On August 28, 2026 Google began expanding AI Overviews for selected queries into full-page, AI Mode-style answers. The collapsed snippet with its "Show more" gate is gone; in its place is an expanded answer with an open "Ask anything" follow-up box, and the first organic result now sits below the fold on many screens. Independent measurement is blunt about what this does: on queries that trigger an AI Overview, Seer Interactive measured organic click-through down 65% and paid click-through down 68%, and Ahrefs found a position-one page 58% less likely to be clicked (January 2026 data). A randomized field experiment from Carnegie Mellon and the Indian School of Business put the outbound-click drop at 39.8% under the older, more conservative design — which the analysis treats as a floor, not a ceiling, for the expanded version. All of this is current as of September 2026, and the numbers are attributed inline so you can recheck them.
The useful response is not to report the loss. It is to change what you measure and what you build, because the same data contains the counter-move: brands that are cited inside the overview receive 35% more organic clicks and 91% more paid clicks than brands left out. Visibility in the overview is not a death sentence; being absent from it is. This guide walks the seven steps that turn that fact into a repeatable practice.
Step 1: Decide what stays citable — and what steps out
For every page that currently earns traffic, make an explicit keep-or-exit call. Google's Search Console AI control toggle lets site owners remove content from generative sourcing without affecting standard rankings, and the off switch is the wrong default. Citable pages are the pages that want the exposure — the ones that answer a category question clearly and can carry a strong call to action. Pages whose traffic is worth more as a direct visit than as a citation — a pricing page that you want the user to reach by clicking, a rent-vs-buy calculator you use as a lead magnet — are candidates for the toggle. The decision, not the toggle, is the deliverable here.
Step 2: Diagnose which queries are actually hurting you
You cannot defend what you have not measured. Use the Search Console generative AI report, rolled out to all sites worldwide on August 31, 2026, for impressions by page, country, device, and date across AI Overviews and AI Mode. Read its limit honestly: it reports impressions but gives no clicks, no queries, and no position, and it is not exposed through the Search Console API. So pair it with your own channel tagging — build a dedicated "AI search visibility" channel group in GA4 so clicks that arrive from AI answers are separated from conventional search clicks, and snapshot the same answer queries monthly. Without dated snapshots you have anecdotes, not a baseline.
Step 3: Build the pages that engines cite
An AI answer does not rank a page, it cites one, and it cites the pages it can quote. That means a page that states a verifiable claim in its first paragraph, carries real data and named definitions, and covers one entity or one question is far more likely to be attached as a source than a homepage with a slogan. Our own research on AI citation sources keeps confirming the same pattern: the engines lean on a narrow, uneven citation base, and a comparison page, a data-backed explainer, or a clearly scoped service page earns the citation more reliably than a broad overview. Build the page that the engine wants to quote, and the answer starts to grow around it.
Step 4: Convert the citations you win
Cited traffic is different from ranked traffic — it is usually more intentionally qualified, because the user asked a specific question and the engine answered it, and your page was a named source. Watch the cited pages for conversion, not just reach, and strengthen their call to action. When a page starts appearing in answers, its job shifts from "get the click" to "turn the citation into the obvious next move." Measure the uplift on your referenced pages the same way the research measures it: a cited brand's advantage over a non-cited one is the number that tells you whether the work is paying.
Step 5: Re-base the KPI from clicks to visibility
The single biggest change is not tactical, it is what you report upward. A click target built for ten-blue-links is now a target for a surface where much of the reading happens inside the answer. Move the north-star metric to visibility + citation: how many of your money queries produce answers that name you, and how often. That is the metric that survived the feature launch, and it is the one a leader can plan against. Impressions are a volume signal, but the citation is the signal that generates the 35% and 91% uplifts.
Step 6: Build the buffer that absorbs a click loss
The same evidence stack shows why some publishers absorb an 80-90% click-through crash with almost no total-traffic impact: they have strong direct and branded traffic, so they are not renting every visit from the search box. This is the diversification lesson, and it is the most durable move in the playbook. Brand the product, capture email, build direct habits, and keep your strongest branded queries healthy. A business that owns its audience can take the hit; a business that rents all of it cannot.
Step 7: Measure monthly and iterate
Treat AI search as a permanent layer, not a temporary wave. Run the same five money queries monthly, compare cited-versus-absent pages, and diff the Search Console impressions and GA4 channel performance against the prior period. Two dated snapshots are the beginning of a measurement system; one screenshot is a hunch. As the engines keep changing the surface under you, the only durable asset is a monthly habit of asking what they say about you — and a site made of pages they want to cite.
The honest limits
Two caveats keep this honest. First, the engine is still early in the change — Google has not disclosed which queries expand, what share of searches is affected, or whether the rollout widens, so treat the shock tests as a floor and expect the surface to keep moving. Second, most of the click-loss figures come from working papers and vendor studies rather than peer-reviewed field experiments; the one true randomized experiment, the Agarwal and Sen working paper, is strong on causation but is not yet peer-reviewed. Use the numbers as directional and keep your own dated evidence as the source of truth. What is not in doubt is the direction: visibility is becoming citation, and clicks are becoming the page you build to convert the ones you still get.
FAQ
Are AI Overviews really taking clicks, or is this overblown?
The direction is well documented. Seer Interactive measured a 65% organic CTR drop on queries with an AI Overview, Ahrefs found a 58% lower position-one CTR, and a randomized field experiment from Carnegie Mellon and the Indian School of Business measured a 39.8% drop on affected queries — treating that as a floor for the expanded design. The magnitude varies, but the mechanism is not in doubt.
Should I opt my site out of generative sourcing?
Only the pages you want held back. The toggle removes content from generative features without changing standard ranking, so use it deliberately for pages that are worth a direct click, and leave the pages that answer a asked-for question eligible to be cited.
How do I measure AI search visibility without a paid tool?
Ask your five money queries monthly, save the answers with dates, and sort yourself into cited-with-source, mentioned, or absent. Pair that with Search Console's AI performance report for impressions and a GA4 "AI search visibility" channel group for the clicks that do arrive. That costs an hour a month and produces the baseline every paid tool will sell back to you.
Related reading
- Google AI Overviews 2026: what changed and what it means — the feature history behind today's expansion
- AI Search Citation Sources by Industry — the citation base AI answers actually lean on
- Programmatic SEO in the Age of AI Search — building pages that get cited, not just ranked
Author: Elena Shaw, AI Search Visibility Analyst at Auspia. Elena tracks how answer engines change the way content gets discovered and cited.
