Google's AI opt-out is live for every site. Should you use it?

Since 31 August 2026 every site in Search Console has a switch that removes it from AI Overviews, AI Mode and Discover's generative features. The switch is easy. Knowing whether to touch it is not, because Google shipped the control and the report at the same time and the report deliberately does not answer the question the control raises.

What actually shipped, and when

Two things landed in Search Console on 31 August 2026, and it helps to keep them separate because they answer different questions.

The first is the Search AI controls: a property-level setting with three options — Include, which is the default, Exclude, and Inherit from parent, which lets a child property follow the choice made on the property above it. Setting a property to Exclude keeps its pages out of AI Overviews, AI Mode and the generative AI features now appearing in Google Discover. It removes you from those surfaces both as a cited link and as grounding material, which is the part most people miss: you are not just giving up the little link card, you are asking Google not to read your page when it builds the answer.

The second is the Generative AI performance reports, a dedicated view of how your pages appear inside those same AI features. This is a reporting change, not a traffic change. That data was already flowing into your overall Performance report; the new view isolates it.

The reports have one significant limitation that shapes the entire decision below. They show impressions broken down by page, country, device and date. They do not show queries, clicks, click-through rate or average position. So Search Console will tell you whether you are visible in AI surfaces. It will not tell you what that visibility is worth.

What Exclude does not cover

Most of the confusion about this setting comes from people assuming it is a general-purpose "keep my content out of AI" switch. It is not. It is narrowly scoped to Google's generative features inside Search and Discover, and every other AI surface is governed by something else entirely.

Google has been clear that the control is not used as a ranking signal for results outside those generative features, and that standard Search, Merchant Center and Ads participation are untouched. That removes the single biggest fear people have about the setting. It does not make the decision easy, because the thing you are actually trading away is harder to measure than a ranking drop.

How much traffic would excluding actually cost you?

You cannot read this number directly, because the generative AI report has no click column. But you can bracket it, and a bracket is enough to make the decision in most cases.

Pull three numbers for the same date range and the same property, filtered to Search:

1. Generative AI impressions (Iai) from the new report.
2. Total impressions (Itotal) from the main Performance report.
3. Total clicks (Ctotal) from the main Performance report.

Because generative impressions are already counted inside the totals, the honest range for what Exclude would cost you is:

A worked example. Say a 40-page B2B services site records 180,000 total Search impressions and 4,200 clicks over 90 days, with 27,000 of those impressions in the generative AI report. The generative share is 15%. The ceiling is 4,200 × 0.15 = 630 clicks over 90 days, or about seven a day. The floor is zero.

Now apply the test that actually decides it: at your real close rate, is the ceiling worth arguing about? If seven clicks a day at a 2% enquiry rate is one enquiry a week and your average project is worth five figures, you are debating whether to voluntarily switch off a lead source to protect a page-view number. If the ceiling is four clicks a month, the whole question is noise and you should spend the afternoon on something else.

The bracket is deliberately crude. It exists to stop the two failure modes we see most: teams who assume AI surfaces are costing them everything without checking the impression share, and teams who assume the traffic is enormous and refuse to look at the ceiling.

If you want a real number, run a staged exclusion

Because the setting works at property level and supports Inherit from parent, a site with verified child properties can test rather than guess. This is the part almost nobody does, and it is the only way to get a click figure out of a surface that does not report clicks.

1. Verify a child property for a section of the site that has its own traffic and is not your primary money path — a blog subfolder or a resources subdomain, not your service pages.
2. Baseline it for at least eight weeks in the main Performance report: clicks, impressions and average position for that folder alone, plus its generative AI impressions.
3. Set that child property to Exclude and leave the parent on Include.
4. Read the difference in the classic Performance report, not the generative one. Total clicks to the excluded folder against the same folder's prior trend, with the rest of the site as your control for seasonality and core updates.
5. Attribute carefully. If sitewide clicks moved the same way, you measured a Google-wide event, not your change.

Two warnings. First, Google has not published how quickly an Exclude takes effect, or how quickly visibility returns if you switch back to Include. Do not design this as a two-week experiment and do not run it on your highest-value templates. Second, run it on a section whose content type matches the pages you actually care about. Excluding a news archive tells you very little about what would happen to your pricing pages.

The decision, by site type

The right answer is genuinely different depending on how your site makes money, because the trade is always the same — you are exchanging discovery for extraction — and different business models value the two sides very differently.

The option nobody mentions: it is not a binary

Include and Exclude are not your only settings. Google's existing preview controls also apply to AI features, which gives you a middle path that most coverage of this rollout skips entirely.

• max-snippet:[number] caps how much text can be used, which limits extraction without removing you from the surface.
• data-nosnippet marks specific elements on a page as off-limits while leaving the rest usable — useful for the one proprietary table you do not want lifted wholesale.
• nosnippet removes text previews entirely, but it also removes your regular search snippet, which is usually a worse trade than the thing you were trying to avoid.

The practical sequence: use data-nosnippet on genuinely proprietary blocks, consider max-snippet on templates where full extraction is the specific worry, and reserve the property-level Exclude for the case where the whole content type is the product.

What to do this month

1. Open the generative AI performance report and record your 90-day impression share. You now have a baseline you did not have in August, and it is worth having whichever way you decide.
2. Run the bracket. One division. Write the ceiling number down and put it in front of whoever is asking for the opt-out.
3. Check your robots.txt separately. If the actual concern is model training rather than Search, Google-Extended and the other vendors' tokens are the controls that matter, and the Search Console setting does nothing for it.
4. Audit your child properties. If you have never verified subfolders, do it now. It is what makes both the reporting and the staged test possible.
5. Leave the setting on Include unless the bracket says otherwise. The default is not an accident, and the reversal timeline is undocumented.

Frequently asked questions

What are Google's Search AI controls in Search Console?

They are a property-level setting in Google Search Console, rolled out to all sites worldwide on 31 August 2026, with three options: Include (the default), Exclude, and Inherit from parent. Setting a property to Exclude keeps its pages out of AI Overviews, AI Mode and the generative AI features in Google Discover, both as a cited link and as grounding material used to build the answer.

Does opting out of AI Overviews hurt my Google rankings?

No. Google states that the control is not used as a ranking signal for search results outside those generative AI features, and that standard Search ranking, Merchant Center and Google Ads participation are unaffected. What you lose is visibility inside AI Overviews, AI Mode and generative Discover, along with the impressions and any clicks those surfaces were sending.

How do I work out how much traffic excluding my site would cost?

Search Console does not report clicks for AI surfaces, so bracket it instead. Take total Search clicks and total Search impressions from the main Performance report, and generative AI impressions from the new report, over the same period. The floor is zero clicks. The ceiling is total clicks multiplied by generative impressions divided by total impressions. For example, 4,200 clicks with a 15% generative impression share gives a ceiling of 630 clicks.

Does the Search Console AI setting stop ChatGPT or Gemini from using my content?

No. The setting governs three Google surfaces only: AI Overviews, AI Mode and generative features in Discover. The Gemini app and Vertex AI grounding are controlled separately through the Google-Extended token in robots.txt, and ChatGPT, Perplexity, Copilot and other assistants each use their own crawler tokens. Excluding in Search Console does nothing to any of them.

What does the Search Console generative AI performance report show?

It shows impressions inside AI Overviews, AI Mode and generative features in Discover, broken down by page, country, device and date. It does not show queries, clicks, click-through rate or average position. That data was already included in your overall Performance report totals; the new view isolates it, so it adds visibility rather than traffic.

Is there a middle option between Include and Exclude?

Yes. Google's preview controls also apply to AI features. Use data-nosnippet on specific proprietary elements to keep them out of extraction while leaving the page usable, or max-snippet to cap how much text can be used. Avoid a blanket nosnippet directive, because it also removes your ordinary search snippet, which usually costs more than the extraction you were trying to prevent.

The takeaway

The control is a real piece of publisher power and it was worth shipping. It is also the wrong first question for most businesses. Excluding can only ever subtract visibility; it cannot add a visit, and it does not touch the assistants most people are actually worried about. Pull your impression share, run the one division, and put a number on the ceiling before anyone argues from instinct. Then go and spend the time on the version of this problem that has upside: making the pages clear, specific and structured enough that an answer engine has a reason to name you rather than the competitor who did that work first.