Google's six conversational attributes, filled in for a CPG product feed

In May, at Google Marketing Live, Google announced a set of new Merchant Center fields it calls conversational attributes. They got less attention than the AI shopping announcements around them, and the official examples (a phone with no headphone jack, a cotton t-shirt) don't help a food or personal care brand see what to do with them. On October 22, Search Engine Land published a clear walkthrough of all six, and it's a good moment to translate them for CPG.

The short version: two of the six are worth real effort for most CPG brands, two are housekeeping, and two depend on your catalog. None of them is required, and Google says adding them won't affect the approval status of your existing products.

What Google says these are for

Google's help page describes conversational attributes as a way to help "AI systems and conversational agents better understand your products' specific nuances," across surfaces like AI Mode in Search. When it announced them, Google said its AI systems use the data to match products with conversational shopping queries across AI Mode, Gemini and other AI-powered surfaces, according to Search Engine Land's May coverage.

Google hasn't published how much weight these fields carry, and we won't guess. What's clear is the direction. Shoppers ask assistants detailed questions, and the assistant can only answer from data it has. A feed with a title, a price and a 300-word description doesn't say whether your granola is made in a facility that also handles peanuts. A question_and_answer field can.

There's a reason to think this kind of question-and-answer data is useful beyond Google, too. In SISTRIX's study of nearly 100,000 URLs published this week, pages with FAQ markup were cited 1.5 times as often in AI answers as pages on the same site without it. That's a different system (web pages, not feeds), but it points the same way: AI answers draw on content already shaped as questions and answers.

The six attributes and their limits

Attribute	What it holds	Limits
question_and_answer	Customer questions and answers about the product	Up to 30 per product; question and answer up to 1,000 characters each
document_link	Links to PDF documents about the product	Up to 5 PDFs; each URL up to 2,000 characters
related_product	Other products and how they relate	Up to 30; types include part of a set, required part, often bought with, substitute, different brand, accessory; identified by GTIN or ID
item_group_title	A name for the whole product family	1 to 150 characters, no variant details; used with item_group_id
variant_option	Name and value pairs that tell variants apart	Used with item_group_id and item_group_title; keep names consistent
popularity_rank	How well the item sells relative to your catalog	0 to 100, higher is more popular, no percent sign

Google's one important caveat: if you already put a detail in description, product_highlight or product_detail, you don't need to repeat it here. These fields are for what's missing, not a second copy.

question_and_answer: where most of the value is

This is the field to spend time on. Thirty questions per product is a lot of room, and most CPG products have more real customer questions than their product page answers.

Good sources for questions, in order of usefulness:

1. Your customer service inbox and chat logs, sorted by how often a question comes up.
2. Retailer Q&A sections on Amazon, Target, Walmart and others, where shoppers ask things they didn't find on the listing.
3. Reviews that mention a surprise ("I didn't realize it had erythritol").
4. The questions in your website FAQ, if it's good. Our FAQ tip covers how to write those.

For a CPG product, the questions shoppers bring to assistants cluster around a few themes: allergens and cross-contact, diets (keto, vegan, gluten-free, low-FODMAP), ingredients they want to avoid, who it suits (kids, pregnancy, sensitive skin), how to use or store it, how long it lasts once opened, and how it compares to the alternative they already buy.

Question	Answer
Is this granola safe for someone with a peanut allergy?	It contains no peanuts, but it's made on shared equipment that also processes peanuts and tree nuts, so we don't recommend it for anyone with a peanut allergy.
How much sugar is in a serving?	A 1/2 cup (45 g) serving has 6 g of total sugar, all from maple syrup and dried fruit, and 4 g of added sugar.
Does it stay crunchy in milk?	It stays crunchy for about five minutes in cold milk. For longer, eat it over yogurt.
How long does it keep after opening?	About four weeks in the resealed bag at room temperature, out of direct sun.

Three rules for writing these. First, every answer must match the packaging and the product page word for word on any fact. If the label says "may contain tree nuts" and the Q&A says "nut-free," you've created the kind of contradiction that makes an assistant distrust both, and a real allergy risk. Our consistency tip applies here more than anywhere. Second, answer the question in the first sentence, then add detail. Third, a Q&A is still a product claim. "Supports digestion" in a Q&A field needs the same substantiation and review as it would on the box.

related_product: tell Google what goes together

The related_product attribute lets you connect items in a typed way, with up to 30 per product. For CPG this is more useful than it looks, because so many shopping questions are about combinations and swaps.

• Substitute: your oat milk barista edition as a substitute for your original oat milk, so an assistant can suggest it when someone asks for one that froths.
• Often bought with: the conditioner for the shampoo, the refill pouch for the starter bottle.
• Part of a set: each flavor in a variety pack linked to the pack.
• Different brand: if you make a sub-brand or a private label line, this is the honest way to mark the relationship.

Use GTINs as identifiers wherever you have them. Tip 16 covers why GTINs matter in Merchant Center generally.

document_link: the PDFs you already have

Up to five PDFs per product. Google's examples are manuals and assembly instructions, which fits electronics better than snacks, but CPG brands usually have documents shoppers would value: a full nutrition and ingredient sheet, an allergen statement, a certificate from a third-party certifier, a usage guide for a skincare routine, or a safety data sheet for cleaning products.

Two cautions. Only link documents you'd be comfortable with a shopper reading in full, because an assistant may quote them. And keep them current: an outdated spec sheet with an old formula is worse than none. If you hold certifications, tip 21 covers showing them so machines can read them.

The housekeeping three

item_group_title and variant_option matter if you sell the same product in several sizes, flavors or scents. The group title names the family ("Cold Brew Coffee Concentrate") without any variant details, and variant_option spells out what differs (size: 32 oz, flavor: vanilla). Keep the option names identical across the catalog, so "flavor" is never "Flavour" or "taste" on another SKU.

popularity_rank is a 0 to 100 number showing how each item sells compared with the rest of your catalog. Google's walkthrough recommends updating it when your best sellers change. It's quick to generate from sales data, and an honest number is the only kind worth sending.

How to roll this out without breaking your feed

Google recommends a supplemental data source rather than editing your primary feed, and we agree. A supplemental source keeps the new fields separate, so a mistake in a Q&A can't touch your prices or availability. Google's help page links a TSV template.

A sensible order for a brand with a few dozen SKUs:

1. Pick your ten best sellers.
2. Write eight to twelve questions and answers for each, drawn from real customer questions, and run them past whoever approves label claims.
3. Add related_product links between those ten and their natural pairs and substitutes.
4. Add document links where you have a current PDF.
5. Upload as a supplemental source, check for errors, then wait a few weeks before judging.

Don't fill all 30 Q&A slots on day one. Ten good answers beat thirty padded ones.

Measuring whether it did anything

Google's AI performance insights report in Merchant Center shows eligible merchants how products and brands perform in conversational queries across AI Mode and AI Overviews, including share of voice against similar competitors, according to Search Engine Land. In May, Google said the report would roll out in Australia, Canada, India, New Zealand and the US in the coming months, so check whether it's in your account yet.

That report only covers Google. Shoppers use several assistants, and Salesforce's 2026 holiday research shows how spread out that use is.

Show the data

Assistant type	Shoppers
Third-party AI chat assistant	21%
AI assistant on a social platform	17%
Brand-owned AI assistant	12%

Source: Salesforce, 2026 Holiday Retail Predictions (July 20, 2026)

For the assistants outside Google, you need another check, whether that's asking them yourself on a schedule or using a tool like SearchPanel from XENA Intelligence, which asks assistants shopper-style questions and keeps the answers. Either way, the test is the same: ask the specific questions your new Q&As answer, and see whether the answers start reflecting your data.

Where this fits

Conversational attributes are Google's version of a trend we've covered before. OpenAI's product feed asks for similar detail, and our ChatGPT shopping feed piece covers what it rewards. The underlying job is the same in every system: get the facts a shopper would ask about out of your team's heads and off the back of the pack, into a structured form a machine can read. The grocery site data we published earlier suggests most brands haven't done this on their own websites either. The feed is a good place to start, because it's a spreadsheet, and spreadsheets get finished.