Paid Search CTR Fell 68% Under AI Overviews. The Budget Case Is Weaker

Paid click-through rate on queries carrying an AI Overview fell from 19.70% to 6.34% between June 2024 and September 2025, a 68% decline. That figure came from 3,119 informational search terms. The same firm's larger study, covering 5.47 million queries, puts paid CTR on AI Overview queries at 15.48% in the same month. Both numbers are honest. Only one of them should touch your budget.

What the 68% figure actually measures

Seer Interactive published the number that everyone now quotes. It is worth reading the method before repeating the headline.

The September 2025 study tracked 3,119 search terms across 42 client organisations, covering 25.1 million organic impressions and 1.1 million paid impressions. The window ran from June 2024 through September 2025. On queries where an AI Overview appeared, paid CTR fell from 19.70% to 6.34%.

Three things about that sample matter and are usually dropped in retelling.

First, the tracked terms are informational. That is the query class where AI Overviews appear most and where paid ads have always converted worst. Applying an informational-query finding to a transactional budget is a category error.

Second, the control group fell too. On queries without an AI Overview, paid CTR fell 32% over the same window, from 19.1% to 13.04%. So roughly half the decline in the treatment group is explained by something happening across all of paid search, not by AI Overviews specifically.

Third, the denominator moved. CTR is clicks divided by impressions. If AI Overviews expand the set of queries that trigger an ad auction, impressions rise, and CTR falls even when click volume is flat or growing. A falling ratio is not evidence of falling clicks.

The same firm's larger sample does not reproduce it

In April 2026 Seer published an update on a much larger panel. It covered 53 brands, 5.47 million tracked queries and 296.9 million paid impressions, running from January 2025 to February 2026. That is roughly 270 times the paid impression volume of the earlier study.

On that panel, paid CTR on AI Overview queries did not collapse. It started 2025 at 14.64% and ended February 2026 at 16.21%. Seer describes paid behaviour in 2025 as linear and stable, and says paid is relatively insulated from the organic effects of AI Overviews.

Set the two studies against the same month and the gap is stark. For September 2025, the narrow study reads 6.34% and the wide study reads 15.48%.

Neither result is fabricated and neither firm is being careless. The two studies answer different questions, and the internet has been treating the narrow answer as the general one. If you have already moved budget on the strength of a 68% number, you moved on an informational-query finding.

The gap that survives both studies

Strip out the headline and one durable fact remains. In February 2026 paid CTR was 16.21% with an AI Overview present and 21.85% without one. That is a relative shortfall of about 26%, not 68%.

A 26% CTR shortfall on a subset of your keywords is a real planning input. It is also a fundamentally different decision from the one a 68% collapse implies.

Why Google's search revenue rose while CTR fell

If paid search were breaking, the seller's revenue line would show it first. It does not.

Alphabet's Form 10-Q for the quarter ended 30 June 2026 reports Google Search and other revenue of $63,271 million against $54,190 million a year earlier. That is growth of about 17%. Total Google advertising reached $81,629 million against $71,340 million.

Microsoft tells a similar story from a smaller base. Search advertising revenue excluding traffic acquisition costs grew 10% in the June 2026 quarter, following 12% growth in the March quarter. Two independent sellers of search ads are not reporting a demand shock.

Impressions are doing the work

Sundar Pichai told investors that AI Mode is driving an incremental increase in Search queries overall and that AI Mode passed 1 billion monthly active users after its October global rollout. More queries means more auctions and more impressions.

Hold clicks constant and add impressions, and CTR falls arithmetically. That is what a large part of the measured decline is. It is a denominator effect wearing the costume of a demand collapse.

Take the seller's framing with appropriate suspicion. Google has an obvious interest in the claim that AI features expand rather than cannibalise search. There is no independent audit of AI Mode query volume, and the "billions of clicks to websites" line is not reconcilable against any published dataset. Treat the revenue figures as verified and the causal story as the seller's.

The real break is eligibility, not click rate

Here is the change that should worry a paid team, and it is not in any CTR chart.

SE Ranking tested 50,032 ad-eligible commercial keywords across 20 niches on 30 June 2026. Text ads appeared in 29.45% of AI Mode responses. Roughly one in three ad-eligible responses carried an ad.

Ad density tracked commercial value closely. Keywords under $2 CPC showed ads 24.33% of the time. Keywords in the $2 to $10 band showed them 32.45% of the time. Keywords above $10 CPC showed them 53.56% of the time.

Placement now depends on campaign type. Eligibility for AI Mode placements runs through Performance Max and the newer AI Max for Search campaigns. If your account is built on manual keyword campaigns, you are not simply performing worse on this surface. You are absent from it.

That is an account architecture decision, and it is the one most paid teams have not made. It also hands more targeting control to the platform, which is a real cost and should be priced as one.

Ad slots and citation slots are almost disjoint

The same study found domain-level overlap between advertisers and cited sources of 11.53%. At exact URL level it was 1.95%. Buying the ad slot and being cited in the answer are close to independent outcomes.

This is the single most useful finding for budget purposes, because it kills the tidiest reallocation story. You cannot fund content work by cutting ads and expect the same surface coverage. They are different inventory. The mechanics of who gets cited are covered separately in the analysis of how AI citations broke away from Google rankings.

What actually breaks in a CAC model

Most acquisition models chain four assumptions. Impressions, click rate, conversion rate and average deal value. AI search damages one of them badly and leaves the others intact.

The conversion rate row is the weakest claim in this table. It rests on the absence of contrary evidence rather than on a study confirming stability, which is a materially lower standard.

The honest reframing is this. Paid search did not get more expensive because of AI Overviews. It got harder to forecast. Those two problems have different fixes, and the second one is not solved by moving money.

If your cost per acquisition has genuinely risen, check competitive density and bid inflation before blaming the answer box. Cross-industry CPC benchmarks are published by agencies with commercial interests and small, self-selected samples, so use your own account history as the baseline instead.

Measure the AIO-present segment separately

The one change worth making inside the model is segmentation. Split your keyword set by whether an AI Overview appears, then compute CTR, cost per click and cost per acquisition for each half.

If the AIO-present half is 26% worse on CTR and equal on conversion, that is a bid adjustment. If it is 60% worse, you have found something the published research has not, and it is specific to you. Either result is more useful than a headline drawn from someone else's keyword set.

Where this argument is weakest

This post argues against a reallocation that many competent teams are making. That deserves the counter-case stated properly.

Start with the wide panel, which has its own selection problem. Seer's 53 brands are agency clients. Agency clients are actively managed, which means their paid accounts were being optimised throughout the measurement window. A stable CTR line at a well-managed account can mask a declining underlying surface that the optimisation is offsetting.

Put plainly, the wide study may be measuring how good agencies are at holding a line rather than whether the line was under pressure. That would make the reassuring reading partly an artefact.

Revenue growth also does not prove buyer outcomes. Google's search revenue can rise while every individual advertiser's return falls, if impressions grow faster than efficiency drops. Seller revenue is a poor proxy for buyer return, and I have used it here as evidence against a demand collapse rather than as evidence of advertiser health.

Single-day snapshots age fast too. The SE Ranking density figures come from one day of collection. AI Mode responses vary between sessions and Google ships changes to this surface continuously. Treat 29.45% as a June 2026 reading, not a constant.

The case for moving money anyway

There is a serious argument I do not think this post defeats. If a growing share of buyer research now happens inside an answer that rarely carries your ad and rarely cites you, then the surface you are buying is shrinking regardless of what CTR does this quarter. Waiting for clean data on a surface that is being rebuilt is a way of arriving late. That case rests on a forecast, and so does mine.

Reallocation options, ranked by what they actually cost

Assume you accept that some shift is warranted. The options are not equivalent, and the cheapest one is usually skipped.

The order matters more than the amounts. Every row below the first depends on the first. You cannot size a bid adjustment or a budget shift without knowing what your own AIO-present segment does. Teams that skip the measurement step and go straight to the reallocation are picking a number from a blog post, which is how the 68% figure ended up in so many plans.

What I would do

I would run the segmentation this month, adjust bids on the AIO-present informational tail, and hold the structural moves until I had one quarter of my own data. I would not restructure the account into Performance Max purely for AI Mode eligibility while ad presence on that surface sits below one in three. The content bet is worth funding, and it should be funded as an addition with its own hypothesis, not as a subtraction from paid. The measurement problem that creates is examined in the piece on attributing dark discovery.

What budgets are actually doing already

The reallocation conversation is loud. The spending data points the other way.

Gartner's 2026 CMO Spend Survey, covering 401 CMOs, found search advertising at 16% of digital budget, up from 14.8% the previous year. Paid media reached 31.4% of total marketing expenses, a five-year high. Marketing budgets overall sat at 7.8% of company revenue, essentially flat on 7.7%.

So the aggregate picture is a slightly larger share of a flat budget going into search advertising, not a retreat from it. The reallocation is happening at the edges of the same pot rather than out of it.

The new money is going somewhere else. Gartner puts mean AI budget allocation at 15.3% across all respondents, funded by reallocation rather than budget growth. In a flat budget environment that money comes from somewhere, and the survey suggests martech is the donor. Martech fell to 19.4% of budget, a five-year low, from 26.6% in 2021.

On the generative engine optimisation side the numbers are softer. Scribewise surveyed 205 US marketing leaders and managers and found 55% had allocated budget for GEO, with 70% of those putting 11% to 20% of budget behind it, as reported by Digiday in March 2026. That is a small self-selected sample and a single vendor survey, so treat it as a signal of direction and not a measurement of scale. The broader question of whether that work pays is examined in the review of what the citation evidence actually shows.

The one number that should move your bids

Seer found that appearing as a cited source in an AI Overview carries a paid advantage of more than 4 percentage points of CTR, and that the advantage persisted every month of 2025. On informational queries in the earlier study, cited brands saw 91% higher paid CTR than uncited ones.

That is the closest thing to a bridge between the two budgets. Citation work is not a substitute for paid search. On the evidence available it makes paid search on the same queries work better, which argues for running both rather than trading one for the other. Whether that effect is causal or a selection artefact of stronger brands is not settled, and nobody should pretend it is.

Frequently asked questions

Where to start this week

One measurement, one adjustment, and a date in the diary.

Pull your last 90 days of search terms and tag each one by whether an AI Overview appears for it. Most rank tracking tools now report AI Overview presence, and a manual sample of your top 50 terms is enough to start. Then compute CTR, cost per click and cost per acquisition for each half.

If the AIO-present half is materially worse on cost per acquisition rather than only on CTR, cut bids there by 15% and read it in three weeks. If it is worse on CTR alone, change nothing, because you are looking at a denominator effect and your money is still working.

Then set a reminder for Seer's next update and for the next Alphabet quarter. Two data points, twice a year, will keep you closer to the truth than any benchmark you can buy.

Related reading