How to optimize your product feed for AI search

Each destination has its own feed specification, and the required fields differ. Check the official specification for your destination.

OpenAI Commerce product feed (native schema)

The native OpenAI product feed specification requires, on every row: item_id, title, description, url, image_url, price (with an ISO 4217 currency code), availability, and brand. It also requires both eligibility flags, is_eligible_search and is_eligible_checkout. When checkout is enabled by setting is_eligible_checkout to true, is_eligible_search must also be true; a false checkout value does not require search eligibility to be true. The is_ads_eligible flag is required for Ads processing and optional for non-Ads feeds. Required merchant fields include seller_name, seller_url, return_policy, target_countries, and store_country; seller_privacy_policy and seller_tos are required when checkout is enabled. Other identifiers (gtin, mpn, condition, dimensions, and more) are optional or recommended on each row.

OpenAI Commerce product feed (Google-compatible upload path)

If OpenAI confirms that your registered feed supports the Google-compatible format, you can upload a Google-format delimited feed without renaming columns. The required fields on every row are then id, title, description, link, image_link, availability, price, and brand. OpenAI maps these to its native schema (for example id to item_id, link to url, and image_link to image_url). This path is conditional on OpenAI confirming your feed supports it; otherwise use the native OpenAI specification.

Google Merchant Center

Google Merchant Center requires id, title, description, link, image_link, availability, and price on every row. The identifiers brand, gtin (or mpn), and condition are conditional on product type and category: brand depends on product type, mpn is required only in specified no-GTIN cases, identifier requirements depend on category, and condition is optional for new products. Submit a valid gtin when one is assigned. If no GTIN is assigned but the product has a brand and mpn, submit both and keep identifier_exists at its default of true. Set identifier_exists to false only when the product truly lacks the applicable identifiers under Google's category-specific rules. An incorrect false value can result in a warning, disapproval, or limited performance.

Keep the title factual and structured (brand plus product plus a key attribute). Google's tips to optimize your product data put the important attributes at the front of the title and ask for the most accurate data you have. Use a single source of truth for these fields so the feed and the product page agree. Missing required fields are the most common reason a feed is rejected or a product loses eligibility.

• OpenAI Product Feed Specification: developers.openai.com/commerce/specs/file-upload/products
• Google Merchant Center product data specification: support.google.com/merchants/answer/7052112
• Google Merchant Center tips to optimize your product data: support.google.com/merchants/answer/7380908
• Google Merchant Center identifier exists specification: support.google.com/merchants/answer/6324478

Keep identifiers, attributes and availability aligned at the source. Paz handles these corrections through the product feeds service.

Stale price or availability is a common cause of a poor shopper experience and can reduce eligibility. Google recommends keeping Merchant Center information current. Submit updates on a cadence that matches how often price and inventory change for your catalog, and use the same values on the feed and the visible product page.

If Paz handles your feeds, we keep them current on the agreed schedule and track what each engine surfaces for your queries, so you can see whether current data is associated with appearances. Platform inclusion is controlled by the platform; Paz records appearances as observations.

Beyond the required fields, add the attributes a shopper in your category actually filters on: size, color, material, and fit for apparel; dimensions, compatibility, and power for electronics; dimensions, material, and capacity for home goods. The exact attributes a platform uses are platform-specific and can change, so check the platform documentation for your category.

Submit a valid gtin when one is assigned. If the GTIN is unassigned but the product has a brand and mpn, submit both and do not set identifier_exists to false. Use identifier_exists=false only when the product truly lacks the applicable GTIN, brand, and MPN identifiers under Google's category-specific rules. Submit only manufacturer-assigned identifiers.

Add Schema.org Product structured data to every product detail page. Include name, description, image, and an offers block with price, priceCurrency, and availability. Where available, add brand, gtin or mpn or sku, and aggregateRating.

Keep the markup consistent with the feed and with the content a shopper sees on the page. Google guidance requires structured data to match the visible content; when the markup and the feed or page disagree (different price or availability), Google may take manual action or limit rich-result eligibility. Pull the feed and the markup from the same catalog source of truth so they agree.

• Google Product structured data: developers.google.com/search/docs/appearance/structured-data/product
• Schema.org Product: schema.org/Product and Schema.org Offer: schema.org/Offer

ChatGPT Shopping ingests product feeds through OpenAI's commerce feed specification. Google AI Mode and Google Shopping draw from Google Merchant Center, which accepts feed formats documented in the product data specification. Each platform publishes its own supported format and submission path; submit the format the platform accepts for your account.

As a practical inference from OpenAI's documentation, meeting its feed requirements establishes eligibility, and selection and placement are decided by the platform. Google states that normal search eligibility is the requirement to appear in its AI search features, and recommends matching structured data to visible content and keeping Merchant Center information current. Treat feed compliance as eligibility.

• OpenAI Product Feed Specification: developers.openai.com/commerce/specs/file-upload/products
• Google Merchant Center product data specification: support.google.com/merchants/answer/7052112
• Google AI features and your website: developers.google.com/search/docs/appearance/ai-features

Before and after submission, validate the feed and the structured data with the platform's own tools. Google Merchant Center reports feed errors and diagnostics; Google Search Console and the Rich Results test report structured-data issues. Fix the specific errors the validator reports before resubmitting.

Validation confirms a feed is well-formed and eligible. Appearance in any specific answer is decided by the platform. Treat validation as a floor for eligibility, not a guarantee of visibility.

Once the required and category-specific fields are complete and validated, improve the context that helps an engine understand the product: clear descriptions, accurate attributes, and consistent taxonomy. Keep factual fields (price, GTIN, availability) accurate.

Two primary sources describe this step. OpenAI's feed best practices ask for concise, factual descriptions, intentional use of optional fields such as description.html and categories.taxonomy, valid encoded URLs, consistent seller links and variants modelled at row level. Google's tips to optimize your product data put key attributes at the front of the title, ask for the highest-resolution images available and require the product data to match the landing page.

This is the core of Paz's product feeds service: a Paz expert finds the attributes that matter for your categories, drafts them from your sources, and publishes within agreed guidelines and brings exceptions to you.

Optimization is only useful if you can tell whether it worked. Run a set of representative shopper queries on the surfaces you care about and record whether your products appear, in what position, and how that changes over time. Measure on a cadence, not once, because results vary by session and over time.

Paz measures this as AI visibility across Google AI Mode, ChatGPT, Perplexity, and Google AI Overviews on a scheduled cadence agreed for the job, including the configured queries, engines and product coverage. See the methodology page for definitions and limitations. Paz records what an engine surfaces as observations.

Fields shown are summaries. Always check the official source for the current, complete, category-specific requirements. Platform requirements change.

• Marketing copy in the title field. Titles are parsed as structured identifiers. Keep titles factual (brand plus product plus a key attribute) and move persuasion into the description.
• Stale price or availability. Submit updates on a cadence that matches how often they change, and keep the feed and the page in sync. Stale data can reduce eligibility and create a poor shopper experience.
• Inconsistent category taxonomy. Mixed or custom taxonomies fragment a catalog. Use a consistent taxonomy across every product.
• Unstructured reviews. Reviews trapped in page HTML may not be usable as signals. Syndicate them as structured data where appropriate.
• Invented or missing identifiers. Submit an assigned GTIN when available. When the GTIN is unassigned, submit brand and MPN when the product has them. Set identifier_exists to false only when the product truly lacks the applicable identifiers, and submit only assigned values.

Completing required fields, adding category-specific attributes, validating, and keeping data current improve a product's eligibility and make it easier for an engine to understand. Each platform then decides whether a product appears, ranks, is cited or remains available. For OpenAI, that conclusion is a practical inference from the distinction between feed eligibility and platform selection, not a quoted OpenAI guarantee statement. Google states that normal search eligibility is the requirement to appear in its AI search features.

Platform inclusion, ranking, publication, and availability are controlled by the platform, not by any feed optimization. Treat visibility as something to measure, not something to guarantee.