Google's AI Overviews Are Eating Shopping Clicks. The Winners Are Being Cited, Not Ranked.
AI Overviews now hit 14% of shopping queries. The data shows the winning move is earning citations, not rebuilding SEO.
Google's AI Overviews now appear on 14% of shopping queries, and the click that used to follow a top ranking is disappearing into the summary. Most brands are responding by rebuilding their SEO and paid stack. The data says that is the wrong fix. The brands holding their discovery are the ones the model cites.
What the shopping data actually shows
The shift is measurable, not anecdotal. Search intelligence firm Similarweb reported in June that AI Overview impressions on commercial-intent queries — "best running shoes under $150," "non-toxic sunscreen for kids" — climbed 61% year-over-year, while click-through on the classic links below them fell.
Semrush found the same pattern from the brand side. Across 1,200 e-commerce domains, category-level informational queries saw average click-through rates drop 28% year-over-year. The traffic did not move to a competitor's page. It stopped at the answer.
Google is building for that behavior on purpose. In May it introduced Universal Commerce Protocol features and AI tools for what it calls a more agentic shopping experience — where the model, not the shopper, does the browsing and comparison. When the machine assembles the shortlist, the question is no longer "do we rank." It is "does the model reach for us when it answers."
By the numbers
| Signal | What changed | Source |
|---|---|---|
| Shopping queries with an AI Overview | ~14% and rising | Report, 2026 |
| AI Overview impressions on commercial-intent queries | +61% year-over-year | Similarweb, June 2026 |
| Click-through on category-level shopping queries | −28% year-over-year across 1,200 domains | Semrush |
The pattern is consistent from both directions: the model shows up more often, and the click that used to follow the ranking shows up less.
Why rebuilding SEO misreads the shift
The instinct to rebuild the SEO and paid stack treats this as a ranking problem with a new interface. It is not. A page can rank first and still be invisible inside an AI Overview if the model synthesizes its answer from three trusted sources and none of them is you.
That is the trap in the current panic: brands are spending against position on a page most shoppers no longer click. The commercial-intent query still exists, the demand is still real, but the surface that satisfies it has changed shape. Optimizing harder for a shrinking click is effort pointed at the wrong layer.
This is what Jaxon Parrott named Machine Relations in 2024 — the discipline of making a brand legible to the machines that now mediate discovery. The framing matters here because it separates two problems people keep merging: being findable by a person scanning results, and being usable by a model assembling an answer. AI Overviews reward the second.
The brands winning are being cited
The mechanism underneath an AI Overview is selection, not ranking. The model pulls from a small set of sources it treats as authoritative for that query, then paraphrases. Get named in that set and you appear in the answer, often with a citation link. Miss it and no amount of on-page work puts you there.
That selection favors earned authority — third-party validation the brand does not own. It is a consistent finding across the space: research on earned versus owned citation rates shows AI engines cite independent, third-party sources at several times the rate of brand-owned pages. The model trusts the outlet that has no reason to flatter you.
Which outlets? That is now a measurable input, not a guess. AuthorityTech's publication intelligence tracks which publications AI engines actually cite by category, and the distribution is narrow — a handful of trusted outlets carry most of the citations in any given vertical. For a shopping brand, being reviewed, tested, or listed in those specific outlets does more for AI Overview presence than another round of title-tag work.
What CMOs should measure now
The scorecard changes with the surface. Three shifts move a brand from fighting for a vanishing click to earning a place in the answer:
- Track citations, not just rankings. Position on the page is a proxy that has decoupled from visibility. Measure how often the model names your brand in answers to your core commercial queries. That is the number that maps to demand now.
- Fund earned coverage in the outlets the model trusts. Identify the small set of third-party publications your category's AI answers pull from, and get your product genuinely tested and listed there. Earned validation is the raw material AI Overviews are built from.
- Make your claims machine-legible. Structured, specific, verifiable product data — third-party test results, clear specs, real review counts — gives the model quotable facts. Vague brand copy gives it nothing to cite.
None of this means SEO is dead. It means ranking is now a means to an end, and the end is citation. The brands that adapt fastest are the ones that stop asking "where do we rank" and start asking "when the machine answers, does it reach for us."
For teams that want a baseline before rebuilding anything, a visibility audit measures where a brand already stands inside AI answers — the starting point that tells you whether the problem is coverage, data, or trust.
FAQ
Are AI Overviews actually reducing shopping traffic, or just changing where it lands?
Both, but the reduction is real. Semrush measured a 28% year-over-year drop in click-through on category-level shopping queries across 1,200 domains. The demand persists; the click that used to carry it to a product page increasingly stops at the summary.
If we rank #1 but aren't in the AI Overview, what fixes that?
Not more on-page SEO. AI Overviews synthesize from sources the model trusts for that query, which skews toward independent third-party coverage over brand-owned pages. Earning genuine reviews and listings in the outlets your category's AI answers already cite is the direct lever.
How do we know which publications to target?
Citation distribution is measurable by category. Tools that track which publications AI engines cite show a narrow set of trusted outlets carrying most citations in any vertical — that list, not a generic PR push, is where earned coverage pays off.