Google AI Mode and AI Overviews: what changes for product pages
Google AI Mode and AI Overviews now shop from your data feed, not your homepage. Here is what product pages need to earn the citation.
Google's AI Overviews stopped being a novelty for shopping queries somewhere around late 2025, and the growth curve since has been steep: an analysis of 20.9 million shopping keywords by Visibility Labs found AI Overviews jumped from 2.1% of shopping SERPs in November 2025 to 14.0% by March 2026, according to Search Engine Land. Add AI Mode's conversational, multi-turn shopping flows on top of that, and distributors and retailers are looking at a search surface that increasingly summarizes and recommends before a shopper ever clicks through. The mechanics of what gets cited are less mysterious than they sound, and they reward exactly the kind of product data discipline most catalogs don't have.
Your product page is no longer the primary source
The core shift: Google's AI shopping features draw heavily from the Shopping Graph, which is populated by your Merchant Center feed and your on-page schema markup, not by your brand copy or hero images. Google's own guidance is explicit that AI-powered features use the same underlying requirements as regular Search results, plus Merchant Center data that has to be accurate and current. There's no special "AI schema." There is, however, a much lower tolerance for gaps.
Practically, that means a shopper asking an answer engine a question never lands on your PDP first. The model reads your feed, cross-references your schema, and only recommends you if what it finds is complete enough to answer the question without guessing. If your attributes are thin, the model doesn't flag it as a data problem, it just recommends whoever answered the question better.
What actually gets rewarded
A few patterns are converging across recent reporting on Merchant Center and AI Overview behavior:
| Signal | Why it matters for AI citation |
|---|---|
Accurate GTIN / sku | Google treats GTINs as the strongest product-matching signal; wrong or missing IDs drop you out of the comparison cluster entirely, per Marcel Digital |
Complete Product schema (name, image, offers, price, priceCurrency) | These are Google's documented minimum required fields for merchant listing markup |
availability, itemCondition, hasMerchantReturnPolicy | Recommended fields Google calls out for stronger listings, and the ones AI models use to answer "can I return this" or "is this new" without a click |
| Feed-to-page-to-schema consistency | Variant attributes (size, color, material) have to match exactly across your PDP HTML, your JSON-LD, and your Merchant Center feed, or Google penalizes reliability, per the same Marcel Digital analysis |
| Explicit attribute depth (material, dimensions, compatibility) | Products missing basic differentiating attributes may never surface in a generative summary at all |
None of this is exotic. It's the same completeness and consistency problem enrichment teams have been fighting for a decade, just with a much less forgiving audience reading the feed now: a language model that either finds the fact or moves on.
A before/after, concretely
Take a mid-market distributor's raw supplier feed for an office chair:
Raw feed description: "Ergonomic task chair, adjustable, mesh back, black, good for office use."
Enriched attribute table:
| Attribute | Value |
|---|---|
material_back | Mesh, polyester blend |
seat_material | Molded foam, fabric upholstery |
adjustability | Seat height, armrest height, lumbar depth, tilt tension |
weight_capacity | 300 lbs |
assembly_required | Yes, tools included |
certifications | BIFMA tested |
warranty | 5-year limited |
color_options | Black, graphite, navy |
Ask an answer engine "office chair for a 250 lb user with lumbar support under $300" and the first version has nothing for it to match on. The second gives it a weight capacity, a specific adjustability claim, and a price band to reason over. That's the difference between being summarized and being skipped.
The consistency trap most catalogs fall into
Here's the part that trips up most mid-size distributors: schema markup is easy to bolt onto a template, but keeping it synchronized with a live feed across thousands of SKUs, multiple suppliers, and constant price and stock changes is not a one-time project. It's ongoing maintenance, and it's exactly where manual processes fall apart, since hand-checking attribute-by-attribute against supplier docs runs in the neighborhood of 30-45 minutes per SKU at any real catalog scale.
Where this connects to enrichment
Your PIM stores the data. Anglera does the work: it plugs into Akeneo, Salsify, inriver, Stibo, Syndigo, Pimcore, Informatica, or a flat file if you have no PIM at all, and it continuously scores, gap-fills, and reconciles attributes against the actual source documents your suppliers send. That's the mechanism that keeps a Product schema and a Merchant Center feed saying the same thing about the same SKU, week after week, without a team manually re-checking it. As AI Mode and AI Overviews keep expanding into shopping, the catalogs that stay legible to a model reading a feed, not a page, are the ones that get cited. Most teams can get a live view of where their own gaps are in about two weeks, not a multi-year systems overhaul.
Sources: