Short answer
Fix price, availability, identifiers and variants first. Then add optional conversational attributes that answer real questions without duplicating product data.
The answer in detail
Start with the fields that can make an AI answer wrong: price, availability, brand, GTIN, color, size and variant grouping. Optional conversational attributes come after that foundation. A missing enrichment field matters less than a product recommendation that contradicts the checkout.
The benchmark
Google's current Merchant Center guide lists six optional conversational attributes. They complement primary product data and do not change existing product approval status. They are not a guarantee of AI placement or a shortcut to direct checkout.
The following limits were checked against Google’s primary attribute specifications on October 29, 2026:
| Attribute | Purpose | Checked limit |
|---|---|---|
question_and_answer |
Product FAQs | 30 pairs; 1,000 characters per question or answer; 10,000 total |
document_link |
Public PDF documentation | Five links; 50 MB per file |
related_product |
Accessories, alternatives and other relationships | 30 linked products |
item_group_title |
Shared variant-group title | 150 characters |
variant_option |
Variant differences | 30 name/value pairs |
popularity_rank |
Relative popularity within a category | 0.0–100.0, one decimal place |
Two established attributes also help: product_highlight for concise benefits and product_detail for factual name/value specifications. The source reports Google's recommendation of four to six highlights.
Core requirements depend on product category, destination and use. Price and availability remain fundamental; brand and valid identifiers have their own conditions. The source's clothing checklist specifically concerned France. For an international feed, consult the current destination-specific specification for color, gender, age group, size and item_group_id; do not assume that one French checklist governs every product worldwide.
The source also mentions an AI performance report available in a limited set of countries. Report access, attribute submission and consumer-facing AI shopping availability are separate questions. Check your account rather than interpreting a missing report as proof that the feed cannot be used.
The rule
Prioritize errors by the customer harm they cause. Wrong stock or price can recommend an unavailable product. Popularity and related items become useful only after basic data is trustworthy.
Write information once in the appropriate field. Google's guide says not to duplicate existing description, highlight and detail data in conversational attributes. Use FAQs for remaining questions about use, care or compatibility. Price, promotions and changing dates belong in their dedicated fields, not in answers that will become stale.
Google can extract FAQ information from linked PDFs. Do not repeat an entire manual as questions simply to occupy more fields. A useful answer is specific and supported by the product documentation.
The steps
- Open Merchant Center's product-issue diagnostics and inspect all relevant problems, not just prioritized fixes. Correct price, stock, identifiers and color first. On Shopify, review suggested category metafields individually; the source reports a home-equipment brand finding both useful and irrelevant suggestions. See Google Shopping product data and GTIN guidance.
- Check that variants share the correct
item_group_id. Use consistent option names, then add shared titles and variant distinctions. - Use a supplemental source where appropriate to enrich app-synchronized data. The source describes enabling Advanced data source management before adding supplemental data. Product IDs and the matching feed label must align exactly; confirm current interface labels and matching rules.
- For 20 bestsellers, complete factual details and highlights, link public manuals and write five to ten questions drawn from support conversations. Omit information already supplied elsewhere.
- Calculate popularity consistently. The source proposes—not Google mandates—
100 × number of lower-selling products in the category ÷ category product count, using recent sales. In its fictional example, third of 40 scores100 × 37 ÷ 40 = 92.5. Document ties, time window and category boundaries; do not describe this as Google's official ranking formula. - Review sample AI answers against the source data monthly. Use the catalog audit for the broader process. Record the query, market and date because results vary.
The pitfall
Treating new fields as keyword inventory makes answers less useful and conflicts with the cited Google guidance. Write as a knowledgeable store associate would answer, using verifiable facts.
Supplemental data also has an owner and maintenance cost. If composition changes in the primary catalog while the supplemental FAQ stays unchanged, the price can be correct and the product advice wrong. Include supplemental sources in every product-change workflow. An outdoor brand's duplicate title formats were discovered through an Ads audit, showing why reading one export alone may miss synchronization problems.
What to tell your team
“In this fictional audit, 18% of products lack relevant color or identifier data. We will resolve the valid missing fields within two weeks, then enrich the 20 bestsellers responsible for 60% of revenue. Better data comes before a larger FAQ.”
