Google AI Overviews opt-out: should your website use it?

For most lead-generation, ecommerce, local and service businesses, AIWIZ’s view is that immediate sitewide exclusion from Google’s generative AI search features is unlikely to be the best default. Keep your site eligible while you establish a baseline and assess what those appearances are worth to the business. The Google AI Overviews opt-out also covers AI Mode and generative AI features in Google Discover, so the decision reaches beyond one type of search result.

Publishers, research organisations and businesses selling original or licensed content may reasonably decide otherwise. Control over that work can matter more than another route to the website, even when the financial effect is uncertain.

This guide is for website owners. The setting does not hide AI Overviews from your own searches: it is not a browser, account or privacy setting, and it is not a copyright remedy. If you want fewer AI answers when you search, that is a different question from whether your site should appear in them.

What is Google’s Search generative AI control?

The Search generative AI control is a Google Search Console setting that determines whether your site’s links and content can appear in specified generative AI search features and help ground their answers. Grounding means using your pages as source material for the answer.

Google’s documentation lists three choices:

• Include. The default for all properties. Google says your content “can appear in Search generative AI features, including showing up as links and helping to ground AI responses”. Inclusion permits participation; it does not guarantee that Google will select your pages.
• Exclude. Your content “is prevented from being visible to users in Search generative AI features, including being linked to within these features”. Google adds: “You won’t receive any traffic or impressions from these features.”
• Inherit control from parent. The property follows the setting of its parent property. Its effective setting may therefore be exclusion even if nobody changed it directly.

The control covers AI Overviews, AI Mode and generative AI features in Google Discover together. There is no documented option to exclude AI Overviews while remaining in AI Mode, or the reverse. Exclusion removes your links and content from those features; it does not remove the features. The answers continue, and other websites can still supply information about the same subjects, or about your brand.

How the control reached UK websites

Google announced the control on 3 June 2026, starting with “a subset of website owners in the UK”, and said it had been engaging with regulators including the Competition and Markets Authority. A note dated 31 August 2026 records that the control and the related reporting reached all websites worldwide on that date.

The regulatory background explains the UK start. On the same day as Google’s announcement, the CMA set out requirements covering publisher control over the use of their content in AI search features, clear attribution through links in AI-generated results and, following consultation, an opt-out from the use of content for AI model fine-tuning. The CMA gave Google nine months to implement all of the changes while expecting the publisher controls sooner, and requires compliance reports every six months for the first year. None of that sets a release date for a dedicated AI click report in Search Console.

One difference in wording remains unresolved. Google’s announcement describes the features as “AI Overviews, AI Mode or AI Overviews in Discover”, while the help page lists “generative AI features in Google Discover” more broadly. Neither explains the relationship between those terms. For decisions about scope, this guide follows the control’s documentation.

How to opt out in Google Search Console

The setting sits under Settings > Search generative AI for the relevant property. Before changing it, record the existing configuration and export the performance data you can access, because a baseline taken afterwards is worth much less.

1. Open the intended property in Google Search Console.
2. Check which domain, subdomain or URL prefix it covers.
3. Open Settings > Search generative AI and note whether the current value is set directly or inherited.
4. Select the exclusion option, then confirm the resulting effective setting.
5. Record the date and allow time for the change to take effect.

For a child property, confirm you have owner access and check the parent hierarchy first. A domain-level change can affect far more of the website than one section requires.

Google’s timing guidance says a change “generally takes a few days”. The same page says content “will be excluded within 1-2 days after the control goes live, but some content may take longer due to caching”. Neither statement is a guaranteed completion time, so do not rely on the setting for urgent removal of sensitive material.

Does opting out affect rankings, training or ads?

Google states that the control “isn’t used as a ranking or inclusion signal affecting other parts of Search”. That is Google’s description of its own systems. It is not a guarantee that your overall organic traffic will be unchanged, because the features you leave still compete for the same searchers’ attention.

Other controls answer different questions, and confusing them is a common and costly mistake:

• Google-Extended is a robots.txt token that governs whether crawled content may be used for training future Gemini models and for grounding in Gemini Apps and Vertex AI. Google’s crawler documentation says it “does not impact a site’s inclusion in Google Search”. The Search generative AI control does not govern training; Google’s help page points to Google-Extended for that.
• Googlebot rules in robots.txt manage crawling. Blocking Googlebot from a page affects ordinary Search as well, so it is not a way to leave AI features while staying in normal results.
• noindex tells Google not to index a page. Google must crawl the page to see the rule, so the page must not be blocked in robots.txt, and Google’s noindex guidance warns that a revisit can take months for a less important page. It is not an instant removal mechanism.
• Snippet controls, including nosnippet, data-nosnippet and max-snippet, limit the previews and content Google may use. Google’s AI features guidance presents them as the way to limit what is shown from your pages in Search generally, so their effects are not confined to AI answers.

Commercial participation is a separate boundary. Google says exclusion “doesn’t override publishers’ other choices to participate in particular services, such as Merchant Center and Google Ads”. That statement concerns your participation choices in those services. It should not be expanded into a promise that every commercial appearance or arrangement is unaffected.

What the AI performance reports show

Google provides separate generative AI performance reports for Search and for Discover. Both document impressions only, and their counting rules differ.

The Search report covers AI Overviews and AI Mode, with breakdowns by page, country, device and date. Without a URL filter, the chart aggregates by property, so two links from your site within one feature count as a single impression. Rows grouped by page aggregate by page, which is why their total can differ from the chart. Applying a URL filter changes the chart to URL-level aggregation.

The Discover report offers pages, countries and dates, with no device split. An impression counts once per result per session after the link scrolls into view, and all data is aggregated by page, so two results from the same property in one Discover list count separately.

Both are views of activity already included in the overall performance reporting. Google’s report launch post says the data “is included in the overall performance report”, and the Search report’s help page says it includes data from the Web search type. Adding AI impressions to overall totals would double-count.

Google’s help pages give three reasons a report may not appear: the property may not yet have access during rollout, the site may have too few impressions in generative AI features, or the site may be excluded. The first reason sits alongside the same pages’ note that the reports reached all websites worldwide on 31 August 2026. Treat that as inconsistent documentation rather than a diagnosis of your account. Check the effective inclusion setting before reading anything into an empty report.

Google’s announcement says it will “introduce additional metrics over time”, and the launch post mentions the same intention. No date for a separate AI click metric has been published.

What the reports cannot tell you about business value

The dedicated AI reports cannot isolate the clicks, enquiries or revenue that follow from your AI appearances. Activity from AI features is included in the Web search data of the standard Search results performance report, which does show clicks, but that report does not separate the AI portion out.

Keep those limits in mind when combining sources. Dividing all organic clicks by AI impressions does not produce an AI click-through rate, because the two figures describe different sets of activity.

Analytics and customer records can show organic visits, qualified enquiries and sales. Placing them alongside AI impressions does not establish which outcomes the appearances caused. Someone might encounter your business in an AI answer and return later through a branded search or a direct visit. Google’s reports cannot demonstrate that connection, and neither can a correlation over a few weeks.

Even a publisher with far more data than most businesses has reached the same conclusion. Phil Andraos, general manager of Reuters Digital, told Digiday on 8 June 2026: “There’s not enough data. If there was, we could make that decision very quickly. Instead, it’s going to be a more careful evaluation, probably testing as well.” That is one publisher’s stated position, not evidence that every business faces the same trade-off, but the response it describes, careful evaluation and testing, is the one this guide recommends.

What independent research adds

A preprint by Wang, Gleason and colleagues, submitted to arXiv on 18 August 2026, reports a preregistered field experiment on Google Search. Of 1,444 people recruited, 1,100 completed the study: three baseline days followed by seven days in one of three assigned search experiences. One group had AI Overviews and AI Mode hidden, one used Google unchanged, and one had every search redirected to AI Mode.

Two headline numbers are easy to misread. The authors estimate that removing AI features raised click-through to external sites by 8.8 percentage points. That figure is adjusted for the fact that a Google interface change during the study meant only about half of AI Overviews were successfully hidden; the unadjusted effect of simply being assigned to the no-AI group was 2.7 points and not statistically significant. Separately, being assigned to AI Mode for every search reduced click-through by 18.8 percentage points compared with unchanged Search, and that estimate needed no such adjustment. The two numbers answer different questions and are not two forecasts for one website.

The sample skewed younger and more highly educated, and was recruited mainly through an online research panel. The lead author’s website lists the paper as under review at PNAS Nexus; at the time of writing it has not been published in a peer-reviewed journal.

The experiment changed what searchers saw. A website opting out changes only whether its own pages can appear: the AI answers remain, drawing on other sources. The study therefore says something about how AI features affect clicks in general, and nothing about whether excluding your site would recover any of them.

Staying included versus excluding: how the options compare

Start by deciding whether your priority is commercial performance or a restriction on how your content is used. Either can justify a decision, but they call for different evidence.

Look at the pages receiving AI impressions alongside their role in the business. A high-volume guide that rarely attracts suitable customers may deserve less weight than a specialist page that supports valuable enquiries. That is a reason to assess lead quality and content purpose, not to assume that either page’s AI impressions caused its results.

For original or licensed material, write down what you want to restrict and why. Production cost, subscriber value and the consequences of misleading summaries can matter even when attribution is incomplete. Keep those judgements separate from any claim of a measured financial return.

Recommendations for different business models

AIWIZ recommends choosing according to how content supports the business, rather than applying one rule to every website. These are starting points for that judgement.

Lead-generation and service businesses

If content helps prospects find and assess your services, exclusion needs a reason beyond frustration with AI search. Examine the pages already receiving AI impressions and the quality of the enquiries those topics generate, while recognising the attribution gap. Our default is to stay included, take baselines and review on an agreed date.

Ecommerce and local businesses

Assess editorial content separately from product and local-business data. Google’s optimisation guide describes Merchant Center feeds and Google Business Profiles as ways for products and services to be visible in AI responses as well as other results. Do not assume that excluding website content withdraws every feed, profile or advertising contribution, and do not assume it leaves them untouched either. The outcome for a particular business needs checking against the services it uses.

Publishers and advertising-funded websites

When page views fund journalism, compare referral trends and advertising contribution with the cost of producing the content. Exclusion can express a publisher’s content policy, but removing one source does not remove the AI answer. Before expecting traffic to recover, identify the mechanism by which readers would return to your pages.

Subscription, research and licensed-content businesses

Review whether publicly available material overlaps with what customers pay to access or license. Exclusion may support your content-use policy, while free previews may still play an acquisition role. The setting does not create a licensing agreement or guarantee better negotiating terms.

Regulated or reputation-sensitive organisations

Identify the specific content that could cause harm if summarised without its qualifications. Record the risk and who owns the decision. A section-level response may be proportionate, but the setting cannot prevent third-party discussion or guarantee that an AI answer will stop mentioning your organisation.

How to test exclusion on one section

A URL-prefix child property can carry its own setting. For example, https://example.com/research/ could cover a research library while other sections remain included. Check the effective parent setting and any more specific overrides before assuming the section behaves uniformly.

For a commercial test, AIWIZ recommends these eight steps. The order matters: the criteria come before the change.

1. Define the decision. Identify the business outcome you want to assess and who owns the test.
2. Record the configuration. Map test URLs, property boundaries, inherited settings and manual overrides.
3. Save a baseline. Export available AI and standard performance data, alongside matching analytics and business records.
4. Choose the comparison. Use a similar, unchanged section where feasible. Record the differences and allow for seasonality and the sales cycle.
5. Agree the criteria in advance. Set the review date, the acceptable downside and the conditions for extending, reversing or declaring the test inconclusive.
6. Make and log the change. Record the date, the resulting effective setting and the propagation period. Avoid other major changes where practical; log any that cannot wait.
7. Monitor separate outcomes. Track AI impressions, organic performance and business results without treating one as a proxy for another.
8. Review the evidence. Examine the comparison pages and rival explanations before deciding whether to retain or reverse the exclusion.

Match reporting periods, filters and time zones before comparing exports; the Search Console reports use Pacific time. Do not draw conclusions from incomplete recent data or a handful of enquiries. A busy shop and a firm that wins a few large contracts a month need very different observation periods.

Lower AI impressions after the change are consistent with exclusion taking effect, but a decline alone does not prove it worked. Confirm the effective setting and allow for reporting delays. An improvement in revenue needs the same scrutiny: a promotion, seasonal demand or one large order could explain it.

If the evidence is too thin to separate a real effect from noise, record the result as inconclusive. You can still make a content-policy decision on other grounds, but do not describe it as a successful traffic experiment.

AIWIZ’s recommendation

For most commercial websites, remain included while you establish a baseline and assess the content at stake. Set a review date rather than waiting indefinitely for a perfect attribution report. Where content is the product, or loss of context presents a material risk, a documented decision to exclude can be reasonable.

Search Console gives you useful evidence, but it cannot settle the commercial question on its own. Combine the available data with a clear account of what the business gains or gives up by participating.

For eligibility checks, read why Google rankings do not guarantee AI search visibility.

Decision checklist

• Confirm who owns each Search Console property and can change its setting
• Record each property’s effective setting and whether it is inherited or set directly
• Export available baselines from the Search and Discover AI reports
• Identify the pages and countries receiving AI impressions
• Gather comparable analytics, lead-quality and conversion evidence
• Identify sensitive, licensable or paid-product content
• Check whether the property structure supports a section-level test
• Agree success, failure, reversal and inconclusive-result criteria before acting
• Record the implementation date and any other changes during the test
• Set a review date and assign responsibility for the decision