AI Mode Picks 4 Products. Be One.

Google’s AI Mode shows about 4 products where standard search shows 22, and the two surfaces pick largely different products. A September 2026 expansion of Merchant Center AI performance insights finally gives sellers the attribute-completeness score behind those picks. Here is what to do on the listing itself.

The shortlist just got brutally short

Over three weeks in July 2026, ProductRise ran the same shopping queries through Google’s standard search and Google AI Mode on the same day — more than 100,000 paired results covering more than 2 million product listings. Standard search returned products on about 88% of queries. AI Mode returned them on 23%. When products did appear, standard search showed about 22.5 per page; AI Mode showed 4.3. Combine the two effects and AI Mode surfaced roughly 95% fewer product listings than the search page most Shopify sellers still optimize for.

A follow-up study by the same team, covering August 9–31, 2026, found the products AI Mode does pick are a different crowd. Only 1.28% of products ranking in traditional search also appeared in AI Mode for the same query on the same day. And when the same product appeared on both sides with different prices, AI Mode showed the more expensive price 68.4% of the time.

The reading most sellers take from a “95% collapse” headline is that the game is over. The more accurate reading is that the game changed surfaces. The crowded 22-product carousel was a volume game where an incomplete listing could still snag a slot. A 4-product shortlist is a quality game where the engine picks the products it can verify — and a Shopify seller controls most of what gets verified.

That is why this is a listing job and not a webmaster job. The attributes AI Mode filters on — brand, product type, size, color, material, price, availability — are fields you edit in your Shopify admin. If you are new to this discipline, start with our GEO for marketplace sellers guide, which walks through the full field-by-field method.

Google now tells you which attributes it wanted

For years the attribute problem was invisible: listings silently lost eligibility and nobody could say which missing field was responsible. That changed in 2026. Google’s AI performance insights in Merchant Center — announced May 27, 2026, rolled out to US merchants starting July, and expanded on September 3, 2026 to Australia, Canada, India, and New Zealand — include an attribute completeness score.

The report does something no SEO audit can: it names the popular product specifications shoppers actually search — color, style, material — and then flags which of your products are missing those structured attributes. Google’s own description says the insights “help provide you with a view into your brand’s performance on AI surfaces and how you can optimize your product data.”

This closes the feedback loop. You no longer have to guess whether “organic cotton” belongs in the title, the description, or a structured field. The report tells you the terms buyers use, and the completeness score tells you which products fail to carry them as structured data.

One caveat before the checklist: the report is currently Merchant Center-centric. Shopify sellers reach it through the Google & YouTube channel, which feeds Google’s surfaces including AI Mode and Gemini. It measures your feed, not your storefront page — but the feed is drawn from your Shopify product fields, so every fix still happens on the listing.

The five fields that decide the shortlist

AI Mode’s shortlist is a filter result. Before any model writes an answer, the shopping system has to match products to the query, and Google’s product data specification says plainly that it “uses this data to match your products to the right queries” and that “missing or incorrect variant attributes (such as item group id, color, or size)” can cause limited eligibility. Here is the operating order we recommend, weighted by what the new report exposes first.

1. Category and product type

Set the Shopify product category (Taxonomy) so the feed carries a structured google_product_category rather than guessing from title text. Category errors are the first named cause of eligibility loss in Google’s spec. An apparel product filed under a generic “Apparel & Accessories” parent misses every sub-category filter — “women’s rain jackets,” “slim-fit chinos” — that a spec-led AI query actually uses.

2. Variant attributes

Declare every option — color, size, material, inseam — as a proper variant with its own option name, not as words buried in the title. Google names missing variant attributes a top disapproval cause, and the August ProductRise data shows AI Mode answers frequently link to a specific variant’s URL, not the parent product. A variant with no structured color cannot be the “navy, size medium” answer to any query.

3. Material and care facts

For apparel sellers this is the field the attribute report surfaces most. Shoppers ask AI “is this warm enough for winter” or “will this shrink.” The model needs fiber content and care instructions as structured facts to answer without guessing. If material lives only in your description prose, it competes with marketing copy for the model’s attention — when it is a structured field, it is simply read.

4. Identifiers

GTIN barcodes on every variant let the shopping graph merge your listing with every other record of the same product. That merge is how the same product earns the review count and price history that make it a confident pick. Our Shopify Product GEO review covers the identifier field map in detail.

5. Price and availability hygiene

AI Mode shows live prices. The August study found price disagreements between surfaces on 38.1% of matched products — often a stale sale price against a current one. The fix is making the Shopify price the single truth: one price, current, no leftover compare-at artifacts feeding your product fields into the feed.

Shopify’s own data says the same thing

The strongest evidence that listing structure decides AI outcomes does not come from Google. It comes from Shopify. In August 2026, Shopify published Q2 commerce data showing that when AI platforms drew on structured Shopify Catalog product data, the shoppers they referred converted 2x better than those referred from scraped or third-party feeds. AI-referred sessions to Shopify storefronts grew 197% year-over-year, and in spec-led categories AI-referred shoppers converted at about double the rate of organic search.

Read that carefully: the conversion gap is not between AI traffic and no traffic. It is between structured product data and unstructured product data. The surface changed; the lever is the same fields.

That 2x conversion figure has a boundary worth stating. It describes shoppers who were already referred — it does not establish that completing attributes causes more AI traffic, and Shopify presents it as an early signal from its own platform data, not a controlled experiment. What it does establish is that when your listing’s fields are complete enough for a platform to consume as structured data, the resulting interaction is worth roughly double. Completing fields is cheap; ignoring a 2x quality gap is expensive.

The fifteen-minute attribute audit

Here is the operating section. Set a timer and do this on your five best-selling products:

1. Open the product in your Shopify admin. Check that the category field is set to a specific Taxonomy node, not left empty.
2. Expand every variant. Confirm each one has its option values filled (no “Default Title” leftovers), a price, and a barcode where the product has one.
3. Read the description as a parser would. Any material, dimension, or compatibility fact that appears only in prose — move it to a metafield or the structured field the Google & YouTube channel maps.
4. Run the Google & YouTube channel’s feed preview (or Merchant Center diagnostics) and note every “missing attribute” warning. These are the exact gaps the AI insights report will score.
5. Re-check in two weeks. The insights report updates as your feed refreshes, so attribute completeness is now a measurable trend, not a one-off chore.

If your catalog is large, do the top five products by revenue first. AI shortlists concentrate on products the graph can verify; the ones with complete records in your best sellers are the ones that show up.

How FirstShelf can help

FirstShelf exists to make products visible and recommended by AI. Our Buyer Match review reads a product listing the way a shopping assistant does — field by field, not page by page — and scores exactly the attributes this article names: category precision, variant completeness, identifiers, and the structured facts buried in your descriptions. You leave with a ranked list of which fields to fix on which products, which is the same list Google’s attribute completeness score will eventually show you. We do not sell schema cleanup; we audit the listing fields AI shopping systems actually match on.