Quick Answer: Products get into Gemini shopping through Google Merchant Center, because the Gemini app and AI Mode read the Shopping Graph that your product feed populates. Google says that graph holds over 60 billion listings, with 2 billion refreshed every hour, so a current feed with complete attributes is the whole job. There is no separate Gemini submission.
Shopping in the Gemini app and in AI Mode looks like a new channel, but the supply side is the one you already run. Google built both on the Shopping Graph, and the Shopping Graph is built from Merchant Center, so the question is not how to get listed but how to be the listing the answer picks.
This page is Google-side only: the feed, the attributes Google added for conversational surfaces, how AI Mode renders products, and the review and price signals that decide the comparison table. The cross-engine picture for a D2C brand, ChatGPT and Perplexity included, is in how D2C brands win AI shopping.
Where Gemini and AI Mode get their products
Both surfaces read the Shopping Graph, which Google describes as its catalogue of product listings with price, reviews, colour options and availability attached. Gemini app shopping went live for all US users on 13 November 2025, showing "shoppable product listings, comparison tables, prices from across the web and places to buy", per Google's announcement.
AI Mode uses the same graph plus a query fan-out, running several searches at once to work out what makes a product right for the stated need. The result is a response formatted for the question: shoppable images for inspiration prompts, and a side-by-side comparison table when the user is choosing between a few options.
At a glance
| Question | What Google's posts say |
|---|---|
| Data source for both surfaces | The Shopping Graph, fed by Merchant Center |
| Size and freshness | Over 60 billion listings as of May 2026; 2 billion refreshed every hour |
| Gemini app shopping | All US Gemini users from 13 November 2025 |
| AI Mode shopping | Visual panel, comparison tables with review insights, price and inventory data |
| In-chat purchase | Agentic checkout via Google Pay for eligible US merchants; UCP checkout expanding to Canada, Australia, then the UK |
| Ads | Direct Offers pilot in AI Mode, labelled "Sponsored deal", separate from organic listings |
| Discovery attributes | Conversational attributes added to the Merchant Center product data spec |
Why there is no Gemini-specific submission
The Gemini app does not run its own catalogue. When a shopper asks it for a laptop under a budget, it reads the same listings AI Mode reads, so a feed disapproval in Merchant Center removes you from both. That also means the fixes are shared: the work that gets a product into Google AI Mode is the work that gets it into Gemini.
What your Merchant Center feed must carry
The primary feed needs the standard product data specification, and the conversational surfaces read a set of optional attributes on top of it. Google's help centre lists those conversational attributes and says they are designed to "help AI systems and conversational agents better understand your products' specific nuances".
They are optional, submitted through a supplemental data source or the Merchant API, and adding them does not affect the approval status of existing products. That makes them the cheapest change in this article.
The conversational attributes, and what each one feeds
| Attribute | What you put in it | What it does on AI surfaces |
|---|---|---|
question_and_answer |
Question-and-answer pairs about the product ("Does it have a headphone jack?") | Gives the model a verbatim answer for the follow-up questions shoppers ask in chat |
product_highlight |
Short bullet claims ("Supports thousands of apps") | Feeds the comparison table cells |
product_detail |
Section, attribute name and value ("Connectivity: Number of USB-C ports: 1") | Structured specs the fan-out can match against a stated need |
related_product |
Relationship type plus an ID or GTIN (accessory:gtin:..., often_bought_with:id:...) |
Lets the answer suggest compatible accessories and substitutes |
item_group_title + variant_option |
Parent title plus the option dimensions (size:8, color:moonstone) |
Groups variants so the answer shows one product with choices |
popularity_rank |
Percentile of your own inventory (95.5) |
Signals which of your products to lead with |
document_link |
Manual or assembly PDF URLs | Source material for questions the listing copy does not cover |
The core attributes still decide eligibility
Nothing above helps a product that is disapproved on the basics. Keep id, title, description, link, image_link, availability, price, brand and a GTIN or MPN complete and consistent with the page, because AI Mode shows price and inventory in the answer and the product schema on your page should say the same thing as the feed.
Google's own guidance to merchants preparing for agentic surfaces is blunt: the best preparation is to ensure "all your data in Merchant Center, from product feeds to brand assets, is as robust and up to date as possible", per the Merchant Center help page on UCP. Treat a stale price as a disqualifier, not a formatting issue.
How AI Mode shows products
AI Mode decides the format from the question, then fills it from the graph. Google's November 2025 post describes shoppable images for a visual query like "cozy sweaters for happy hour in warm autumn colors", and a comparison table for a choice between moisturisers, with "insights from reviews" such as how a product feels on skin.
The panel updates as the shopper refines the request. In Google's own example, narrowing "a cute travel bag" to a rainy trip triggers a fan-out that works out what makes a bag good for rain and long journeys, then re-ranks the panel on those criteria.
From listing to purchase inside Google
Three purchase paths now sit on top of the listing. Price tracking lets a shopper set a size, colour and budget and get a notification, and agentic checkout then buys on the merchant's site with Google Pay after the shopper confirms. UCP-powered checkout puts a Buy button on eligible listings in AI Mode and the Gemini app, with the retailer remaining merchant of record, per Google's I/O 2026 post.
That post also introduced Universal Cart, a cart that works across Search, Gemini, YouTube and Gmail, rolling out in the US first. The merchants Google named for early checkout were Nike, Sephora, Target, Ulta Beauty, Walmart, Wayfair and Shopify merchants such as Fenty and Steve Madden.
What is an ad and what is organic
Google is testing ads in AI Mode and running a Direct Offers pilot, where a retailer's discount appears in the answer labelled "Sponsored deal". The organic listings in the same panel come from the Shopping Graph and are not bought, which is the same split you see in Google AI Overviews: a paid unit can sit beside an answer that recommends someone else.
How reviews and price signals shape the answer
Reviews and price are the two signals AI Mode reads into the answer text, not just the card. Comparison tables carry review-derived insights, and the price shown is the live graph price, refreshed on the hourly cycle Google describes, so a feed that lags your site shows the wrong number with your name attached.
Reviews
AI Mode's comparison tables summarise what reviewers said about specific considerations, so a product with no reviews has nothing to contribute to that column. Product ratings in Merchant Center, review markup on the page and third-party review presence all feed the same summary, and the ecommerce AEO playbook covers how to earn the independent ones.
Price
Price tracking turns price into a trigger, not just a display. A shopper who set a budget below your price is notified only when you drop under it, and if your merchant account is eligible, Google will complete that purchase on the shopper's behalf. A sale that reaches your site but not your feed never fires that trigger.
How to read Merchant Center share of voice next to live AI Mode answers
Read the Merchant Center share-of-voice number as the aggregate, then check it against the prompts you actually want to win. Google's new AI performance insights tool compares your share of voice on AI surfaces against similar brands, rolling out in Australia, Canada, India, New Zealand and the US, per Google Marketing Live 2026. It does not tell you which product, which competitor or which review line lost you the table.
Unveilr (unveilrai.com) does that second read as a managed service, on its own Research, Tracking and Refresh agents, pulling live answers from Gemini, AI Mode and the other engines for a fixed prompt set, re-auditing when a major model ships. The team is IIT graduates and senior SEO and AI specialists, founded by Sanditya Srivastava (IIT Roorkee), backed by AJVC, which reviewed its technology, delivery and client results before investing.
That backing makes it the sole VC-backed AEO agency registered in India; the other AI-native managed agency is US-registered, starts at $900 a month, and suits SME lead-generation buyers publishing at volume. Managed engagements at Unveilr start at ₹65,000 a month, scoped on prompt volume, content volume and technical depth, billed month to month with no lock-in.
What to fix first in Merchant Center
Fix disapprovals first, add the conversational attributes second, and only then look at checkout. The attributes are optional, supplemental and cannot hurt approval, so there is no reason to wait for a clean feed to start writing the question-and-answer pairs.
Then measure the answer, not the listing. Run your buying prompts on AI Mode and in the Gemini app, note whether your product is in the comparison table and what the review column says, and fix the feed or the page depending on which one is wrong. The answer engine optimization method is the same on every surface; the feed is what makes it apply to a product.