Should You Opt Out of AI Overviews? What the Data Says

Should You Opt Out of AI Overviews? What the Data Says

For most businesses, opting out of AI Overviews is not worth it yet, and the first public test suggests it can cost organic clicks as well as AI ones. Google's Search Console now includes a setting that removes a site's content and links from AI Overviews, AI Mode and generative features in Discover. With AI Overviews appearing on a large majority of searches, many site owners are asking whether to flip it.

This article explains exactly what the control does and does not do, what the first published opt-out test found, how it differs from the separate AI training controls announced in late October 2026, and the decision framework we use with clients.

What the Search Console control actually does

Google's help documentation for the Search generative AI control describes a setting under Settings in Search Console. It controls whether your site's links and content appear in:

• AI Overviews;
• AI Mode;
• generative AI features in Google Discover.

Google notes the list may expand over time. The default for every property is "Include my site's links and content".

Just as important is what the control does not do, according to the same page:

On 6 October, Search Engine Roundtable reported an update to the documentation: owners of a top-level domain property now see a notification when a child property overrides the parent setting. That matters for larger organisations, where one team could opt a subfolder out without the main site owner knowing.

Why the question is urgent now

Two developments in October pushed this decision up the agenda.

AI Overviews now appear on most searches

On 26 October, Search Engine Roundtable reported that recent data from Semrush, SE Ranking and seoClarity shows AI Overviews on over 80% of Google searches. The figures vary by provider and query set: SE Ranking reported 82.5% of US brand searches, while seoClarity reported over 62% of queries, ranging from 55% to 84% by industry. Whatever the exact share in your sector, an AI answer is now the first thing most searchers see.

Measurement is still incomplete

Search Console's generative AI performance reporting shows impressions, but not clicks. That leaves site owners making a traffic decision without the traffic data, which is exactly the condition under which people make expensive mistakes.

What the first opt-out test found

On 30 October, Search Engine Roundtable covered a small case study by Cyrus Shepard, who used the control to block his own site from AI Overviews and AI Mode and compared the data before and after. His findings, as reported:

• AI impressions were about 8% of the site's total impressions;
• by comparing overall clicks before and after, he estimated AI impressions had a click-through rate of roughly 0.9%;
• regular organic clicks also dropped when the AI features were disabled.

The caveats are significant. Barry Schwartz described it as "a super limited study based on a small sample size", and AI Overviews still appeared on the results pages during the test (they simply no longer cited the site), so it does not compare clicks with and without AI Overviews present. One site, one period, one estimate.

Even so, the direction of the result is the important part. The common assumption behind opting out is that AI answers only take clicks away. This test, limited as it is, found the opposite effect on organic clicks. One plausible explanation is that being cited in the AI answer above the results reinforces a searcher's choice to click the organic listing below it. That is a hypothesis, not a finding, but it is enough to make opting out a decision to test rather than assume.

Don't confuse it with the AI training controls

A second set of controls was announced at the end of October, and it is easy to mix the two up.

On 30 October, Search Engine Roundtable reported that Google plans to adopt a work-in-progress IETF proposal for a shared AI training preference in robots.txt, allowing a site to state its training preference across multiple AI providers in one step. If a site opts out of training through the shared signal without separately specifying Google-Extended, Google will treat it as though Google-Extended were disallowed. Google aims to support the control in 2027.

The same day, Search Engine Roundtable reported that Microsoft Bing is introducing separate controls for AI training and AI grounding, stating that "the traditional Bing rankings are unaffected."

The distinction that matters:

• Training controls decide whether your content can be used to train models. They do not decide whether you appear in AI answers today.
• The Search Console generative AI control decides whether your content and links appear in Google's AI search features. It does not affect training.

A publisher worried about its content training a model and a business worried about losing clicks to AI Overviews need different settings. Choosing the wrong one solves neither problem.

Our decision framework

We do not recommend opting out by default. We recommend it only where the evidence for a specific site supports it, and we recommend testing rather than switching sitewide.

When staying in is the better choice

For most service businesses, local businesses and B2B companies, the case for staying in is strong:

• Your content is a route to an enquiry, not the product. If a searcher reads your answer in an AI Overview and then calls you, the AI citation did its job.
• AI citations follow organic rankings. Ahrefs tracked about 37,500 AI citations across five AI tools for 20 days and found that pages ranking in Google were about seven times more likely to stay cited in AI Overviews and AI Mode for at least a week. Being present in both reinforces visibility; it does not split it.
• Your competitors are likely staying in. Opting out does not remove the AI Overview. It removes you from it, leaving the citation to someone else.

When testing an opt-out may make sense

There are cases where the trade-off is genuinely different:

• Your content is the product. Publishers, data providers and reference sites whose revenue depends on the page view may lose more from a complete AI answer than they gain from a citation.
• Your Search Console data shows high AI impressions and falling clicks on the same pages, sustained over weeks, not days.
• A specific section is affected, such as a set of definitions or calculators that AI answers reproduce in full.

How to test without risking the whole site

Because the control works per property and child properties can override the parent, it is possible to test on part of a site:

1. Verify a URL-prefix property for one section, such as a glossary or a single content folder.
2. Record 28 days of baseline clicks and impressions for that section, alongside a comparable section that stays included.
3. Opt the test section out and leave everything else untouched.
4. Compare both sections over the following four to six weeks, allowing for the few days the change takes to apply.
5. Keep the opt-out only if the test section clearly outperforms its comparison.

Annotate the start date. If a Google update lands during the test, as one easily could in November, the comparison section is what keeps the result interpretable. Our timeline of confirmed Google updates lists the dates to check against.

Earning citations is usually the better use of time

For most sites, the more valuable work is improving how often AI answers cite them. Studies published in October point in a consistent direction, with the caveat that all of them describe correlations rather than causes.

Search Engine Journal reported on 28 October on TJ Digital's analysis of 193,507 AI answers. Cited pages opened with a direct answer 23% of the time, against 14% for pages the AI retrieved but did not cite. After adjusting for Domain Rating and page type, a publish or update date within the past 12 months was associated with 88% higher odds of citation, tables with 69% higher odds and first-hand experience with 45% higher odds.

The practical steps that follow are familiar to anyone who has worked on featured snippet optimisation:

• open each key page with a clear, direct answer to the question it targets;
• use tables for comparisons, prices and specifications;
• show genuine first-hand experience, with specifics only you could know;
• keep important pages current and show when they were last updated;
• keep ranking well in classic organic search, because citation stability follows it.

If you want to know whether AI assistants mention your business at all, our free AI visibility check asks ChatGPT the kinds of questions your customers ask and shows which businesses it names.

What to watch over the next few months

Three things would change this advice, and we are tracking all of them.

Click data in Search Console. Google has said it will add metrics to the generative AI reporting over time. Once clicks from AI features are measurable per page, the opt-out decision becomes a straightforward comparison rather than an estimate. Until then, the safest assumption is that the first published test is directionally right: AI citations and organic clicks support each other.

Larger opt-out studies. One site's results can be distorted by seasonality, a single article going viral or an unrelated ranking change. Results from dozens of sites across sectors would carry far more weight, and we expect them as more publishers experiment with the control.

How AI answers link out. Google changed the way AI Mode and AI Overviews display links again on 15 October, adding more inline links. Each change of this kind shifts how much traffic a citation delivers. If AI answers start sending meaningfully more clicks, staying in becomes even more clearly the right call; if they send fewer, the case for selective opt-outs strengthens for content-led businesses.

Key takeaways

• Google's Search generative AI control removes a site from AI Overviews, AI Mode and Discover's AI features; Google says it does not affect regular ranking or AI training.
• The first public opt-out test, a single small case study, estimated a roughly 0.9% CTR on AI impressions and found that regular organic clicks also fell after opting out.
• AI training controls (Google-Extended, the shared IETF signal planned for 2027, Bing's new options) are a separate decision from AI search visibility.
• For most businesses, staying in and earning citations is the better default; where opting out is tempting, test it on one section first.
• Click data for AI features is still missing from Search Console, so annotate every change and judge the result on your own clicks.