Fill Google Merchant Center Attributes AI Mode Can Actually Use

Google's conversational attributes do not replace a clean product feed. They add the facts a shopper asks after the title has done its job: Does this have a headphone jack? Which width is this shoe? Where is the assembly PDF?

This is the worksheet. The Google AI Mode shopping audit is the five-stage operating system. Do not start here if Merchant Center and your storefront still disagree on price, stock, or URL.

Google's conversational attributes help page is the source for every field below. The attributes are optional. Google says they will not change the approval status of products you already submit. Google recommends a supplemental data source, or the Merchant API, instead of overloading a fragile primary file.

If a fact already lives in description, product_highlight, or product_detail, do not paste it again into question_and_answer. Google tells you not to duplicate.

Start from the core specification

The product data specification still decides whether the item can appear at all. Conversational fields sit on top.

Minimum before you add AI Mode attributes:

• Stable id
• Accurate title, description, link, image_link
• availability and price that match the live offer
• brand plus gtin or mpn where required
• item_group_id when the shopper can buy more than one variant

Shopify stores that use Google & YouTube already sync a primary feed. Put conversational fields in a supplemental source keyed on the same id. WooCommerce and other platforms submit the same attributes through a Merchant Center file or the Merchant API. The field names do not change.

The attribute set, in the order you should fill it

Google's own example on the conversational-attributes page is a Pixel phone variant: one id, one item_group_id, a family title, variant options for display / memory / colour, Q&A about the headphone jack and Bluetooth, highlights, details, related accessories, a manual PDF, and a popularity rank of 95.5. Copy the structure. Do not copy the phone.

Write a before-and-after row for one real product

Use a product you actually sell. Apparel, bags, furniture, and device accessories show the gaps fastest.

Before (what many feeds still send)

A shopper who asks for a 26-litre clamshell backpack that fits a 16-inch laptop will not find this record unless luck intervenes. The feed never said those facts.

After (same SKU, conversational-ready)

Notice what moved: identity, family, readable variant, measurable constraints, and questions the description does not already answer.

Apparel example

A wool sweater feed that only says Navy Knit will lose “machine washable merino crewneck, women's medium, not itchy.” Put fiber, care, and fit in product_detail. Put “Is it machine washable?” and “Does it pill in the first month?” in Q&A only if those answers are true and not already in the details. Size belongs on variant_option, not in a paragraph the model has to guess from.

Furniture example

A dining table that ships in two cartons needs related_product for the required hardware kit if that kit is a separate SKU. The Q&A should answer assembly time and whether chairs are included. Dimensions go in product_detail. A lifestyle title that omits seat count wastes the conversation.

Common submission errors

Re-open the Merchant Center product after the supplemental upload. If the attribute is not on the item, the transport failed. Fix the join on id before you write more copy.

Q&A is a repeated group attribute. Colons, commas, and quotes inside a question or answer will split the row if you treat a CSV like a document. Google recommends TSV or separate question_and_answer columns, and tells you to escape special characters. If Merchant Center shows one mangled FAQ instead of five, the file broke before Google ever judged the copy.

Respect the official limits

From Google's specification:

• product_highlight: optional. Google's product highlight spec allows 1–150 characters per highlight, requires at least 2 and at most 100, and recommends 4–6. Product only. No promotional shouting, no keyword lists, no comparisons, no compatible-product essays, no repeat of details you already submitted elsewhere.
• product_detail: name/value pairs with their own character limits. Use real spec names, not marketing adjectives.
• question_and_answer: group attribute. Max 30 pairs. Max 10,000 characters total. Each question and each answer max 1,000 characters. Both sub-attributes required.
• Do not put price, sale dates, shipping promises, or your company name inside Q&A. Those belong on offer attributes.
• Do not submit Q&A if document_link already contains the same FAQ. Google says it will extract from the document instead.
• document_link: public PDF URLs, comma-separated if more than one.
• related_product: relationship type, identifier type, and identifier. Google's examples include required_part:id:AZ7B, accessory:gtin:811571013579, and often_bought_with:id:AZ7C.
• popularity_rank: a percentage of your own inventory performance. Higher means stronger relative to *your* other products. Do not invent a marketplace rank.

Grammar and spelling matter. Google says customers rely on the answers. A wrong “yes, it is waterproof” is worse than an empty field.

Shopify path vs WooCommerce path

Shopify. Install and connect Google & YouTube. Let the channel own the primary sync. Export the product id values it uses. Build a supplemental sheet or Merchant API upload for highlights, details, Q&A, documents, and related products. Re-upload when the catalog changes, not once a year.

Metafields can hold the source facts. They do not reach Merchant Center unless you map them. If material and laptop size live in metafields, decide which field is canonical and write that into the supplemental source.

WooCommerce and other platforms. Use the same attribute names. Your plugin or feed app is only a transport. Open the Merchant Center product, not the plugin UI, to verify what Google stored. Custom order meta that never leaves WordPress cannot help AI Mode.

A practical fill order for a 50-SKU test

1. Export 50 SKUs that already approve in Merchant Center.
2. Repair titles, identifiers, and variant groups. Stop if more than a handful still mismatch the site.
3. Write four to six highlights per SKU (Google's recommended range). Read them aloud. If they sound like ads or name another product, cut them.
4. Add details only for measurable specs.
5. Write three to eight Q&A pairs from real tickets or PDP questions: fit, compatibility, inclusions, care, what the product is *not*.
6. Attach PDFs that already exist. Do not upload a marketing brochure and call it a manual.
7. Add related products only when a shopper would be stuck without the second item.
8. Skip popularity_rank until you can calculate it from your own units or revenue share.
9. Submit the supplemental source. Wait for Merchant Center to show the new attributes on a sample of products.
10. Re-run the six-prompt matrix from the Google AI Mode audit.

What these fields cannot do

They cannot override a disapproved item. They cannot invent stock. They cannot make a prohibited product eligible. They cannot guarantee that AI Mode will quote your Q&A or show your SKU.

They can remove a class of failure you can prove: the conversation asked a constraint, and your feed never contained the answer.

If you need the wider catalog and measurement loop, return to the audit. If you need the Shopify-wide channel map, use the Agentic Storefronts hub. This page has one destination: a Merchant Center row that can survive a follow-up question.