Skip to content
All field notes

Google AI Mode / Gemini / Google Merchant Center / Product Data / AI Shopping

How to Optimize Product Data for Google AI Mode and Gemini

Prepare Merchant Center product data for Google AI surfaces, add optional conversational attributes, and use current reporting without treating visibility as guaranteed.

Matheus Reis

/ 9 min read

Optimize product data for Google AI Mode and Gemini in two stages. First, make the core Merchant Center record accurate and consistent with the product page and checkout. Then add Google’s optional conversational attributes where they supply useful facts that are not already represented.

The second stage cannot repair the first. A detailed FAQ does not fix the wrong GTIN. A variant-option field does not fix an availability mismatch. Google states that incorrect, missing, or conflicting product data can cause disapprovals, limited eligibility, or incorrect displays in Merchant Center programs.

Google also does not promise that completing a field will produce a particular AI answer or ranking. Treat the work as product-data quality and eligibility work, not a guaranteed visibility formula.

Establish the core Merchant Center record first

Google’s current product data specification governs the core record used for ads and free listings. Google’s conversational attributes are part of that same specification and explicitly complement the primary Merchant Center data, so this guide uses the core record as the first implementation stage.

Audit these areas before adding AI-specific fields:

AreaWhat to verify
IdentityStable ID, legitimate GTIN or MPN where applicable, brand, product category
Product copyTitle and description accurately identify the represented item and match the landing page
VariantsItem group ID and distinguishing values such as size, color, material, or capacity
OfferPrice, sale price, currency, availability, and availability date where required
MediaMain image represents the item; additional media is accessible and correctly licensed
FulfillmentShipping and return information matches the actual purchase conditions
DestinationProduct URL resolves to the same item and offer represented in the feed

Conflicts deserve priority over missing optional fields. If Merchant Center says an item is in stock while the product page or checkout says it is unavailable, add no enrichment until the systems agree.

Use the platform-neutral AI shopping product data checklist to audit the source record before translating it into Google’s schema.

Google’s six conversational attributes

Google currently documents six optional attributes designed to add product nuance for AI-driven and traditional search experiences. They complement the main Merchant Center specification.

AttributeWhat it representsSafe implementation rule
question_and_answerProduct-specific question and answer pairsUse approved factual answers; do not invent shopper questions or promises
document_linkRelated public PDF documentsLink the current manual, assembly guide, specification, or other relevant document
related_productDeclared relationships such as required parts, accessories, substitutes, or often-bought-with itemsUse stable identifiers and a relationship the merchant can defend
item_group_titleA shared name for a product familyUse with item group ID so variants have a clear parent identity
variant_optionName-and-value pairs that distinguish variantsRepresent the actual options a buyer can select
popularity_rankThe item’s popularity as a percentage relative to the merchant’s own inventoryDo not describe it as a global Google ranking or independent market popularity

Google says these attributes are optional, can be submitted through a supplemental data source, the primary source, or Merchant API, and do not affect the approval status of existing products. Google recommends a supplemental data source as one way to add them.

Optional status is not evidence that every catalog should populate every field. Add an attribute when the merchant has a current authoritative source and a process to maintain it.

1. Build Q&A from real decision gaps

Use question_and_answer for product questions whose answers are specific, stable, and supported by merchant documentation.

Good candidates include:

  • compatibility with a named product or standard;
  • whether a component is included;
  • care or installation requirements;
  • material or fit details not clear from the title;
  • a product-specific warranty condition.

Avoid answers that depend on a location, promotion, or stock state unless that value is maintained with the same discipline as the main offer. Do not use Q&A to repeat the title and description in different words.

Google explicitly says merchants do not need to duplicate details already submitted in description, product-highlight, or product-detail fields. Decide which field owns each fact.

Use document_link for a current, publicly accessible PDF that materially describes the product. A technical manual, compatibility chart, ingredient sheet, assembly guide, or warranty document may answer questions that do not fit cleanly in a product description.

Before submitting a link, verify:

  • the document matches the exact product or product family;
  • the revision is current;
  • the URL works without authentication;
  • the document does not conflict with the landing page;
  • the merchant has the right to distribute it.

A document library needs an owner. An obsolete manual can make a complete feed less accurate.

3. Declare relationships instead of implying them

Use related_product to connect products through a supported relationship and identifier. Google’s current examples include required parts, accessories, substitutes, and products often bought together.

Relationship data can be operationally sensitive. A required part should actually be required. An accessory should be compatible. A substitute should satisfy the relevant product need without creating a false equivalence.

Do not convert a merchandising preference into a technical compatibility claim.

4. Clarify product families and variants

Use item_group_title with item_group_id to give the family a clear shared identity. Use variant_option to represent the properties that make one purchasable variant different from another.

For every variant, confirm that the option values agree with:

  • its title;
  • its image;
  • its URL or selected landing-page state;
  • its price and availability;
  • the value displayed at checkout.

Variant consistency matters more than adding many option names. A field called finish on one item and surface on another may describe the same commercial choice but create an avoidable mapping problem.

5. Use popularity rank within its documented scope

Google defines popularity_rank relative to the merchant’s own inventory. A higher submitted value indicates that the item performs better than other products the same merchant sells.

That does not establish:

  • market share;
  • popularity across Google;
  • a guaranteed position in AI Mode or Gemini;
  • a reason a shopper should prefer the item.

Document the calculation and use a stable method across the submitted catalog. Omit the field when the underlying measure cannot be explained or maintained.

Submit conversational attributes without creating a second source of truth

Google currently supports three submission paths:

  1. a supplemental data source;
  2. the primary product data source;
  3. Merchant API.

A supplemental source can be useful when the core commerce platform does not yet model the new attributes. It also creates another mapping and update path, so define which system owns each field.

Use a simple governance record:

AttributeAuthoritative sourceSubmission pathOwnerValidation event
Product identityCommerce or product-information systemPrimary source or APICatalog ownerItem or variant change
Price and availabilityCommerce systemPrimary source or APICommerce operationsOffer-state change
Product Q&AApproved product knowledgeSupplemental source or APIProduct/content ownerProduct or policy revision
DocumentsControlled document librarySupplemental source or APIProduct/compliance ownerNew document revision
RelationshipsMerchandising or compatibility systemSupplemental source or APIMerchandising ownerAssortment change

The table is a control model, not a Google requirement. Its purpose is to stop the same fact from being edited independently in several systems.

Shopify merchants have a specific Google path

For Shopify Agentic Storefronts, product discovery in Google AI Mode and Gemini currently uses the Google & YouTube sales channel, which syncs product and store information to Google Merchant Center. Shopify instructs merchants to enable automatic product and shipping-information syncing.

The Shopify integration remains early access and is not available to every store. Direct-checkout eligibility is a separate issue from product discovery. See the Shopify Agentic Storefronts setup guide for the current channel and checkout distinctions.

Use AI performance insights only if the account has access

Google announced AI performance insights for Merchant Center in May 2026. Its current help documentation describes the report as a pilot for a limited number of US Merchant Center accounts, despite earlier expansion announcements.

If the report appears in the account, Google says it can provide views across AI Mode and AI Overviews, including:

  • share of voice relative to the competitor set available in Merchant Center;
  • query frequency and query type;
  • discovery, evaluation, and purchase phases;
  • product terms and structured attributes that may be missing;
  • organic AI traffic rather than paid-ad traffic.

Use those views to find questions for investigation. Do not treat share of voice as proof that one field caused visibility, or that the report captures every Google AI interaction.

If the report is absent, that does not imply a setup error. Continue with Merchant Center diagnostics, source consistency checks, and the channel’s available performance reporting.

Common optimization mistakes

Treating optional attributes as ranking factors

Google says the fields can help systems understand product nuance. It does not promise a ranking change for completing them.

Writing Q&A from a model without product review

Generated answers can introduce unsupported compatibility, warranty, or performance claims. Require approval from the owner of the underlying product fact.

Repeating the same content across every field

Duplication increases maintenance work without adding information. Use the field designed for the fact.

Optimizing the feed while the landing page disagrees

Google’s core specification requires important values such as title, description, price, and availability to match the purchase surface. Resolve those conflicts first.

Assuming current access is global

Conversational attributes, AI performance insights, and Shopify’s Google Agentic Storefronts integration have different rollout states. Check the live Merchant Center account and current help pages before writing an implementation plan.

Primary sources

The Google product and rollout details in this guide were verified on July 12, 2026.

Further reading

Written by

Matheus Reis Co-founder at kn8 · Ecommerce AI

Matheus Reis is a product executive and co-founder at kn8, building the Storefront Agent for ecommerce brands. He writes about AI in retail, agentic commerce, and the future of the buying experience.

Private beta / hands-on demo

See kn8 on your storefront.

Bring us one customer request. We’ll show how kn8 answers in chat and completes the task in your storefront.

  1. 01Your storefront
  2. 02A customer request
  3. 03Live walkthrough