Google Shopping AI: How Products Get Picked in AI Mode and Gemini

1. 01

   Google shopping AI lives in AI Mode and AI Overviews in Search, plus the Gemini app. All of them pull product picks from one place, Google's Shopping Graph of more than 50 billion listings.

2. 02

   AI Mode doesn't run one search per prompt. It uses query fan-out to split the request into many related searches, and every sub-query behaves like a filter on your product data.

3. 03

   Google AI Mode chooses products it can match and explain. It can only match what your data actually says, so complete, consistent product data in Merchant Center and on the page decides most of the outcome.

4. 04

   Six optional conversational attributes in Merchant Center, from question_and_answer to popularity_rank, let you hand Google's AI the answers shoppers ask for in chat. Start with the Q&A one.

5. 05

   You can finally measure it. Merchant Center's AI performance report shows your share of voice in AI Mode and AI Overviews, and Search Console reports AI impressions by page.

Where the Shopping Graph gets its data

Google says the Shopping Graph holds more than 50 billion listings, with over 2 billion refreshed every hour (Google). Two inputs fill most of it:

• Merchant Center feeds. Your structured data: title, description, product_type, google_product_category, gtin, brand, color, material, size, price, availability, shipping and returns settings, and now the conversational attributes. A sloppy Google product category mapping puts items in the wrong candidate pool before any attribute is read.
• Your product pages. Googlebot reads PDP text and structured data. Google asks that structured data "matches the visible text on the page" and lists keeping Merchant Center information current as an SEO basic for AI features (Google Search Central).

The two get compared. If your markup says €89 and the feed says €79, Merchant Center raises a "Mismatched value (page crawl)" issue and the item can be disapproved. We check that diagnostic first in every audit.

How Google AI Mode chooses products

Google hasn't published a ranking formula, but its documentation and the AI performance report describe a clear sequence:

1. 1

   The shopper asks in plain language

   "A carry-on backpack for a rainy weekend in Dublin, fits a 16-inch laptop, under $150." Five constraints in one sentence.

2. 2

   AI Mode fans the question out

   It runs related searches across subtopics and data sources, which Google calls query fan-out (Google Search Central).

3. 3

   Candidates come from the Shopping Graph

   Listings whose category and attributes plausibly fit: travel backpacks, waterproof fabrics, laptop sleeves, anything under $150.

4. 4

   Explicit attributes do the filtering

   If no field says "water-resistant" or "fits 16-inch laptops", the product can't be confirmed for that constraint and drops out.

5. 5

   It ranks and explains

   AI Mode shows the products it can justify, building comparison tables from fields like capacity and weight, next to price and reviews.

6. 6

   The shopper acts

   They click through to your PDP, track the price, or buy with agentic checkout where the merchant is eligible (Google).

Every reason in that answer maps to a field. The other bags may be just as good. Their listings said less.

Eligibility basics

For web pages the bar is low. A page has to be indexed and eligible to show in Search with a snippet: "There are no additional technical requirements," and there's "no special schema.org structured data that you need to add" (Google Search Central).

Plenty of guides still treat schema as the main lever for AI Mode. For products the feed matters more; markup is mostly a consistency check.

Product listings need an approved Merchant Center account with free listings on. For checkout inside AI Mode and Gemini, Google's Universal Commerce Protocol offers a native checkout or a cart handoff to your site, and Google approves every integration. Our guide to agentic commerce protocols explains how UCP fits next to the other standards, and the agentic commerce guide covers the bigger shift.

The six conversational attributes

Google announced them with AI performance insights at Google Marketing Live 2026, rolling out globally (Search Engine Land). They're optional and they don't affect product approval. You can send them through a supplemental data source or the Merchant API (Productsup).

Sources: question_and_answer, item_group_title, popularity_rank in Google Merchant Center Help; limits for the other attributes from Productsup.

We add them through a supplemental data source keyed on the same id as the primary feed (a Google Sheet is fine for a pilot). The primary feed your ads team relies on stays untouched.

Do question_and_answer first, then item_group_title and variant_option if variant families are messy. Leave popularity_rank for last. It adds little while the attributes underneath are thin.

How to fill `question_and_answer` well

Google's rules here beat most third-party advice (Google Merchant Center Help):

• Product facts only

  No prices, sale dates or shipping promises. Offer data belongs in other attributes.

• Real questions, no keyword lists

  Write what a shopper would actually type or say, in plain language.

• No duplication

  Don't repeat what's already in title, description, product_detail or product_highlight.

• Spread the topics

  Specs, materials or ingredients, what's in the box, plus the occasions or activities the product suits.

• Skip what a PDF already answers

  If the facts are in a document_link file, Google can extract FAQs from it.

• Verifiable answers

  Every answer should be checkable on the product page or the label.

A practical way to source the questions

Use the questions shoppers already ask: support tickets and on-site search logs first, then reviews and the AI performance report's top terms. Answer each in one or two verifiable sentences. The same pairs belong on the PDP as a visible FAQ; whether to add FAQ schema on product pages on top is a separate decision.

Don't rush to fill all 30 pairs. Eight answers with new facts beat thirty that restate the description, and padding breaks Google's no-duplication rule.

Related: Google now asks merchants to send AI-generated descriptions in structured_description with digital_source_type set to trained_algorithmic_media, and it recommends putting the most important details in the first 160 to 500 characters of the description (Google Merchant Center Help). Our Google Merchant Center guide covers the full attribute set.

Merchant Center: AI performance report

It lives under Merchant Center, Analytics, Products, in the AI performance tab (Google Merchant Center Help). It covers organic conversational shopping queries (free listings, no paid ads) on AI Mode and AI Overviews. What you get:

• Your share of voice. Your AI impressions divided by total impressions for you and the competitors you define.
• Shopping stages. Share of voice split by stage: discovery, evaluation, ready to buy.
• Top terms and search intents, plus popular attributes, one product category at a time.
• Products showing. How many of your products appear for each term or intent.

It's limited to English-language queries, and to accounts in five countries: the US, Canada, Australia, New Zealand, India. That leaves out the UK and Ireland for now.

Google's advice is blunt: add relevant top terms to titles and descriptions, and fill missing attributes, most popular first. One caveat. Add a term only when the product really has that property.

Search Console: generative AI performance reports

Google introduced these reports in June 2026. They show how often your URLs appeared in AI Overviews and AI Mode, by page, country, device, date. Access is rolling out to sites in stages (PinMeTo).

AI feature traffic also sits inside the standard Performance report under the Web search type (Google Search Central). Filter on your PDP path (/products/ on Shopify) to see which product pages AI surfaces.

GA4: AI referral traffic

In Acquisition, then User acquisition, filter first user source / medium for Gemini and other AI assistants. Track sessions and revenue monthly, with key events alongside.

AI Mode and AI Overviews clicks generally land in GA4 as plain google / organic, so use Search Console for that split.

Expect small, valuable volumes. Adobe measured 53% more revenue per visit from AI-referred retail traffic than from non-AI traffic in July 2026 (Adobe).

For cross-engine measurement, see how to measure AI visibility.