Conversational attributes in your product feed: prepare for AI-driven shopping queries in a few clicks

Shopper queries are getting longer and more specific. People asking Gemini or AI Mode about a product don't type "running shoes" anymore. They type "running shoes for flat feet under $150 that don't squeak on wet pavement." That kind of query needs richer product data than a traditional title and description can provide.

Google rolled out six new conversational attributes in the Merchant Center product data specification, designed to help AI systems and conversational agents better understand the specifics of your products. They're optional. Adding them won't affect your existing product approval status. But they're a clear signal of where Google sees shopping search going, and they give brands a structured way to add the kind of detail that AI surfaces can pull from.

Here's what each attribute does, why the FAQ attribute might be the most interesting of the bunch, how to think about popularity rank, and how to add this data to your feed without doing it all by hand.

Why AI shopping queries are a problem Google needs help solving

Before going into the attributes themselves, it's worth understanding why Google is investing in this at all. AI shopping queries are extremely long tail. They're often unique. They've probably never been searched before in that exact form, and they might never be searched in that exact form again.

Compare two queries: "running shoes" and "running shoes for flat feet under $150 that don't squeak on wet pavement." For the first one, Google has years of behavioral data. Millions of people have typed it, clicked on results, bounced back, refined, and bought. Google can rank confidently based on what historically works for that query.

For the second query, none of that exists. There's no historic click-through data. No bounce rate signals. No "people also searched for" pattern. Google might be seeing this exact combination of constraints for the first time, and it still needs to give the shopper a useful answer.

This shifts the burden. When Google can't rely on historic SERP data, it leans harder on the underlying product data itself. The more structured, specific, and accurate the data you submit, the easier it is for Google to match your products to queries it has never seen before. Which is, in part, why Google is offering these new conversational attributes. They make Google's job easier, and brands that adopt them give Google better material to work with.

Before going further: the usual cat-and-mouse caveat

SEO has always been a cat-and-mouse game between marketeers and Google. A new signal gets introduced, smart marketeers figure out how to use it, then some less smart ones figure out how to abuse it, and Google eventually adjusts how much weight that signal carries. The cycle keeps going.

It's not yet clear how much weight conversational attributes will carry in practice. Google has shipped the specification, but how AI surfaces actually use the data, and how much it influences what gets surfaced, is something we can only speculate about right now. Worth submitting these attributes honestly, but worth keeping expectations grounded too. These types of features come and go in the SEO world.

That said, this is going to be interesting to test. More on that below.

What conversational attributes are

Conversational attributes sit alongside your existing Merchant Center product data. According to Google's documentation, they're designed to complement the data you already submit, not replace it. If you already include information in your description, product_highlight, or product_detail fields, you don't need to duplicate it.

You can submit them through a supplemental data source (Google's recommendation), through your primary data source, or via the Merchant API. The full list and specification is available here: how to use conversational attributes.

There are six attributes in total. Let's walk through them.

Question and answer [question_and_answer]

This is the one most worth paying attention to. The question_and_answer attribute lets you submit FAQ-style content as structured product data. Each entry is a question paired with an answer, both written in plain language.

A good example from Google's documentation: for a phone, you might submit "Does it have a headphone jack?":"This version doesn't have a headphone jack." and "Does it support Bluetooth?":"It has full Bluetooth 6.0 support."

Why this matters: conversational AI agents need to answer specific questions. When a shopper asks Gemini "does the iPhone 17 still have a headphone jack," the AI needs structured product data it can pull from. A general product description that mentions "premium audio features" doesn't help much. A direct Q&A pair does.

This is where the FAQ attribute becomes genuinely useful for narrowing down on specific user queries. Conversational shopping queries often contain very specific constraints: "is this waterproof," "does this fit a 15 inch laptop," "is the battery replaceable," "does this come with the charging cable." General product descriptions rarely answer all of these directly. A well-built Q&A library does.

If you can mine your support tickets, product page FAQs, customer review questions, and pre-sale chat logs, you can build a Q&A library that maps directly to the questions shoppers ask AI assistants. The brands that already have rich FAQ content on their product pages are in a strong position. The brands that don't have a clear opportunity to build it.

A few practical tips for FAQ content:

• Focus on questions that are specific to the product, not generic shop policy questions
• Keep answers short, factual, and to the point
• Cover the things shoppers usually need to confirm before buying: compatibility, materials, sizing, warranty specifics, included accessories
• Update them when product specs change

Building that library by hand for a large catalog is real work, though. This is exactly what the new conversational attributes generation type in the Productrise supplemental feeds feature takes care of: we read your product pages, your existing data, and your product feed, then our carefully tuned AI model generates ready-to-go question_and_answer content for all your products in just a few clicks.

A fun experiment worth running

Here's an idea worth testing if you have a catalog and some time to play with this. Pick a product, and submit a question_and_answer pair that's so utterly unique that no other product on the internet could plausibly match it.

For example, if you ran a store that sold flip flops, you could add: "Are these flip flops able to let you walk on molten lava after a volcano erupts?":"Yes! These flip flops are the only ones that actually do this!"

Then, after some time has passed, type that exact query into AI Mode and see whether your flip flops get surfaced. If they do, you've confirmed that question_and_answer pairs are being used as a direct retrieval signal for AI surfaces. If they don't, the attribute might still influence things, just not in a way you can isolate with a single weird question.

Google is usually well ahead of abuse like this, so anyone trying to game the system at scale will probably get filtered out fast. But as a one-off experiment to learn how the signal actually behaves, it's a fun way to probe what's happening behind the scenes. If you run it, let us know what you find.

Document link [document_link]

This attribute lets you link to PDFs related to a product, such as a user manual or assembly instructions. You can submit multiple PDF URLs by separating them with commas.

It's most useful for products where documentation matters: electronics, furniture, tools, appliances. Less relevant for clothing or simple consumer goods. AI agents can reference these documents when answering detailed product questions, which can be helpful when a shopper asks something the product description doesn't cover.

The new conversational attributes feature in Productrise supplemental feeds covers document_link too, alongside question_and_answer, so you can add both attributes to your entire catalog in one go.

Related product [related_product]

Related product lets you signal relationships between products in your catalog. You can specify a relationship type (like accessory or required_part), an identifier type (id or gtin), and the identifier itself.

For example, if you sell a camera, you might submit accessory:gtin:811571013579 to link a compatible lens. This helps AI agents suggest complementary items in a natural way during a shopping conversation.

Item group title [item_group_title]

If you sell products with variants (different sizes, colors, materials), the item_group_title attribute gives the whole group a clean, human-readable name. So instead of every variant carrying its own long title, the group as a whole gets called "Organic Cotton Men's T-Shirt," and individual variants inherit that grouping.

This helps AI surfaces describe variant families in a way that makes sense to shoppers, rather than listing each variant as a separate product.

Variant option [variant_option]

This pairs with item_group_title. The variant_option attribute lets you specify the properties that distinguish one variant from another. Each option has a name and a value, like Shoe width:narrow and size:8.

Together with item_group_id and item_group_title, this gives AI surfaces a clear picture of what makes each variant different, which matters when a shopper asks something like "do you have this in a wide width."

Popularity rank [popularity_rank]

Popularity rank is a self-reported value between 0 and 100, where higher values indicate better-performing products in your catalog. Google's example uses 95.5 for a top performer. Brands decide what counts as popular: units sold, revenue, page views, or conversion rate.

Worth keeping expectations realistic here. This attribute is most likely used (if it gets used at all) to rank your own products against each other, not to outrank a competitor's catalog. Marking every product at 100 to try to beat someone else's listings isn't going to work. The point of the attribute is to tell Google which of your products are the strongest, so AI surfaces can prioritize accordingly. A realistic distribution (some top performers at 90+, most products in the middle, some at the bottom) is the right approach.

The same fundamentals still apply

Worth emphasizing: conversational attributes don't replace the core principles of organic product visibility. Your titles, descriptions, images, structured data on product pages, accurate pricing, and product reviews all still matter. Conversational attributes add a layer on top, giving AI surfaces extra context that makes products easier to surface for specific queries.

If your feed has unresolved issues with required attributes, missing images, or mismatched structured data, those are still the first things to fix. Conversational attributes are an addition, not a replacement. For a deeper look at the basics, see how to audit your product feed for organic visibility.

Add conversational attributes in a few clicks with Productrise

The recommended way to add these attributes is through a supplemental feed, which keeps your primary feed clean and makes it easier to update conversational data without touching your main product data. Read more about how supplemental feeds work here: supplemental feeds: fix product feed issues for organic visibility.

And this is where Productrise makes things easy. Our supplemental feeds feature now includes a dedicated generation type for conversational attributes. In just a few clicks, you can add question_and_answer and document_link data to all your products. We read your product pages, your existing data, and your product feed, and our carefully tuned AI model turns that into structured, ready-to-go conversational data. This sits alongside all the other supplemental feed generation types we support, like titles, descriptions, product categories, and more.

Want your products ready for AI-driven shopping queries? Start your free trial today and generate conversational attributes for your entire catalog in minutes.