AI and search··Snomi Team

Google AI Mode in shopping: how it works and what to fix in your product data

Google AI Mode turns product search into a conversation. Gemini, query fan-out, and Shopping Graph all run on the data in your Merchant Center feed.

Last updated:

Google AI Mode turns product search from a list of links into a conversation. Gemini interprets the shopper's question, query fan-out breaks it into related searches, and Shopping Graph supplies products. All three layers run on the data in your Merchant Center feed, so completeness and freshness decide visibility.

Key takeaways

  • Product discovery in Google Search has moved from keywords to conversational queries, visual search, and follow-up questions. Queries in AI Mode are on average about three times longer than classic ones.
  • Visibility in AI results depends on product data. Shopping Graph holds more than 60 billion offers, with more than 2 billion refreshed every hour [1].
  • Shopping is becoming visual. Google Lens records more than 25 billion searches a month, and roughly one in five has shopping intent [2].
  • AI Mode has been available in Poland since October 2025, but it does not yet pull Shopping Graph data for Polish results. Shopping queries therefore end in citations and store links rather than panels of specific product offers. Google rolls Shopping Graph features out on the US market first.

Table of contents

  1. What powers AI Mode: Gemini, query fan-out, and Shopping Graph
  2. How visual search changes image requirements
  3. Your Merchant Center feed is now your most important SEO asset
  4. Price tracking and agentic checkout: what to prepare
  5. What this means for sellers today
  6. You do not have to do this by hand
  7. Why feed quality decides visibility now
  8. FAQ

What powers AI Mode: Gemini, query fan-out, and Shopping Graph

Layer What it is What it means for sellers
Gemini Google's reasoning model. It interprets longer queries, catches intent, constraints, and preferences, and handles follow-up questions Data must clearly say what the product is, who it is for, and which traits make it a good match
Query fan-out Splitting one complex question into many related searches: category, use case, price, material, size, reviews, availability An incomplete feed drops out of some sub-queries even when the product physically fits the need
Shopping Graph Google's product data set - more than 60 billion offers with prices, availability, reviews, colors, and store information Your feed and on-page structured data are what Google matches to queries

The practical takeaway is the second layer. A shopper asking for a "waterproof hiking backpack for a two-week trip under €70" triggers several parallel searches at once. A product missing a material attribute falls out of the waterproofness sub-query even if it meets that condition in real life.

How visual search changes image requirements

Product discovery no longer starts with text alone.

Shoppers can take a photo, search from an image, scan a product in a store, or circle an item on a phone screen to find similar ones. Google Lens then shows price, promotions, reviews, and places to buy, and Circle to Search brings that behavior into social media, videos, and apps.

For the feed, that means two separate requirements that are easy to mix up.

The image must match a specific SKU. If you sell a jacket in five colors, each variant needs its own photo in that color. One shared image across variants breaks visual matching.

Attributes must describe in structured form what is visible in the photo. Matching still runs on fields, not pixels.

Google AI Mode results for a conversational travel bag query - conversation on the left, product offers on the right

Virtual try-on

Google is expanding virtual try-on for apparel. Eligible products can receive the relevant marker in free product listings and product ads. The Merchant Center documentation is specific: a supported apparel category, an image of at least 512 by 512 pixels (ideally above 1024), and one garment shown in full, without hands, accessories, or wrinkles covering details [3].

That is a clear example of how technical image requirements stopped being about aesthetics and became a condition for accessing a feature.

Google Shopping virtual try-on - from query through look generation to the result on a model

Your Merchant Center feed is now your most important SEO asset

In AI-driven shopping results, the product feed does a lot of the work that category pages used to do in classic SEO.

Merchant Center uses product attributes to understand what you sell, match products to queries, and show correct details in ads and free listings. When that data is incomplete or wrong, Google flags issues in Merchant Center and ads may not serve [4].

Attributes that matter most

Attribute Why it matters
Product title The first thing Google reads. It should include brand, product type, and the main differentiating trait
Description Adds context: use case, features, material, fit, compatibility, audience
GTIN, brand, MPN Identify the exact product and link it to the same item from other sellers
Price Must match the landing page and stay current
Availability Tells Google whether the product is in stock, out of stock, preorder, or backorder
Image link The main image in ads and free listings. Google recommends high resolution, around 1500 by 1500 pixels or higher
Category and product type Classify the product and place it in results
Color, size, material, gender, age group Drive matching for more detailed queries, especially in fashion, home, and accessories
On-page structured data Helps Google connect the product page with Merchant Center data

Conversational attributes

In May 2026, around Google Marketing Live, Google added six optional attributes described explicitly as conversational. They help AI Mode, Gemini, and Business Agent understand the product in natural language [5].

Attribute What it describes
question_and_answer Answers to product questions, stored as question-and-answer pairs
related_product Links to other products in the catalog: accessories, spare parts, substitutes
document_link Links to documents such as manuals, catalogs, or specifications
item_group_title A shared title for a variant group, used with item_group_id
variant_option Traits that distinguish a variant, stored as name-and-value pairs
popularity_rank The product's popularity position in your catalog, as a percentage

In practice, start with the first three. Question and answer, related product, and document link did not exist in a standard feed at all, so they carry information Google cannot invent elsewhere. The other three refine things already covered by item_group_id, color, and size, so they can wait.

It is worth separating two field groups, because they are easy to confuse. Product highlight and product detail are not conversational attributes, even though Google reads them. Google also says not to duplicate content: if a fact is already in the description, product highlight, or product detail, do not repeat it in conversational attributes.

All six fields are optional, and adding them does not change approval status for products already in Merchant Center. Google recommends sending them through a supplemental data source so they stay separate from the primary feed.

Product feed is the foundation.

See how Snomi prepares your assortment for AI shopping - Google Shopping, AI Mode, and ChatGPT Ads.

See how

Price tracking and agentic checkout: what to prepare

Google is building a path where a shopper tracks a product price, picks a size or color, sets a target amount, gets a drop alert, and confirms the purchase. For eligible sellers, Google can add the product to the cart and finish payment with Google Pay.

For sellers outside the first rollout markets, two points matter.

Agentic checkout is still limited to selected markets and sellers. Google described it as rolling out first for product offers in the United States, in English, with sellers that accept Google Pay. Elsewhere, the transaction still completes on your site.

Price tracking, however, needs real-time accuracy: base price, sale price, variant prices, and availability. If a shopper waits for a discount alert, the data behind that alert must match what they see after arriving at the store. A mismatch at that stage costs more than missing visibility, because you lose a shopper with the highest intent.

It is also worth watching Universal Commerce Protocol, an open protocol Google describes as a payment standard using credentials stored in the Google wallet.

What this means for sellers today

Many markets are in a transition period. AI Mode answers shopping queries but leans on page content rather than Shopping Graph offers. Shops win when they have well-written categories, guides, and product pages that models can read, plus presence in external sources, because the model likes citing comparisons and discussions.

When Shopping Graph is wired into those results, the criterion shifts from content to product data. Whether your product enters a comparison at all will then depend on the state of the feed at that moment.

Cleaning a catalog takes weeks, not days: filling attributes across tens of thousands of products, fixing variant structure, syncing stock. That work has to happen before the switch, not after it.

You do not have to do this by hand

Everything above can be checked manually, but with a catalog of tens of thousands of products it takes weeks, and the result goes stale with every assortment change.

Snomi includes an audit module that walks the feed and flags concrete products to fix: missing attributes, titles that are too short, inconsistent variants, availability mismatches, and empty conversational fields. Most of those issues can then be optimized in the tool itself, without touching the source feed in the shop.

Create an account and try it free for 7 days to see how many products in your catalog need work.

Why feed quality decides visibility now

Shoppers have moved past keywords: they ask questions in natural language, search visually, try on virtually, and track prices.

The Merchant Center feed is what lets Google understand what each product is, which category it belongs to, whether it is available, how much it costs, and which shopper it fits.

Product data is becoming the interface between the store and AI-driven discovery. A visually polished product page will not make up for missing attributes, inconsistent variants, or stale availability. Structured, current catalog data will increasingly decide whether AI can recommend the right product at all.

Run a free product feed audit

See how many products have titles that are too short, missing attributes, or are not ready for AI. It takes five minutes.

Try for free

Frequently asked questions


Sources

[1] Google I/O, May 2026 - Shopping Graph: more than 60 billion offers, more than 2 billion refreshed hourly [2] Google, October 2025 - Google Lens: more than 25 billion searches a month, about 1 in 5 with shopping intent [3] Google Merchant Center Help - image requirements and virtual try-on [4] Google Merchant Center Help - Product data specification [5] Google Marketing Live, May 2026 - conversational attributes in the product specification

Figures current as of 12 September 2026. AI Mode and Shopping Graph features often roll out by market in stages, so check the latest Google documentation before operational decisions.

KNOWLEDGE BASE