Four numbers get quoted in almost every discussion of AI Overviews: a 34.5% drop, a 58% drop, a 47% drop, a 61% drop. They are all real. They also measure four different things, and if you pick the wrong one you will build the wrong strategy.
The short version: clicks per search have fallen sharply on queries where Google shows an AI answer, the decline appears to have stabilised in early 2026, and the single most actionable finding in the whole body of research has nothing to do with the size of the drop. It is that being cited inside the AI answer changes your outcome dramatically.
Sorting the Four Numbers
Each study answers a different question. Lining them up side by side is where most reporting goes wrong.
| Source | What it measured | Finding |
|---|---|---|
| Ahrefs, April 2025 | CTR on the top-ranking page for keywords triggering an AI Overview, across 300,000 keywords | 34.5% reduction |
| Ahrefs, February 2026 | Same method, extended to December 2025 data | 58% reduction |
| Pew Research Center | Real browsing behaviour: share of visits where a user clicked any traditional result | 8% with an AI summary vs 15% without |
| Seer Interactive | Brand-level Search Console data across dozens of brands | Organic CTR on AI Overview queries fell from 1.76% to 0.61% |
| Randomised field experiment, 2026 | Causal effect on outbound clicks | 39.8% drop |
The distinctions matter. Ahrefs measures one page’s click rate on one class of keyword. Pew measures whether a human clicked anything at all during a session. Seer tracks a longer time series at the account level. None of them is measuring total site traffic, which is what most people assume when they hear “traffic is down 58%.”
One number does deserve extra weight. The 2026 randomised field experiment is the first causal evidence in this debate rather than correlation. Its finding, a 39.8% drop in outbound clicks, sits in the middle of the range, which is roughly where you would expect a well-controlled estimate to land.
If you want a single working assumption, use that one. If you want to understand what your own numbers are telling you, our explainer on what click-through rate actually measures is the right starting point, because the studies above are all CTR studies and CTR behaves differently from sessions.
The Stabilisation Nobody Reports
Coverage tends to freeze at the worst moment in the data. Seer’s tracking shows organic CTR on AI Overview queries bottoming out around 1.3% in December 2025 and recovering to roughly 2.4% by February 2026.
That is still well below pre-AI Overview levels. But a floor followed by a partial recovery is a different strategic picture from a straight line heading to zero. Planning as if clicks are about to disappear entirely is not supported by the most recent data.
Meanwhile, zero-click behaviour has been rising for reasons that predate AI. SparkToro’s analysis found the share of searches producing at least one click fell about 9.5 percentage points between 2024 and 2026, while the share of searches leading to another Google search rose. Google answering questions in place is part of that. So is Google keeping users inside Google.
The Finding That Should Change Your Plan
Buried in Seer’s 2026 analysis is the number that matters most for anyone making decisions: brands cited inside an AI Overview earn roughly 120% more organic clicks per impression than uncited brands on the same query.
Read that carefully. Same query, same AI answer present. The variable is whether you are in it.
That reframes the whole problem. The question is not how to survive AI Overviews. It is how to be the source they quote. Which is a familiar problem wearing new clothes: it is what featured snippets asked of publishers a decade ago, at higher stakes.
Two related data points sharpen the target. Surfer SEO’s analysis of 46 million citations found YouTube accounted for roughly 23% of AI Overview citations and Wikipedia for about 18%. And Authoritas found roughly 70% of the pages cited in AI Overviews changed over a two- to three-month window, with those changes not tracking traditional organic rankings.
Citation is volatile and only loosely coupled to your blue-link position. Ranking first does not guarantee being cited, and being cited today does not mean being cited next quarter.
The UK Just Gave Publishers a Real Lever
On June 3, 2026, the UK’s Competition and Markets Authority imposed a conduct requirement on Google under the Digital Markets, Competition and Consumers Act 2024. It followed the CMA’s designation of Google with strategic market status in general search.
The requirement has three planks that matter commercially:
- Publishers get a granular control over whether their content is used in AI features including AI Overviews and AI Mode, without losing their place in ordinary search results.
- Publishers can separately opt out of their content being used for fine-tuning AI models.
- Google must attribute publisher content properly, with clear links, when it appears in AI-generated answers.
The CMA called it a world first. Its stated purpose is to strengthen publishers’ hand in negotiating content deals. Google’s position has consistently been that its AI features are driving search growth, a claim publishers and several independent studies dispute. Main obligations are expected to take effect from December 2026.
Before treating the opt-out as an obvious win, think about it as a collective action problem. Opting out means forgoing AI Overview traffic while competitors who stay in occupy the answer space alone. The right decision depends on what your market does, not only on what you do.
What Actually Works Now
Five things follow from the evidence rather than from vibes.
- Separate your query types before you panic. AI Overviews cluster on informational queries. Measure the CTR change on your commercial and navigational terms separately, because the damage is uneven and your averages are hiding it.
- Optimise for citation, not just position. Clear, extractable, factually specific passages. Original data. Direct answers near the top of the page. The pages that get quoted answer a question cleanly in a way a model can lift without ambiguity.
- Stop treating search as your only distribution. ChatGPT referrals to publishers grew substantially but still account for under 1% of referrals. Chatbots are not replacing search traffic. Email, direct, and community are.
- Build things an AI answer cannot substitute for. Tools, calculators, comparison data, ongoing coverage. If your page can be fully summarised in three sentences, it will be.
- Watch impressions, not just clicks. Visibility inside an AI Overview may not send a click today and may still build brand recall. Judging AI-era SEO on clicks alone under-measures it.
None of that is a trick. It is the same discipline that always separated content worth citing from content that merely ranked, applied under harsher conditions. Our older piece on growing traffic without paid ads holds up on the fundamentals, though the search chapter of that playbook now needs the caveats above.
Update, September 2026: The Same Pattern Is Hitting Advertisers
The automation shift described above isn’t limited to organic search. Google began automatically upgrading Search campaigns to a new AI-driven optimization layer called AI Max starting September 1, 2026 — expanding keyword matching and ad-asset generation with less advertiser control over the specifics, the paid-search equivalent of the visibility trade-offs publishers are already navigating with AI Overviews.
Where This Stands
Information here is current as of September 5, 2026. The research base is moving quickly, methodologies differ, and Google has not published auditable click data of its own, which is precisely the gap the CMA’s intervention is designed to close. Treat every figure above as a measurement of a specific thing, not a universal truth about your traffic.
The most honest summary is uncomfortable but useful. Nobody outside Google knows the true aggregate number, the range of credible estimates is wide, and the one lever you fully control is whether your content is worth quoting.
