Are these recommendations paid?

Organic recommendations are separate from ads on the major platforms, though sponsored placements are being tested and sold in some shopping surfaces. Label your reports by surface.

For branded products sold by multiple retailers, yes. Identifiers let assistants match your listing to the same product elsewhere and to its reviews.

How often should feeds update?

Price and stock should update whenever they change. Assistants that recommend an out-of-stock item learn to trust the source less.

Engine guide: shopping

How to get recommended by AI shopping assistants

Shopping assistants do not browse your store the way a person does. They read feeds, product markup and reviews, then match attributes to the shopper's stated needs.

The short answer

AI shopping assistants recommend products using structured data: merchant feeds such as Google Merchant Center and OpenAI's product feed, schema.org Product markup, marketplace listings and review text. They match attributes against what the shopper asked for. To be recommended, fill every relevant attribute in your feeds, keep price and stock current, and collect reviews that mention use cases.

What ranked on 26 September 2026: Commerce platform blogs and feed-tool vendors dominate this query. The useful primary sources are the merchant documentation pages from Google and OpenAI, which spell out the required and recommended product attributes.

Where each shopping assistant gets its data

There is no single shopping index. Each assistant draws on its owner's own catalogue first, which is why a product can appear in one and not the others.

Google's AI shopping features draw on the Shopping Graph, fed by Merchant Center feeds and crawled product pages
ChatGPT shopping uses merchant feeds submitted to OpenAI plus pages crawled by OAI-SearchBot
Amazon's Rufus answers from Amazon's own catalogue, listings, reviews and customer questions, so off-Amazon content matters little there
Review text is mined for use cases. A review saying a bag fits a 16-inch laptop answers a query no spec sheet does
Returns, shipping speed and availability are attributes too, and assistants can filter on them

Things that sound right and are not

ClaimOur product descriptions are great, so we will be recommended.
Marketing prose is the least structured part of a listing. Assistants match on attributes such as size, material and compatibility. An empty attribute is a filter you fail.
ClaimOne feed covers every assistant.
Google, OpenAI and Amazon use different specifications and different submission routes. Map one master catalogue to each spec rather than hoping one export fits all.
ClaimStar rating is what matters.
Ratings help ranking, but what the reviews say decides the match. Prompting buyers to mention how they use the product produces the text assistants quote.

Where to start

01Audit your Merchant Center feed for missing recommended attributes and fix the ones shoppers filter on.
02Make sure product pages carry Product markup with price, availability and GTIN matching the feed.
03Ask assistants the needs-based questions your buyers ask, such as best waterproof jacket under a set price, and log who is named.
04Change your post-purchase review request to ask what the customer uses the product for.
05If you sell on Amazon, answer customer questions on listings, since Rufus reads them.

Questions

Are these recommendations paid?: Organic recommendations are separate from ads on the major platforms, though sponsored placements are being tested and sold in some shopping surfaces. Label your reports by surface.
Do we need GTINs?: For branded products sold by multiple retailers, yes. Identifiers let assistants match your listing to the same product elsewhere and to its reviews.
How often should feeds update?: Price and stock should update whenever they change. Assistants that recommend an out-of-stock item learn to trust the source less.

Doing this yourself and want a second opinion on what the assistants say about you? Tell me what you are seeing, or read the AEO Playbook for the whole framework.