AI Mode Now Sells in Canada. What Google Reads Before It Picks Your Product

On 4 November Google Canada announced that Canadian shoppers can now buy directly from eligible Canadian retailers without leaving AI Mode in Google Search or the Gemini app. The shopper asks, the assistant picks a product, and the purchase goes through Google Pay. The retailer stays the seller of record. The launch names lululemon, Shoppers Drug Mart, Simons and a set of Shopify merchants, and says that from next year any retailer will be able to switch direct checkout on from inside Merchant Center.

Google's own figure in the same post is that 59% of Canadian holiday shoppers plan to use an AI tool to help them shop this season. When the sale happens inside the assistant, the product page never gets its turn to persuade. Whatever Google read before it chose is the whole pitch.

The questions are three times longer, and they have one answer

A narrow corridor of hundreds of identical closed doors, with one thin teal ribbon of light running its full length to a single far door standing ajar and glowing warm. — A long, specific question does not browse. It travels past every door that almost fits and stops at one.

On Alphabet's third-quarter call on 29 October, Sundar Pichai said queries in AI Mode are on average three times longer than traditional searches, and that one in five now uses voice, images or documents. Philipp Schindler, Google's chief business officer, said the Universal Commerce Protocol now enables direct checkout for hundreds of thousands of brands across Google.

A query three times longer is not “rain jacket”. It is “a packable rain jacket for a wet Vancouver commute that fits over a blazer and does not look like hiking gear”. A question that specific has one good answer in your catalog, and the assistant has to find it from the facts it holds about each product. It cannot find “fits over a blazer” in a record that says “great jacket, ships fast”, and your theme and homepage never enter the decision, because they are not in that record.

The product it picks comes out of your feed

Plain wooden blocks riding a conveyor into a large glowing lantern and coming out the far side with tiny drawers pulled open, each holding a small lit object. — The feed is the record Google shops from. The new attributes are extra drawers in it, and they start out empty.

The record is your Merchant Center product feed. On Shopify the Google & YouTube channel builds it from the product data already in your admin, and WooCommerce has an equivalent plugin, so the feed can only say what the product record says. Until now it had no field for the kind of thing a shopper asks in a sentence.

There is now. At Google Marketing Live in May, Google introduced what it calls conversational attributes, in its own words a way for retailers to update product descriptions so they match the more conversational way people search. The Merchant API reference spells them out. A product can carry up to 30 question-and-answer pairs, each question and answer up to 1,000 characters and 10,000 in all. It can carry links to PDF manuals and assembly instructions, a list of related products, the title of the group a variant belongs to, and a popularity rank from 0 to 100 within your own inventory.

Read that list as what AI Mode wishes it knew about your products. The same announcement promised AI performance insights in Merchant Center, a report comparing your share of voice on Google's AI surfaces against similar brands, rolling out to Canada, the United States, Australia, India and New Zealand. Google is about to tell you how often it picked you, a number most merchants have never had.

Thirty questions per product is a writing job, not a settings job

The attributes are optional and an empty one costs nothing today. But a catalog with the answers filled in gives the assistant more to match a long query against, and that is the entire game. A sensible first pass, for the products that matter most:

Pull three months of support tickets, chat transcripts and review questions and group them by product. The questions shoppers already ask are the ones AI Mode will be asked.
For each product, write five to ten pairs from facts you can stand behind: dimensions and weight, what is in the box, materials and ingredients, what it fits or works with, how to care for it, the occasions it suits. Keep prices and dates out; they belong in the offer fields and go stale.
Do not restate the title or the description. The pairs are for what those fields do not already say.
Then reread the description. If the most common question is not answered there either, fix that first; it is the field every assistant reads, on every platform.

The page still has to agree with the feed

A glass vase on a plinth beside its twin made of layered teal sheets, a single beam passing through both and shining amber where their outlines meet but splintering where they differ. — Feed, page and markup describing the same product in the same words. Where they differ, the light breaks.

The feed is not the only thing Google reads. Its documentation for merchant listings says a product page carrying Product structured data is eligible across Google's shopping surfaces, and it lists what the markup should hold: name, image and an offer with a price are required; description, brand, availability, material, colour, size and a GTIN are recommended. It also says Google may attempt to verify that data before showing it, and that only a page where a shopper can actually buy the product qualifies.

So three copies of every product exist: the feed, the visible page and the markup on that page. Google compares them. A feed that says in stock while the page markup says out of stock, or a page with no Product markup at all, is a product Google has a reason to doubt at the moment it is choosing on someone's behalf. Most themes ship some of this markup, few ship all of it, and nobody checks 2,000 pages by hand.

Finding the pages that disagree, at catalog size

This is where CortexCommerce fits, on the structured data side of the problem. The free scan crawls your live storefront the way a crawler does and reports, per product page, exactly what is missing: Product structured data, meta title and description, Open Graph tags, the canonical link, image alt text and whether the description is thin. It also checks that Google-Extended and the other AI crawlers can reach the store at all. That turns “do my pages agree” from a 2,000-tab afternoon into one ranked list.

On Shopify, where it installs as the CortexGuard app, a theme app embed then adds Schema.org Product JSON-LD to every product page the theme left without it, filling only the gaps that are open so nothing is duplicated, and the next scan verifies the fix is live. By hand, keeping feed, page and markup in step is a job you do once and then stop doing. With a scan that re-checks after every change, it stays done.

We cannot tell you whether AI Mode will pick your product, and Google's own share of voice report will be the first honest measure of that. What we can say is that when it reads your feed and then checks your page, both will say the same complete thing. To see what your product pages are missing today, run a free scan at cortexcommerce.ai/scan and start with the pages it flags for structured data.

The questions are three times longer, and they have one answer

The product it picks comes out of your feed

Thirty questions per product is a writing job, not a settings job

The page still has to agree with the feed

Finding the pages that disagree, at catalog size

Sources