Checkout moved into the answer. Customers write the line the shopper reads.
Google does not say how AI Mode ranks products. It does say what goes into the comparison table where the shopper chooses, and that table is where a product feed stops working.
On November 4, 2026, Google began letting Canadian shoppers buy from lululemon, Shoppers Drug Mart, Simons and select Shopify merchants without leaving AI Mode in Search or the Gemini app. The shopper never reaches the brand's site. The decision happens in an answer the engine writes, and a product feed buys a brand eligibility for that answer and nothing more. Google does not publish how the answer ranks products. It does publish what the answer shows the shopper at the moment of choice: a comparison table in which the seller's claims sit beside customers' reports. The engine picks the brand whose condition, meaning the reviews, mentions and reputation it already reads, was pulling the customer before the question was asked. Spend and access reach the feed and the offer. They do not reach what customers wrote.
A product feed buys eligibility, and in 2027 eligibility becomes the floor
Eligibility, in agentic checkout, is everything that lets AI Mode sell a product at all: the product feed, the Universal Commerce Protocol integration, Merchant Center data, accurate price and stock. Google's November 4, 2026 Canadian announcement says that in 2027 it will make simplified protocol integration available directly within Merchant Center, so retailers can enable direct checkout from their own accounts. Early access is an advantage until that date. After it, eligibility is the floor.
The comparison table in AI Mode, as Google describes it in its U.S. and India announcements, can be read line by line against the documents that name who supplies each line.
Price and stock: the seller, through the product feed.
Product attributes: the seller, through the Merchant Center data attributes that TechCrunch reported in January 2026 Google added so sellers can feature their items within AI search surfaces. Each one is the seller's statement about its own product.
Star rating: businesses, review aggregators, review sites and Google users, per Merchant Center Help. This is reputation written across the web, away from the brand's site, compressed into one figure.
Review insight: customers, in Google's words "insights from reviews."
Any competitor can fill the seller's lines on the same terms. No competitor can fill its rival's customer lines, and no feed can rewrite its own.
Google does not disclose how AI Mode ranks products. It discloses the table.
Google does not disclose the inputs AI Mode uses to rank products. Its January 2026 Direct Offers announcement says only that Google "already elevates the most relevant products to meet your search criteria," and it never says what relevance is measured on. No Google document says whether reviews feed that step. An argument about AI Mode that rests on the ranking rests on something no one outside Google can check.
The comparison table is documented. Google's November 2025 description of AI Mode shopping follows a shopper choosing among options "like moisturizers suited to your skin type," who sees "a comparison table with a side-by-side view of considerations specific to that product," "including insights from reviews (like how a moisturizer feels on skin)." The table is where the shopper chooses, and Google has said what goes in it.
The skeptic's reply is obvious. The Merchant Center attributes TechCrunch reported could carry a seller-written "for dry skin," which would put the seller in the skin-type column after all. It would put a claim there: the seller's statement about its own product, entered by the party that profits if the shopper believes it. Every competitor can enter the same claim at no cost, which leaves it unable to separate one moisturizer from the next. The review insight sits beside it as customers' report on that same criterion, how the moisturizer felt on their skin. Asked to name the consideration that answers the skin-type question, Google named the report, not the claim. The column that answers the shopper's question is the one the seller did not fill.
The moisturizer is Google's example, and the table is general. The November 2025 post says the experience began rolling out "in Search for categories like toys, health and beauty and electronics in the U.S.," and Google India's April 2026 announcement describes "a side-by-side view of considerations specific to that product, including insights from reviews to help you quickly understand the differences" with no product type attached.
Whatever the ranking weighs, the choice is made in a table where the only answer to the shopper's question not written by the seller is condition, set down before the shopper asked.
The strongest objection: money reaches the answer and the reviews
The strongest case against the claim that condition decides inside AI Mode is that spend reaches the answer. TechCrunch reported in January 2026 that Google will let brands offer a special discount to users while they are looking for a product recommendation in AI Mode; Google's own announcement names the pilot Direct Offers and labels the unit "Sponsored deal." Money reaches reviews as well: sampling programs, incentivized reviews and syndication of a brand's own reviews into Merchant Center all turn budget into review volume. The first Canadian retailers are large names with the resources to integrate early. On this view, money and access decide the purchase, and condition is a comforting story for brands that lack both.
Direct Offers shows where the advertiser's money enters. In Google's example, "Google already elevates the most relevant products to meet your search criteria." Then: "Now relevant retailers have an opportunity to also feature a special discount." A discount, "like a special 20% off discount," changes what the moisturizer costs. It does not change what customers wrote about how it feels on skin.
Review spend buys volume. In the United States, the Federal Trade Commission's final rule of August 2024 bars compensation or other incentives conditioned on reviews expressing a particular sentiment. Google holds its own paid inventory to the same line: Google Ads Help documents that review summaries in Search ads are aggregated from online reviews and cannot be edited by advertisers. The brand can deliver the reviews. It cannot write what they say. It can shape what they cover: a summary works from whatever reviewers chose to discuss, and a brand that prompts reviewers about fit, feel and durability decides which questions the insight line can answer. Prompting sets the topic. Customers still set the verdict.
Condition is also not wholly the brand's to earn. In the Canadian launch the retailer is seller of record, and a customer reviewing the brand's product is reviewing a purchase the retailer delivered and serviced. The brand's reviews will absorb the retailer's delivery and service. For a brand sold through retailers, choosing the retailer is a decision about condition, not only about distribution.
A discount does decide some sales. When two products' review lines are equal on the question asked, the discount decides.
The objection has a clear test. If answer engines sell placements that reach the decision with no review evidence beside them, spend sets the terms and this argument fails. While the table carries what customers said, a discount is one more number read against the review line.
The column that answers the shopper's question is the one the seller did not fill.
The test runs today in any category AI Mode covers. Put the category's real questions to it: which moisturizer suits dry skin, which feels light under makeup, which earbuds stay in during a run. Read the review insight beside your product in each answer. A question your reviews never address is a gap no feed can close, and an attribute claiming the answer does not close it either. The gap closes when customers are asked about use, such as fit, feel and durability, and say what they found. Condition is written before the shopper asks, and in the table the shopper reads, it is the one answer no seller can write.
Further reading
- Google Canada, announcement of direct checkout in AI Mode and the Gemini app for Canadian shoppers, the primary document for the launch blog.google/intl/en-ca/products/explore-get-answers
- Google, description of AI Mode shopping and agentic checkout, including review insights in comparison tables and Shopping Graph scale blog.google/products/shopping/agentic-checkout-holiday-ai-shopping
- Google Merchant Center Help, About Product ratings, documenting that star ratings are compiled from multiple independent sources support.google.com/merchants/answer/14620705
- TechCrunch, report on Google's announcement of the Universal Commerce Protocol, its co-developers, Merchant Center attributes and brand discount offers in AI Mode techcrunch.com/2026/01/11/google-announces-a-new-protocol
- Google India, announcement of AI shopping in Search and the Gemini app, showing that the comparison table with review insights is described with no product type attached blog.google/intl/en-in/products/explore-communicate/new-ways-google-is-using-ai-to-make-shopping-easier
- Google Ads Help, About Review summaries, documenting that review text in Search ads is aggregated from online reviews and cannot be edited by advertisers support.google.com/google-ads/answer/18760807
- Google, January 2026 announcement of the Direct Offers ads pilot in AI Mode, showing the paid discount added to products Google has already surfaced blog.google/products/ads-commerce/agentic-commerce-ai-tools-protocol-retailers-platforms
- Federal Trade Commission, announcement of the final rule banning fake reviews and testimonials, including the ban on incentives conditioned on review sentiment ftc.gov/news-events/news/press-releases/2024/08/federal-trade-commission-announces-final-rule-banning-fake-reviews-testimonials