual URLs that had the strongest AI visibility. Those URLs showed a decline during exclusion and a rebound after restoration. Brand-search behavior appears relevant at the site level, although the post does not quantify it.

The practical impact may therefore differ between citation-dependent landing pages and pages supported by direct demand, but the account cannot establish a universal pattern.

What should publishers monitor next?

Greg Sterling’s comment added a separate observation from Near Media research. He said:

“We’ve seen in our research at Near Media that AI overviews create a kind of brand lift or awareness that translates into subsequent searches and clicks, not consistently but many times.”

His comment supports the halo hypothesis, but it is not a controlled replication of Shepard’s test and does not validate the 0.9% estimate. Its stated inconsistency also makes the effect unsuitable as a universal rule.

The account ends with a call for more sites to run the same opt-out comparison. The most useful next evidence would come from replicated tests that publish dates, sample sizes, query- and page-level click data, average positions, brand-search effects, and whether AI Overviews remained on the SERP.

Those records could separate traffic generated by an AI citation from the broader halo attributed to AI visibility.