Every site owner on Google can now switch their site out of AI Overviews and AI Mode, and very few have thought about what that switch will be worth in three years. Google finished rolling the Search generative AI control out to every website worldwide on August 31, a few months after the UK Competition and Markets Authority imposed a conduct requirement obliging Google to offer exactly this kind of opt-out. The setting sits in Search Console under Settings. Google’s documentation says that once you exclude your site, its content is no longer eligible as an input to AI responses or linked from them, and that the choice is not used as a ranking signal anywhere else in Search.
Now picture the company that flips the switch. Today it looks like a contained decision, because you give up whatever AI Overviews and AI Mode send you while your organic listings stay exactly where they were. In my forecast last week, I put the window for the ranked results page to stop being Google’s default first response at 2028 to 2030, with the links one click away rather than gone. Run the opt-out forward into that window, and the company has removed itself from the surface that became the default, while keeping its place on the surface that moved behind a click. Suppose that costs another 20% of its traffic, a figure I am using to show the scale of the bet rather than one anyone has measured. Google’s position does not change either way, which is the whole point of a no-retaliation rule.
The evidence that does exist points the same direction. In Pew Research Center’s browsing analysis, users clicked a traditional result in 8% of visits when an AI summary appeared, against 15% when one did not. A preregistered field experiment by researchers at Penn and Northeastern found that routing every search into AI Mode for a week cut click-through to external sites by 18.8 percentage points compared with normal Google Search. That study forced the condition, ran before AI Mode carried ads, and used participants who had rarely touched AI Mode, so it is a directional signal about a default shift rather than a forecast of one.
Here is the change underneath all of it. Being indexed used to buy a visit, and once the answer layer becomes the default, being indexed buys only eligibility, the right to be considered when a system assembles an answer, with no promise that anyone arrives at your site.
The Index Stopped Being A Shelf And Became A Candidate Pool
For more than 20 years, the bargain was easy to read. A crawler took your pages, and the search engine sent people back. Getting indexed put you on a shelf a person would scan, ranking decided how close to eye level you sat, and the visit was the payoff, with everything else happening on your own property.
In the answer layer, indexing puts you into a candidate pool, meaning the set of documents a system can draw from before it decides what to say. The system reads across that pool, selects, synthesizes, and then decides whether a link is needed at all. Google’s own wording for the new control gives the game away, because the thing you surrender by opting out is eligibility to be used as an input. That is what the index now sells. The visit has become a second event that depends on whether the answer leaves the person with a reason to keep going.
I have written before that answer engines test familiar best practices earlier in the decision chain, and that still holds for the practices themselves. What has changed is the value of the indexed state. Being in the index used to get you most of the way to a visit, and it is becoming the first of several steps, any of which can end the journey before your site sees it.
Travel Agents Already Lived Through This
The closest precedent I know is the travel agent. The Bureau of Labor Statistics counted 61,500 travel agent jobs in 2025 and projects almost no change through 2035, and an analysis of BLS data put the drop between 2000 and 2021 at roughly 70%. The industry’s trade body has long argued that the federal count misses independent advisors, and ASTA works from a Census figure closer to 105,000. Whichever count you prefer, the volume left once travelers could book a flight themselves, and it did not come back.
What stayed was the work the self-service channel handled badly: complicated itineraries, luxury and adventure travel, group trips, and fixing things when a connection collapses mid-journey. Host Agency Reviews’ research shows most leisure agents now run their own businesses, and that adventure and luxury specialists out-earn generalists. Their clients arrive having already done the basic research, and they are paying for judgment rather than access.
The agents who made it did not try to win the routine booking back. They rebuilt around what the new channel could not do well, priced that work properly, and accepted a smaller, more committed client base. There is a second lesson in the same BLS page, which now lists AI tools among the reasons demand may stay flat. The industry that adapted once is being asked to adapt again, and the ones best placed are those who learned the first time that the job is whatever the automated channel cannot do.
What The Work Becomes
SEO does not die when the SERP stops being the default; success stops being measured by where you rank and starts being measured by whether a system selects you, which is harder to earn and harder to count. The work shifts in three ways.
The first is that content moves from capturing a query to settling a question. A page built to rank for a phrase can be summarized away in one sentence of an answer. A page that holds original data, a defined position, a clear claim with its qualifier, or an explanation nobody else has written is something a system struggles to answer well without. The test for every important page becomes whether an answer engine would produce a worse response if that page did not exist.
The second is that structure moves from keyword silos to decisions. Most sites are still organized the way crawlers and rankers rewarded, by topic and by phrase. Both the answer layer and the person who clicks through later in their thinking are better served by organization around the decisions people actually make, such as comparing options, qualifying a provider, choosing a configuration, or fixing a problem. Information architecture becomes the map that machines and people read at the same time, which is why it matters more now, not less.
The third is that the site’s job splits in two. Part of it becomes the source of record that answer engines draw from, which means consistent facts, clear entity information about who you are and what you offer, and pages that can be crawled and rendered without friction. Those are the entry fee, and they buy eligibility and nothing more. The other part becomes the conversion point for visitors who arrive having already read a synthesized answer. Those visitors skip the explainer. They want pricing, proof, comparisons, availability, and a way to reach a person. Most sites built their best real estate for the first-time visitor, and the answer layer now handles that visitor before they ever arrive.
Is The Traffic That Is Left Actually Better?
The comforting claim is that you will get less traffic but better traffic, and the evidence on that is split. Microsoft Clarity’s study of more than 1,200 publisher and news sites found AI referrals made up under 1% of traffic but converted to sign-ups at 1.66%, against 0.15% from search. A peer-reviewed paper in *Marketing Science* by Maximilian Kaiser and Christian Schulze, covering 973 ecommerce sites over twelve months, found ChatGPT referrals converted about 13% below organic search and beat only paid social, though the gap narrowed as the year went on. Different verticals, different definitions of conversion, and different time windows mean both results can be true at once.
So treat “higher quality” as a hypothesis to test on your own site, not a promise to plan around. The harder part is keeping two kinds of evidence apart. A referral session that converts is a counted number. The influence of an answer that named you without linking is an inferred one, and most dashboards blend the two without saying so. Separating them is the problem I spend most of my time on at CitationIQ, and whatever tools you use, make that separation before you base a decision on either.
So, About That Toggle
The opt-out is the first time anyone has been asked to put a price on being indexed for AI features. Google’s Generative AI performance report will show you impressions from AI Overviews and AI Mode by page, country, and date, which gives you a rough view of what you would give up today. It cannot show you what you would give up in 2029, because the surface that matters most then has not become the default yet. A blanket opt-out decided on this year’s numbers prices the choice against the surface that is leaving.
Two details change the reasoning. The control does not affect AI training, which Google handles separately through Google-Extended, so anyone flipping it to protect their content from model training gets neither the protection nor the traffic. And the CMA’s deadline for page-level controls falls in March 2027, which turns an all-or-nothing choice into a page-by-page one. Protecting a proprietary research page while leaving in the pages whose job is to be selected is a defensible use of that control. Pulling the whole site out of the surface your future customers will start on is a harder case to make.
The Window Is Open Now
The forecast range gives most businesses two to four years, and the slowest work is exactly what cannot be rushed in 2029. Rebuilding content around questions you can settle, reorganizing a site around decisions, and retooling measurement so counted and inferred numbers stay separate all take longer than a redesign cycle. Start with the pages that carry the most commercial weight and ask, honestly, whether a system could answer the buyer’s question just as well without them. If it could, that page has no reason to be selected, and no ranking will save it.
When the ranked list stops being the default, being indexed buys eligibility rather than a visit, and the businesses that keep earning from search will be the ones whose content a system cannot answer well without.
If you have already flipped the toggle, or decided not to, tell me how you decided that in the comments, because I would like to see the reasoning people are using.
More Resources:
- What Opting Out Of Google’s AI Search Features Means Now
- Google Gives Sites AI Search Opt-Out, But Not The Data To Use It
- Study Confirms Google AI Overviews Cut Organic Clicks 38%
This post was originally published on Duane Forrester Decodes.
Featured Image: Kues/Shutterstock
