- Conversational attributes are six new optional Merchant Center feed fields Google announced at Google Marketing Live 2026 and rolled out globally.
- They help AI surfaces like AI Mode in Search and Gemini understand and recommend your products when a shopper asks a real question in plain language.
- The six fields are question_and_answer, document_link, related_product, item_group_title, variant_option, and popularity_rank.
- They are optional, they do not affect the approval status of your existing products, and Google recommends submitting them through a supplemental data source.
- Google has not documented a Shopping ranking boost for using them. The upside is machine-readable product facts for conversational queries, not a bidding lever.
Shopping has always rewarded one behaviour: giving Google more accurate, more specific information about a product than the merchant next to you did. That principle sits underneath everything we do on e-commerce accounts, and conversational attributes do not change it.
What they change is who is reading. Your feed now has to answer to a system holding a conversation with your customer, not just matching a keyword against a title.
Most merchants will read the announcement, file it under "nice to have," and move on. That is exactly why the fields are worth filling in now.
What are conversational attributes in Google Merchant Center
Conversational attributes are six optional product data fields in Google Merchant Center that help AI systems and conversational agents understand the specific nuances of your products. Google describes them as complementing your primary product data specification rather than replacing any part of it.
They were teased in January 2026 and formally launched at Google Marketing Live in May 2026. They are available in all countries.
The important detail is who reads them. Your title, price, and GTIN are written for a matching system. Conversational attributes are written for a system that has to answer a question it has never seen before, using only what you gave it.
Not sure which of the six actually apply to your catalogue? Read "Which conversational attributes should you prioritise" further down for the order we work in.
Why did Google launch conversational attributes
Google launched conversational attributes because shopping queries inside AI Mode and Gemini are longer, more specific, and more comparative than anything the traditional feed was built to answer.
A shopper no longer types "waterproof jacket." They type something closer to "is this jacket warm enough for a Berlin winter commute, and does it come in a wider cut."
AI Mode and Gemini do not browse your product page the way a human shopper does. They read your feed.
A feed built only for keyword matching gives an AI system almost nothing to reason with when the question is conditional, comparative, or specification-level. It can tell a shopper what the product is called and what it costs. It cannot tell them whether it will work for them.
This is also the second kind of update in a year where Google has quietly changed what advertisers control. The removal of campaign-level language targeting moved a decision into the system. Conversational attributes move in the opposite direction, giving you a lever you actually control.
What are the six conversational attributes
There are six conversational attributes, each closing a different gap in what an AI system can say about your product.
| Attribute | What it does | Format notes |
|---|---|---|
question_and_answer | Product FAQs in question and answer pairs | Question and answer sub-attributes, each up to 1,000 characters. Repeats up to 30 times, capped at 10,000 characters per product |
document_link | Links to authoritative PDF documentation such as manuals, assembly instructions, and package inserts | PDF only, up to 2,000 characters per URL, up to 5 documents per product |
related_product | Declares typed relationships between products in your catalogue | Relationship type, identifier type, and identifier sub-attributes. Repeats up to 30 times |
item_group_title | The family name for a product sold in multiple variants | Used with item_group_id. Example: Organic Cotton Men's T-Shirt |
variant_option | The properties that actually distinguish one variant from another | Name and value sub-attributes, used with item_group_title and item_group_id |
popularity_rank | How well a product sells relative to the rest of your own inventory | A number between 0.0 and 100.0, one decimal place, no percentage sign |
Three of these had no equivalent in the standard feed at all: question_and_answer, related_product, and document_link.
The relationship types in related_product are fixed. You choose from part_of_set, required_part, often_bought_with, substitute, different_brand, and accessory.
That is a structured map of your catalogue, handed to a system deciding what to suggest alongside the product a shopper just asked about.
variant_option fills a real hole. Google's existing spec covers colour, size, material, pattern, age group, and gender. If your product varies along any other axis, such as graphics card configuration or shoe width, you previously had no structured way to say so.
Wondering whether any of this touches your live Shopping campaigns? Read "Do conversational attributes affect Shopping ads performance or product approval" next.
Do conversational attributes affect Shopping ads performance or product approval
No. Adding conversational attributes does not change the approval status of products already in your Merchant Center account, and Google has not published any claim that they improve Shopping ads ranking.
This matters for how you brief a client. The honest framing is that these fields buy you eligibility to be understood on AI surfaces, not a performance lift you can forecast in a spreadsheet.
It is the same trap advertisers fall into with Quality Score. A number moves because the underlying account got better, not the other way round.
Google's stated benefit is that the data helps customers discover product information across AI-driven surfaces while also enhancing traditional search experiences. That second half is deliberately soft language. Treat it that way.
The risk of filling them in is close to zero. The risk of ignoring them is that AI Mode answers a detailed question about your category using a competitor's feed instead of yours.
How do you add conversational attributes to your feed
Google recommends adding conversational attributes through a supplemental data source joined to your primary feed on the id attribute. You can also add them to your primary data source directly, or submit them through the Merchant API.
The supplemental feed route is the right default for agency work. It keeps the new fields separate from the feed your platform generates, so you can test, update, and roll back without touching the source of truth for your live products.
About to spend hours filling these in? Read "What mistakes should you avoid with conversational attributes" before you build the sheet.
A practical sequence that works:
- Export your product IDs and sort by revenue or margin contribution over the last 90 days.
- Take the top slice of that list, not the whole catalogue. These fields are written, not generated, and quality decides whether they get used.
- Build a supplemental Google Sheet or TSV file keyed on id.
- Use TSV rather than CSV. Question and answer values are full of commas and colons, and CSV escaping turns into a debugging exercise you do not need.
- Populate the question and answer pairs first, then catalogue relationships, then document links.
- Upload, confirm the attributes are being read, and expand outward from there.
Pull the questions from somewhere real. Your support inbox, on-site product Q and A, live chat transcripts, and returns reasons are far better sources than anything invented in a briefing document.
Which conversational attributes should you prioritise
Start with question_and_answer, because it has the most surface area and is the one most likely to answer the exact query a shopper types into AI Mode.
A sensible priority order for most catalogues:
- question_and_answer is the highest-value field, and the one that maps directly to how people phrase conversational queries.
- related_product tells AI systems what to suggest alongside or instead of the product, which is where basket value lives.
- document_link is strongest for technical, assembled, regulated, or spec-heavy products. If you already have manuals as PDFs, this is close to free.
- item_group_title and variant_option are essential if you sell anything with variants outside Google's six standard dimensions.
- popularity_rank is cheap to generate from your own sales data and easy to automate, but the least differentiating on its own.
If you sell simple, single-variant, self-explanatory products, the last two do very little for you. Spend the effort on question and answer depth instead.
High-consideration categories get the most out of this. Furniture is the clearest example, where shoppers research dimensions, materials, assembly, and delivery for weeks before buying. Every one of those questions is a pair you could be feeding to AI Mode right now.
One caveat before you invest the hours. If your conversion tracking is broken or incomplete, you will not be able to tell whether any of this changed anything. Fix measurement first, then enrich the feed.
What mistakes should you avoid with conversational attributes
The most common mistake is duplication. Google explicitly tells merchants not to repeat information already submitted in description, product_highlight, or product_detail.
Five things to get right:
- Do not duplicate existing attributes. Repeating your description inside a question and answer pair adds nothing and wastes your character budget.
- Do not submit pairs that are already inside a linked PDF. If you are also using document_link, Google will extract the FAQ information from the document instead.
- Keep offer data out. No prices, no dates, no promotional language. Those belong in the attributes built for them.
- Do not stuff keywords. Google's guidance is explicit about avoiding keyword or search term lists in these fields. This is a comprehension channel, not a ranking one.
- Be accurate with popularity_rank. The value is meant to reflect genuine relative performance within your inventory. Marking every product at 95 makes the signal useless.
There is one more failure mode worth naming. Writing 30 shallow pairs per product from a template is worse than writing 6 specific ones that answer questions a customer actually asked. Volume is not the metric here.
Frequently asked questions
Are conversational attributes required?
No. All six are optional for every product, and leaving them empty has no effect on your existing listings or their approval status.
Will conversational attributes get my products into AI Mode?
They make your products easier for AI surfaces to understand and recommend, but Google has not guaranteed placement or documented a ranking benefit. Think of it as eligibility, not a lever.
Can I use conversational attributes outside the US?
Yes. All six attributes can be used in all countries.
Do I need a developer to implement them?
Not necessarily. A supplemental data source built as a Google Sheet or TSV file joined on id covers most merchants. The Merchant API is the better route only if you are populating these fields at scale or automating popularity_rank from live sales data.
How many questions and answers can I submit per product?
Up to 30 pairs, with each question and each answer capped at 1,000 characters, and a total limit of 10,000 characters across all pairs for that product.
Should I write these myself or generate them?
Generate the first draft if you like, but ground every question in a real customer question from support tickets, on-site reviews, or your inbox. Invented questions produce generic answers, and generic answers are exactly what an AI system already has.
Get your feed ready before your competitors do
Right now most catalogues in your category have six empty fields where their product knowledge should be. The merchants who fill them in first are the ones AI Mode has something to work with.
At Fuel Digital we build and manage Shopping feeds for e-commerce brands across the US, Australia, and Canada. You can see what that has produced for our e-commerce clients, or start with a full Google Ads account audit to find out what your feed is currently telling Google before you add anything to it.
