Google added six new optional attributes to help AI systems understand your products. What they do, a worked example, and how to start filling them in.
Shopping has only ever rewarded one behavior: giving Google more accurate, specific information about a product than the next merchant did. Fill in a field well, and Google's systems can match that product to the shopper looking for it. Fill it in vaguely, or duplicate what you've already said elsewhere, and it does nothing for you.
Google Merchant Center's new conversational attributes don't change that rule. They extend it to a new kind of reader: an AI system having a conversation, not just an algorithm matching keywords.
What you need to know right away
Google added six new optional attributes: question_and_answer, document_link, related_product, item_group_title, variant_option, and popularity_rank.
They complement your existing feed, not replace it. Adding them doesn't affect the approval status of products already live.
The recommended path is a supplemental data source, built and maintained through a feed tool like Channable rather than by hand.
The bar isn't "fill in every field." It's fill in every field you can back with something real and specific — the same standard that's always separated good feeds from bad ones.
Google hasn't listed these as a Shopping or Performance Max ranking factor. It doesn't need to be one for this to matter: your feed is what your campaigns run on.
What conversational attributes actually are
Google's own framing is that these attributes help AI systems and conversational agents better understand your products' specific nuances. In practice, that means giving Google structured answers to the questions a shopper would otherwise have to ask a salesperson, or dig through five browser tabs to find.
The Six Attributes Explained
Question and answer (question_and_answer) FAQ pairs about the product. Example: "Does it have a headphone jack?": "This version doesn't have a headphone jack."
Document link (document_link) Links to related PDFs: manuals, spec sheets, assembly instructions. Multiple files are comma-separated.
Related product (related_product) Explicit relationships to other products in your catalog: required parts, accessories, frequently bought together. Built from a relationship type, an identifier type, and an identifier.
Item group title (item_group_title) A shared title across every variant of a product, paired with item_group_id.
Variant option (variant_option) The specific attributes distinguishing one variant from another within that group, such as size or color.
Popularity rank (popularity_rank) Where a product ranks by popularity as a percentage of your total inventory. Higher means better-performing relative to the rest of your catalog.
None of these are new concepts for a merchandiser. FAQs, related products, and variant data already exist somewhere in most catalogs, usually scattered across the product page, a PDF, and a CMS field nobody exports. What's new is that Google now has a dedicated place in the feed for it, specifically so an AI system doesn't have to guess.
Why conversational attributes matter
This lands at the same moment shopping is becoming conversational across the entire category, not just inside Google. It's a shift the team at Flowboost has been tracking closely across both organic and paid channels, since it touches how products get discovered no matter which surface a shopper starts on. Google's own AI Mode and Gemini are moving Search in the same direction. Conversational attributes are Google giving merchants a formal channel to feed that shift.
AI Mode and Gemini don't browse your site the way a shopper does. They read your feed. A feed built only for keyword matching — title, price, a generic description — gives an AI system almost nothing to reason with when a shopper asks a comparative or conditional question. Conversational attributes close that gap, but only if you actually fill them in.
What this looks like on an actual product
Take a mid-range waterproof shell jacket, sold in four sizes and two colors.
The base feed already covers the fundamentals: ID, title, price, GTIN, brand, availability. Layer in item_group_title ("Classic waterproof shell jacket") and variant_option ("size:M,color:forest green") and each individual SKU becomes legible as one option inside a family, not an isolated product.
Add two or three question_and_answer pairs built from the questions people actually ask before buying a jacket — is it machine washable, does it have underarm vents, how does the fit run — and an AI agent can answer those directly instead of guessing or defaulting to a competitor's product page.
Add a document_link to the care and sizing guide, and a related_product entry pointing at the packable stuff sack customers usually buy alongside it, and the product now carries the context a good in-store salesperson would give in person.
None of this changes what the jacket costs or whether it's in stock. It changes whether an AI system can talk about it accurately, and whether it recommends this jacket over one with a thinner feed.
The part most feeds aren't ready for
Google's documentation is careful to note these attributes are optional and won't affect approval. That's true, and it undersells the real barrier: writing good conversational data at catalog scale is a harder problem than writing good keyword-matching data.
A title field tolerates some sloppiness, because algorithmic matching is fuzzy by design. question_and_answer doesn't work that way. A vague, wrong, or copy-pasted-across-products answer doesn't just fail to help; it actively erodes trust in your feed as a source once an AI system catches the pattern. That's a worse outcome than not submitting the attribute at all.
This is a data-quality problem, not a field-mapping problem, and most catalogs aren't set up to solve it yet.
How to start without rebuilding your feed
Build it as a supplemental feed, in a proper feed tool. A platform like Channable lets you layer conversational attributes onto your existing primary feed through rules and mapping, so you're not hand-editing exports or waiting on a developer to touch your core feed. This is the practical version of Google's own "supplemental data source" recommendation.
Check for duplication before you write anything new. If you already populate description, product_highlight, or product_detail with the same information, don't recreate it under a conversational attribute. Duplicated, inconsistent data across fields is worse for an AI system than a gap.
Start with your highest-traffic, highest-variant products. item_group_title and variant_option pay off fastest on products with real variant complexity — apparel with size and color, electronics with configuration options. A single-SKU product gets little from this pair.
Write question_and_answer from actual customer questions, not from what you assume they'd ask. Pull from support tickets, product reviews, and on-site search queries. Five accurate, specific pairs beat twenty generic ones.
Treat popularity_rank as a maintenance commitment, not a one-time export. It's a relative ranking across your inventory. Set it once and never update it, and it becomes actively misleading as your catalog and sales mix shift.
Watch your Shopping and Performance Max data after rolling these out, not just your feed's approval status. Since the effect on campaigns is indirect, you'll want to track it in your actual account data rather than assume it's working. A dedicated Google Ads analysis tool like Wolfy can help surface whether feed-quality changes like these are actually showing up in performance, instead of guessing from the dashboard alone.
The takeaway
Fill in every field you can back with something real and specific. That's the entire discipline, and it hasn't changed just because the reader on the other end is now a conversational AI instead of a matching algorithm.
The merchants who treat these six attributes as a checkbox, filled in vaguely just to have them filled in, will see nothing from this. The ones who treat them the way they've always treated good feed data (as an extension of genuinely useful product content) will be the ones AI Mode, Gemini, and whatever comes next actually recommend when the question gets specific.
