Shopping Ad CTR Is Rising While Impressions Fall: Why That Is Not a Win

Fewer impressions can look exactly like better performance

The Monday report says click-through rate is up. It is hard to argue with a line that goes up, and the obvious read is that last quarter's creative work landed: the Shopping campaigns are converting attention better than they did a year ago. Then somebody scrolls one column to the left and finds that impressions fell off a shelf.

On September 7, 2026, Search Engine Land reported an analysis by Mike Ryan, head of ecommerce at Smarter Ecommerce, describing that exact shape. Across thousands of Shopping and Performance Max campaigns spanning hundreds of advertiser accounts, he found that "clicks were often relatively flat or slightly down, while impressions had fallen more noticeably". His proposed explanation is AI Overviews, and the most important thing about his proposal is how he labels it. Search Engine Land's own line: "Ryan is clear that the data doesn't prove AI Overviews are causing the change."

So there are two separate things on the table, and keeping them apart is the whole job. One is a measured pattern with a very large amount of data behind it. The other is a candidate explanation with one analyst's name on it and a warning label he attached himself. Both are useful to an advertiser. Only the first one is settled.

What the analysis actually measured

The numbers are worth reading before the theory, because they stand on their own even if the theory does not.

Impressions down, CTR up, clicks roughly where they were

Search Engine Roundtable covered the same underlying LinkedIn post the same day and put figures on the shape: "between mid-2025 and mid-2026, median Shopping ad impressions fell from roughly 1.85 million to 1.4 million, while median click through rate climbed from 1.20% to nearly 1.55%".

Search Engine Land compresses the mechanics into one sentence that is easy to skip and shouldn't be. Shopping ads "aren't necessarily generating significantly more clicks. They're being shown less often, which is pushing CTR higher."

Read that against the SEO story everyone already knows and the symmetry is the interesting part. Publishers have been gaining impressions and losing clicks, the pattern Ryan calls the crocodile effect. In Shopping ads he is describing the mirror image: fewer impressions, higher CTR.

The dataset is large, and size buys exactly one thing

The analysis covers 175 billion impressions. That number does a specific job and it is smaller than it looks: it buys confidence that the pattern is real rather than a quirk of a handful of accounts, and it buys nothing at all about the cause. A very large correlation is still a correlation.

This distinction matters more than usual here, because the headline figure and the headline theory travel together in every summary of this story. The figure is durable. The theory is a proposal.

The cross-check that started it

Ryan did not go looking for AI Overviews. He had been watching Shopping CTRs climb in Smarter Ecommerce's own benchmark data, and then, as Search Engine Land tells it, "separate ecommerce data from Optmyzr showed CTR up 17% year over year", close to the roughly 20% he was seeing. Two datasets agreeing is what prompted the deeper look.

Notice what that agreement covers. A second, independent dataset corroborated the pattern. Nothing in it corroborated the explanation.

The hypothesis, and the word doing the most work in it

Ryan's theory is specific enough to be testable, which is more than most platform speculation manages. It is also explicitly unproven, and he is the one who says so.

Experience switching, in his own words

The mechanism has a name. Ryan calls it "experience switching", and the version he posted himself is blunter than any paraphrase: "Google checks the predicted CTR of a given query and serves AIOs preferentially for lower-probability queries in order to preserve revenue." The consequence he draws from it is a two-part one: "Thus CTR climbs partly due to a shrinking denominator (fewer impressions) and partly due to selection/pruning (higher click propensity)."

Search Engine Land's rendering is the same idea in platform terms. Google may be more likely to serve AI Overviews for queries with a lower predicted likelihood of generating an ad click, while preserving Shopping ad placements for searches with stronger click propensity. If that were happening, Google could cut total Shopping impressions without cutting clicks at the same rate, which is a tidy explanation for a ratio that rises while the raw counts do not.

The hedge is his, not a caution we added

Every "may" in this post is copied from the source rather than layered on top of it. Search Engine Land sets out three ways the theory could be wrong, in its own words rather than a critic's:

• The timing could be coincidental. Two lines moving in a year is not one line moving the other.
• AI Overviews could be contributing through another mechanism. Right suspect, wrong mechanism is still wrong.
• Something else entirely could be responsible for the decline in impressions.

The verdict at the end of that paragraph is one short sentence: "For now, it's a hypothesis."

Google has not confirmed the reading, and it is worth being clear that Google confirming a third party's theory about its own ad-serving selection was never a realistic outcome. The absence of a Google statement is not evidence against the theory. It is also not evidence for it.

The leg with the least data under it

One part of the mechanism is weaker than the rest, and Ryan flags it as such. He says that anecdotally he rarely sees Shopping ads and AI Overviews together on the same result page, which is what makes "either/or" feel plausible. That is a personal observation, offered as one, and it is the piece most likely to be revised by anyone who measures it properly.

He seems to expect the revision himself. He told Search Engine Land he expects Google to move from more "either/or" search experiences toward "both/and" experiences as it develops AI-native formats and puts more Shopping ads directly inside AI Overviews. If that is where this goes, today's pattern is a phase rather than a new normal, which is its own reason not to rebuild your reporting around it.

Two ways a CTR can rise, and only one is a win

Strip the platform story out entirely and a useful piece of arithmetic is left over. It applies to your account whether or not AI Overviews have anything to do with it.

The denominator route

CTR is clicks divided by impressions. Cut impressions and hold clicks and the ratio rises without a single thing improving. This is the route that flatters you, because the number moves in the direction everyone celebrates while the business result stays exactly where it was.

The selection route

The other route is real. If the impressions you lose are the ones that were never going to click, what remains is a better-qualified pool and the higher CTR reflects a genuine change in who is seeing the ad. Ryan's theory says both routes are running at once, which is why he splits it into a shrinking denominator and selection toward higher click propensity.

The two are indistinguishable from CTR alone, which is the entire practical point and survives even if every word of the AI Overviews explanation turns out to be wrong. Telling them apart takes a second column:

• Call it selection when impressions fell, clicks held or grew, and conversion rate held or improved.
• Call it a shrinking denominator when impressions fell faster than clicks, and conversion rate and CPA drifted the wrong way.
• Call it a genuine creative win when impressions held steady and clicks grew against them.
• Call it unreadable when you only have the ratio, which is more often than most reporting admits.

A worked example: the same clicks, a much better-looking ratio

Take the reported medians and run one account through them. Nothing about the ad changes, no bid changes, no new creative:

Metric	Mid-2025	Mid-2026	Change
Impressions	1,000,000	761,000	Down 24%
Clicks	12,000	11,800	Down 1.7%
CTR	1.20%	1.55%	Up 29%

The illustration is built from the medians reported above, not measured on a real account, and that is the point of it: the exact shape those medians describe produces a 29% CTR improvement and 200 fewer visitors. A dashboard that shows CTR and hides impressions reports this as a good year. If you have ever wondered whether a high CTR means your ads are working, this table is the reason the honest answer is "not by itself".

What else changed in the same twelve months

The reason the causal claim stays open is not that the analysis is sloppy. It is that the window contains more than one large change, and a twelve-month comparison cannot separate them.

Amazon left most U.S. Shopping auctions

The clearest example has nothing to do with AI. Reporting on Tinuiti's Q2 2026 benchmark notes that "Amazon, for example, withdrew from most U.S. Google Shopping auctions in July 2025, opening more Shopping inventory to other retailers", and that "by Q2 2026, Tinuiti reported Amazon's Shopping impression share was still at 0% against the median retailer".

That is a structural change to who competes in the auction, it started at the beginning of the same window Ryan measures, and it persisted through all of it. Money kept arriving too: in the same panel, "Google paid search spending increased nearly 14% year over year in Q2, while Google Shopping investment grew 18%". More advertisers spending more in a channel changes impression distribution per account on its own.

A second panel measuring the same direction, with no theory attached

Optmyzr's Q1 2026 benchmark, reported separately in May 2026 and "based on more than 21,000 accounts", has "click-thru rate increased from 1.83% to 2.22%, a 21.31% gain year-over-year" while "impressions also dropped by roughly 11% year-over-year". Those are account-wide Google Ads figures rather than a Shopping-only cut, and the phrase AI Overviews appears nowhere in that coverage.

This is the strongest thing that can be said about the story and it is worth saying precisely. The impressions-down, CTR-up direction has two independent measurements from two vendors with different account bases. The AI Overviews explanation has one origin and one disclaimer.

Twelve months is a window, not a control

Put those together and the honest summary is that the analysis observes two lines moving in opposite directions over a year in which at least four things moved at once:

• The competitor set, after Amazon left most U.S. Shopping auctions in July 2025.
• The money, with Google Shopping investment up 18% year over year by Q2 2026.
• The result page, with the pattern itself dated to the expansion of AI Overviews.
• The ad formats, which Ryan expects to keep changing as Google develops AI-native units.

The analysis does not isolate any one of them, and Ryan does not claim it does.

That is not a reason to ignore the finding. Practitioner panels are often the only place a shift like this is visible at all, months before any platform documents it, which is the same reason it pays to watch how AI summaries are reshaping the ads around them and what Google is shipping into AI Mode. It is a reason to keep the finding and the theory in separate boxes.

The metrics to read next to CTR

None of what follows depends on the AI Overviews question being settled. It is the part you can act on this week.

Impressions and absolute clicks, in the same view

Search Engine Land's guidance lands in the same place: "advertisers may need to look beyond CTR and pay closer attention to impression volume, clicks and overall traffic when judging performance". If your reporting shows a ratio without the two numbers that produced it, the ratio can move 29% while your traffic falls and nothing in the report says so.

The practical version is a four-line check on any CTR movement:

• Did impressions move more than clicks did, in percentage terms?
• Did absolute clicks hold, or did only the ratio improve?
• Did conversion rate and CPA follow CTR, or diverge from it?
• Did the mix of queries change shape, or only the volume?

Conversion rate and cost per acquisition, which moved the other way

The Optmyzr benchmark is more useful here than the headline number it is usually quoted for. In the same period that CTR rose 21.31%, "conversion rate declined slightly by 0.96%, while CPA increased 4.41%". Optmyzr's own summary of the shift is the cleanest published statement of the problem: "more clicks, from a smaller impression pool are converting at a marginally lower rate".

Higher CTR, slightly worse conversion, slightly more expensive acquisition. Three metrics, one direction of travel, and the one that looks best is the one that means least. This is exactly why creative analytics is worth setting up as a habit rather than a quarterly panic, and why the metric you optimise toward should survive a change in how often you are shown.

The question CTR was never able to answer

The missing question is "was I shown at all", and impression share answers it where CTR cannot. In a year when the result page is being rebuilt around AI answers, the share of available impressions you captured is a more honest scoreboard than the share of impressions you converted to clicks.

Everything above is measurement. The half of the problem that measurement does not touch is what the ad says once someone does see it, which is where the numbers stop helping and testing more creative starts, whether you shoot those variants or generate them in Novoads.

How Novoads solves the creative half of a shrinking impression pool

If the impression pool really is narrowing toward higher-intent queries, then fewer people see your ad and the ones who do are further along. Both halves of that raise the bar on the asset itself.

A narrower query set raises the bar on the asset

There is a version of this shift that is good news. Being shown less often to people who were never going to buy is not a loss. But it removes the cushion that volume used to provide: when impressions are plentiful, a mediocre creative still finds its buyers eventually, and when they are not, the creative has to work on the first view. That is a production problem before it is a media problem, and the working assumption of roughly ten tries to find one winner does not get cheaper because the auction got tighter.

The Shopping and Performance Max campaigns in this analysis are the same ones that consume video assets. Google's own Performance Max documentation lists videos among the creative assets an advertiser provides, so the asset group that runs your Shopping inventory is also the one asking for footage most ecommerce teams do not have the budget to shoot every month.

From a product photo to the assets an asset group needs

That is the half we build, and the loop is short enough to run monthly:

• Upload a product image. No URL to paste and no plugin to install.
• Write or auto-generate the script, then pick an AI actor to deliver it.
• Render vertical or horizontal, in 9:16, 1:1 or 16:9, as a downloadable file.
• Repeat for the next angle, which is what turns one idea into a testable set.

It rents the same frontier engines every serious tool rents, which makes the render quality a tie and moves the argument to what it costs to get an ad you can actually run. Novoads starts at $49/month on the Starter plan, with 50 credits per month, and all plans are published on the pricing page. Start for $49/month. 50 credits every month. Cancel anytime.

If you would rather see the format first, our ecommerce video ad breakdown covers what actually goes in the asset group.

CTR became a mix metric and stopped being a scoreboard

That is the sentence worth keeping from all of this, and it needs no position at all on AI Overviews. For most of the history of paid search, click-through rate was a reasonable proxy for whether an ad was doing its job, because the pool of impressions it was measured against was roughly stable. When the platform starts choosing which queries get ads at all, the denominator becomes a variable the advertiser does not control and cannot see in the ratio.

Mike Ryan may turn out to be right. He may turn out to be right for the wrong reason, or wrong in a way that teaches the industry something about how AI answers and ad inventory share a page. The useful part does not wait on that verdict: the moment a metric's denominator is being set by someone else's product decisions, that metric describes the mix you were given rather than the job you did. Read it that way, put impression volume and absolute clicks beside it, and the Monday report stops being able to flatter you.