Google AI Mode Lands in Gemini and Circle to Search, Shopping Graph Crosses 50B SKUs: Re-Rank Your Feed
Three new surfaces, three behaviors
April 7 was the day Google quietly broke the assumption that AI Mode shopping lived inside Search. It now runs in three distinct places: classic Search AI Mode, a shopping panel inside the Gemini app, and the results of a Circle to Search gesture. Each surface pulls from the same Shopping Graph, but each weighs your feed fields differently.
The Gemini app panel is conversational. A shopper types “find a birthday gift for my mom, budget around 80 dollars” and Gemini composes a carousel of products with prices, availability and merchant ratings inline. In that flow product_highlight and description carry serious weight, because Gemini is reading them as bullet points and short-form context, not as fallback metadata.
Circle to Search runs on a different logic. A user screenshots an Instagram or TikTok still, circles a handbag, and Google reverse-matches against the Shopping Graph using vision. The fields that matter here are image-side: image_link, additional_image_link with multiple angles, and structured visual attributes like color, material, pattern. A listing with a single hero image is effectively invisible.
Search AI Mode is still text-led, but AI Overviews now appear on roughly 14 percent of shopping queries. A search for “best running shoes for flat feet” returns an AI summary above the fold with product cards sourced from your feed. The summary is generated, but the cards are pulled by structured attribute match.
What 50 billion SKUs changes about feed competition
The Shopping Graph crossed 50 billion SKUs the same week. For context, it was roughly 35 billion two years ago and 45 billion last year. The growth curve is steepening, which means basic differentiation by category, price and brand no longer separates a listing from the crowd.
What the model now rewards is structured attribute completeness. For a running shoe, fields like age_group, gender, size_system, size_type, material and pattern together let AI Mode answer a long-tail query like “flat feet” with precision. A feed that only ships title and price will not surface against that intent at 50B-SKU density.
The March 2026 Core Update, which finished its 21-day rollout on March 31, reinforced the same direction. Search Central reported original product content gaining an average of 22 percent visibility lift after the update settled. For feeds, “original” maps directly to descriptions written with real usage context, spec detail and material sourcing, not the templated copy lifted from a supplier sheet.
A second shift. AI Max, previously a limited beta, has now rolled out to Standard Shopping campaigns. AI Max generates dynamic creative by pulling every available field in your feed. A thin feed produces thin creative, which produces thin click-through.
The feed audit checklist
Below is the audit list I recommend running before your next AI Max push or budget reshuffle, prioritized by impact.
| Priority | Field / Area | Check | Surface affected |
|---|---|---|---|
| P0 | title | Core attributes (brand + category + key spec) in the first 70 characters | All three |
| P0 | description | Real-world context, minimum 500 characters | Gemini + Search |
| P0 | image_link + additional_image_link | At least 3 distinct angles, 800x800 minimum | Circle to Search |
| P1 | product_highlight | Up to 5 bullet points, one selling point each | Gemini |
| P1 | gtin / mpn / brand | Complete and consistent with authoritative DBs | All three |
| P1 | availability + price | Real-time sync, latency under 1 hour | Search AI Mode |
| P2 | material / pattern / color | Visual attributes fully populated | Circle to Search |
| P2 | size_system / size_type | Region-specific labels | Gemini conversational |
| P3 | product_detail (key/value pairs) | Minimum 5 structured pairs | AI Max creative |
If you fail P0 or P1 anywhere, do not raise budgets yet. Fix the feed first.
Attributes that now directly drive AI recommendations
A few fields shifted noticeably in weight after the April 7 rollout. Flag these to your feed engineer or data team.
product_highlight. Historically optional, now rendered directly as bullet points inside the Gemini shopping card. Aim for 3 to 5 bullets, 75 characters or less each, one focused selling point per bullet. Stacked adjectives waste the slot.
Semantic density inside description. AI Mode can now distinguish “premium fabric” from “50 percent Merino wool plus 50 percent recycled polyester.” The second parses into the Shopping Graph attribute map. The first is ignored. Write descriptions like spec sheets, not like ad copy.
Image angle diversity via additional_image_link. Circle to Search reverse matching needs front, side, detail and lifestyle shots at minimum. A single hero image will almost never trigger a match in Circle to Search.
shipping and return_policy. AI Mode now surfaces explicit trust signals. Listings with vague or missing return terms get deprioritized regardless of price advantage. This is a behavior shift, not just a display change.
Sponsored Stores and Direct Offers are the two new ad surfaces tied to AI Mode. Sponsored Stores promote an entire merchant as an entity inside the AI flow, which favors brand-driven catalogs. Direct Offers inject single-SKU promotions into Gemini conversations, but require clean sale_price and sale_price_effective_date data to qualify.
Measurement adjustments
The traditional funnel (impression, click, conversion) undercounts performance on AI surfaces because shoppers often complete information consumption inside the AI summary without clicking through.
Add two new metrics. First, the AI surface impression metric in Merchant Center, which tracks visibility across AI Mode, Gemini and Circle to Search separately. Second, a query-level informational versus transactional split, since the two intent types now have radically different click rates but both contribute to assisted conversion.
Attribution needs revisiting too. Last-click is no longer reliable. A shopper might ask Gemini first, circle a product in Circle to Search later, then open Search and type the brand name directly before purchase. GA4 data-driven attribution combined with Merchant Center cross-surface reports is the minimum stack to see the full path.
Budget allocation also needs revisiting. With AI Max now covering Standard Shopping, the split between Performance Max and Standard Shopping campaigns no longer works the way it did six months ago. A reasonable starting point is 70 percent to Performance Max plus AI Max as the primary surface engine, and 30 percent to manually curated Standard Shopping for control coverage. Run two weeks, then rebalance using the cross-surface report.
Related Articles
Amazon Synthetic Performer Image Labeling: XMP Metadata Compliance Guide for Sellers
Starting July 22 2026, Amazon requires all product images and A+ Content featuring photorealistic AI-generated people to carry contains-synthetic-performer XMP metadata. This guide covers what needs tagging, how to batch-add metadata with ExifTool and Adobe Bridge, and what happens if you skip it.
EU AI Act Article 50 Takes Effect: AI Content Compliance Guide for Cross-Border E-Commerce
As of August 2 2026, EU AI Act Article 50 requires machine-readable markers and visible disclosure on AI-generated content shown to EU consumers. This guide covers what cross-border sellers need to do with product images, video ads, and chatbots, plus how to implement C2PA metadata.