How Merchant Center sends product facts to AI Mode

Merchant Center is Google’s service for sending product information such as names, prices, availability, images, and other required attributes. Conversational attributes are not a second product listing. They are an optional extension for details that are difficult to express in a short title or standard product field.

Google says these details help AI systems and conversational agents understand product nuances. That is a statement about the intended use of the data, not a promise that every submitted sentence will appear in a user-facing answer or change a product’s position.

• The main product record remains the base data source.
• Extra attributes are for useful clarifications, not repeated copy.

Section sources:[1] Google Merchant Center Help

The documentation lists several field types. question_and_answer stores a user query and a brand answer; document_link points to supporting documentation; related_product connects one product to another; item_group_title describes a shared group. variant_option covers product variants, while popularity_rank can carry a product’s position in a real ranking.

The field names remain in their original form because they are the names used in a data source or API. Their meaning matters more than the label: a merchant should provide a concrete, checkable fact rather than a promotional phrase. Google also says not to duplicate details already present in description, product_highlight, or product_detail.

• Do not add a field merely to repeat a keyword.
• Do not present a subjective marketing claim as a ranking or verified fact.
• Check that document_link is reachable and supports the statement.

Section sources:[1] Google Merchant Center Help

Google describes two practical routes: conversational details can be included in a supplemental data source or the primary data source, and they can also be submitted through the Merchant API. A supplemental source is an additional set of data that complements the main catalog; it does not remove the requirements for the base product record. The Merchant API is a programmatic interface for automated updates.

The right route depends on where the catalog is maintained and who updates it. Before shipping, record the schema version, submission date, product identifier, and exact text of each extra fact. This makes it possible to separate a transmission problem from a case where Google received the data but did not show it in a particular response.

• Validate the primary product source first.
• Add only details with a clear origin and review date.
• Keep a submission log separate from observations about AI Mode displays.

Section sources:[1] Google Merchant Center Help

In a September 16, 2026 publication, Google says that during testing with lululemon, attributes submitted by the brand were incorporated into 50% of relevant product recommendations in AI Mode. This is useful evidence of a possible path for product facts to enter an answer system.

The page does not provide the total number of recommendations measured, sample size, query set, repetition period, recommendation count, or independent validation. The careful wording is therefore: Google reported 50% incorporation in relevant recommendations in one test with one brand. It should not be rewritten as “Merchant Center raises ranking by 50%” or “every product will be shown in AI Mode.”

• Observation: 50% is the number in Google’s publication.
• Unknown: the number of runs and recommendations.
• Inference: treat it as a signal for measurement, not a universal benchmark.

Section sources:[2] Google

Separate four events: Google received the data; the product appeared in an AI Mode response; AI Mode used the specific fact; and a user clicked or purchased. Conversational attributes may matter at the first or second level, but the documentation does not prove the third or fourth. A bot visit to a page is also not proof that the page was named as a source in an answer.

For an internal test, fix the product set, language, country, date, availability, and exact user queries. Save the full response, visible links, and whether the specific attribute appeared. Compare groups with the same period and do not change price, description, images, and extra fields at the same time; otherwise causality cannot be isolated.

• Data availability is not a display.
• A display is not use of a specific fact.
• Use of a fact is not a click or purchase.

Section sources:[1] Google Merchant Center Help[2] Google

The official pages do not promise a position in AI Mode, a required display, a website link, the brand being named as a source, a click, or a sale. They also do not say that adding an attribute by itself is a universal ranking factor. Google describes a possible way to help AI systems understand product nuance and reports one test result; that is a narrower claim.

A useful strategy is therefore not to fill fields mechanically, but to send a small number of precise, checkable details missing from the main product record. If an attribute duplicates the description or has no evidence, it adds measurement noise rather than independent value.

• Google does not publish the total number of recommendations measured in the lululemon test.
• There is no evidence of a universal visibility lift.
• There is no guarantee of display, link, click, purchase, or causality.

Section sources:[1] Google Merchant Center Help[2] Google