AI Shopping Surfaces Make Product Data Integrity a Technical Ops Job

On May 20, 2025, Google announced a shopping experience in AI Mode that it described as supporting the full path from inspiration to purchase. In that same announcement, Google said the experience combines Gemini capabilities with the Shopping Graph, which now holds more than 50 billion product listings and refreshes more than 2 billion of them every hour. That matters for a simple reason: Google is not treating AI shopping as a loose recommendation layer. It is grounding the experience in product data such as reviews, prices, color options, availability, and freshness. For merchants, that shifts catalog integrity out of the SEO backlog and into day-to-day operations.

AI shopping runs on merchant data

The business implication is hard to ignore. If Google is using merchant data inside AI-assisted shopping flows, then product data quality is no longer a back-office cleanup task. It becomes part of what customers see while they compare options, judge price, and decide whether to buy. Google’s merchant listing documentation reinforces that by requiring an active price and currency for merchant listing experiences, recommending availability, and documenting shipping and return properties that can travel with the offer itself.

That is why this now belongs with technical operations. A well-written product page is not enough if the underlying offer data conflicts across the visible page, structured data, feed fields, and Google-managed settings. Google explicitly recommends putting Product structured data in the initial HTML for best results. It also warns that JavaScript-generated Product markup can make Shopping crawls less frequent and less reliable, especially for fast-changing details like price and availability. If pricing or stock moves often, that is not a theoretical concern. It is an engineering constraint.

Variants are not a cosmetic detail

Variant handling is where a lot of stores quietly break down. Google’s product variant documentation says that variant markup can make products eligible to appear with variant information in merchant listing experiences. It expects merchants to model the relationship clearly using ProductGroup and Product, with identifiers linking the parent group to each individual variant. Each variant needs its own unique ID, and the product group needs one too.

Google goes further than markup. It says the site must be able to preselect each variant directly with a distinct URL so Google can crawl and identify that specific version. Those preselected URLs are expected to show the correct image, price, and availability, and still let the shopper add that exact variant to the cart. For single-page setups, Google expects one canonical URL for the overall ProductGroup. For multi-page setups, each page must include full, self-contained markup for the entities defined there.

In practice, this touches routing, canonical logic, template output, and catalog structure. Size, color, pattern, or material variants cannot just live as informal UI states. If someone lands on a green medium product, Google wants to understand that it is the green medium product, not a generic parent page with a client-side selector layered over it. Stores that blur that distinction create uncertainty around what is actually for sale, and uncertainty is exactly what AI shopping systems are trying to remove.

Returns and shipping now influence the visible result

Google’s return policy documentation makes the same point from a different angle. It says MerchantReturnPolicy can be used so Google Search may show return policy information alongside products and in knowledge panels. A standard business-wide policy can be nested under Organization, while product-level return policies can override it for specific items. Merchant listing guidance also states that if both organization-level and product-level return policy markup are present, Google defaults to the product-level return policy.

Search Console adds another operational layer. Google says online merchants can use Search Console to view, add, and manage shipping and return policies under Settings > Shopping > Shipping and returns, including delivery time, shipping cost, return window, and return cost. Shipping details are automatically approved, but return policies are manually verified and can take about 10 to 13 days. Google also warns that shipping costs submitted to Google should exactly match the shipping costs shown on the website, and that a listing may be rejected if Google finds a lower shipping cost on the listing than on the site.

That is the shift many teams underestimate. Return and shipping data is no longer passive policy copy sitting on a footer page. Google treats it as structured, customer-facing commerce data with approval states, precedence rules, and rejection risk.

The precedence model matters when several teams or systems touch the same information. Google documents an order of precedence for shipping and return configurations: product-level feeds submitted in Merchant Center are strongest, followed by Content API settings, then settings in Merchant Center or Search Console, then product-level merchant listing markup, and finally organization-level markup. Those layers are not interchangeable duplicates. Someone needs to decide which one is authoritative and keep the weaker layers aligned.

AI-generated titles create a governance problem

Title quality now has a governance angle as well. Google’s Merchant Center help says product titles should clearly describe the product shown on the landing page, distinguish between variants, and avoid promotional copy such as shipping details or sale dates. It also says that all titles created using generative AI must be submitted with the structured_title attribute rather than the standard title field, and that the digital_source_type value must be trained_algorithmic_media. If both structured_title and title are provided, Google says it will use the plain title field.

That detail creates a real control problem. A merchandising team can think it has implemented the correct generative-AI field while an older export still fills title and quietly overrides it. The same help page also notes that structured_title is not supported by Schema.org at this time. In practical terms, AI-title governance sits in Merchant Center and feed management. On-page schema will not solve it by itself.

What the technical ops job actually looks like

For ecommerce teams, the response should not be a one-off schema cleanup. It needs to be a repeatable operating model that checks the same product across the page template, structured data, feed fields, and Google-managed settings.

Verify that core Product markup is present in the initial HTML, not only after client-side rendering.
Check that active price, currency, and availability stay consistent between the page, structured data, and feed output.
Make sure each variant has a unique ID, a clear group relationship, and a crawlable preselected URL that resolves to the correct image, price, and stock state.
Decide where shipping and return truth lives, whether that is Merchant Center, Search Console, product-level markup, or organization-level markup, then align the weaker layers to that source of truth.
Review product titles so variant details are explicit and any AI-generated titles are submitted through the correct field without being overridden.
Use Google’s validation and monitoring surfaces after template or merchandising changes so catalog updates do not silently break visibility.

This is the kind of work Greg can handle in a practical, hands-on way. That might mean auditing Merchant Center inputs against on-page structured data, fixing variant and return markup in WordPress or Drupal, cleaning up feed fields and AI-generated title disclosure, and putting simple repeatable checks in place so merchandising changes do not quietly erode visibility. Google’s own documentation now treats these data points as active parts of the buying experience. Stores that treat them as technical operations will be in a much better position for AI-assisted discovery than stores still treating them as occasional SEO maintenance.

Need help with this kind of work?

Talk to Greg about a product data integrity audit Get in touch with Greg.

Sources

Last modified

2026-06-07