A pattern is circulating in paid search circles: Shopping ad click-through rates are climbing, up roughly 20% year over year, while impressions fall more steeply. Mike Ryan of Smarter Ecommerce spotted it across thousands of Shopping and Performance Max campaigns and proposed a hypothesis, carefully labeled as such: that Google may be serving AI Overviews on queries with low predicted ad-click propensity while preserving Shopping placements where click propensity is strong. The observation is sound and the hypothesis is reasonable. The framing that has grown around it, that advertising opportunities are being suppressed, is where I part company. Read the same numbers from inside the auction and they describe something else entirely.
What Shopping used to be
To see why, you have to remember what Shopping actually is. It was built as a comparison surface, and for years it behaved like one: it covered the whole funnel. Enormous impression volume, low CTR, mediocre conversion rate. Most of those impressions came from people who were browsing, comparing, forming an idea of a category, not buying. That was never a defect. Shopping earned its keep as a support layer for Search and as a genuine lift lever, and the long tail of weak-intent impressions was simply the cost of covering everything.
But it did mean that Shopping was doing two jobs with one instrument, and the metrics of the two jobs pulled in opposite directions. Discovery wants reach and tolerates a low click rate. Purchase wants qualification and rewards a high one. Blended together, they produced exactly the profile Shopping always had: huge numerator of impressions, thin conversion.
The numbers describe pruning, not amputation
Now look at what is actually reported. Impressions down noticeably. Clicks flat, or slightly down. CTR up.
Hold those three together, because the combination is the whole argument. If AI Overviews were suppressing valuable advertising opportunities, clicks would have fallen roughly in proportion to impressions. They did not. Substantially the same click volume is arriving on a smaller number of impressions, which means the impressions that disappeared were, in the main, the ones that were not converting into clicks anyway. That is the weak-intent tail leaving the auction. The buyers who were going to click are still clicking.
Clicks held while impressions fell. That is not an opportunity being taken away. That is a low-intent tail being removed from a channel that was never good at serving it.
Which reframes the whole picture. If AI Overviews are absorbing the research-and-compare stage, Shopping stops being a comparison surface stretched across the funnel and becomes what it is genuinely best at: a lower-funnel, high-intent asset. Fewer impressions because the intent filter is tighter. Higher CTR because the audience is more qualified. And, critically, a better conversion rate, because the traffic arriving is closer to purchase. Meanwhile Performance Max continues to work the upper and mid funnel across its own surfaces. The portfolio did not lose a stage. It reallocated one.
How to tell which reading is right
Both readings are compatible with rising CTR, which is precisely why CTR is the wrong metric to argue about. It is a ratio, and a ratio can rise because the numerator improves or because the denominator shrinks. An advertiser celebrating a 20% CTR increase driven entirely by falling impressions has not improved anything. They have received the same clicks on fewer chances and mistaken the arithmetic for performance. This is basic auction hygiene, and it is the first thing I check when a client brings me a metric that looks like good news.
The discriminating metrics are further down. If the specialization reading is correct, conversion rate should be rising alongside CTR, and CPA and ROAS should hold or improve, because the traffic you are still buying is better qualified. If the suppression reading is correct, you would see click volume and revenue eroding without any qualitative compensation: fewer clicks, same or worse conversion rate, deteriorating efficiency. Those are two clearly different signatures, and any advertiser can check which one their own account shows this week. That is a more useful exercise than debating the hypothesis in the abstract.
The part nobody should wave away
Now the honest complication, because the specialization reading has a real weakness and it deserves naming rather than dismissing. That weak-intent tail was not worthless. It was the discovery layer, the lift lever Shopping genuinely provided. Saying "Performance Max covers the upper funnel" is only partly true, because PMax works the upper funnel across surfaces that are themselves increasingly contested by AI-generated answers. If AI Overviews are absorbing the comparison stage, that stage is not simply migrating to another line in your media plan. It is migrating out of the auction.
So there is a gap. Not a gap in your Shopping performance, which may well be improving, but a gap in your presence during the stage where the buyer is building a shortlist of who to consider. And naming that gap is not a concession. It is the entire point.
The stage you cannot buy
Earlier this year I argued that your Citability Score had become a lever on your paid-media economics, because on a commercial query that triggers an AI Overview you could either earn presence organically or pay a rising CPC for it. That framing now looks too modest. If Shopping and AI Overviews are tending to be either/or rather than both, then on a growing class of queries there is no auction to enter. The shortlist is being assembled inside an answer, and no bid participates in that process.
This splits your keyword universe into two territories with different rules. In one, the buyer is close to purchase, the ad serves, and you compete on bid, feed quality, offer and landing experience. In the other, the buyer is still deciding who belongs in the consideration set, an AI answer frames that set, and the only way in is to be cited. Paid closes. Citability qualifies. And the qualifying stage no longer has a price list.
That is a structural change in what a media plan even is. Knowing which of the two territories a keyword sits in becomes a planning decision, not a reporting detail, and it is measurable: the AI Visibility Index shows where you currently appear across the queries that frame your category, and the Citability Score explains whether you are built to hold that position.
A note on where this goes
One reason to hold all of this loosely: Ryan himself expects the current either/or pattern to be temporary, with Google moving toward both/and as it builds native ad formats inside AI surfaces, and the direction of travel across the industry supports him. That does not weaken the argument. It sharpens it. In the either/or world, citability is the only way into the consideration stage. In the both/and world, you will be bidding for placement inside an AI answer whose content is assembled from sources chosen on structural grounds, which means your citability determines the context your ad appears in and, almost certainly, what that placement costs you. Either way, the variable you control is the same one.
Fifteen years in the auction taught me to distrust a metric that improves for reasons nobody has explained. Shopping CTR is up, and the reason matters more than the number. If the reason is that the comparison stage has moved somewhere you cannot bid, then the most important line in your paid media strategy is no longer in your paid media account.