Skip to content

Library Guides

Conversational commerce: the product data AI assistants ask for

What AI shopping assistants need from product data: Google's six conversational attributes, a worked example, a field map and the status of each surface.

By CentralUpdated 20 November 202610 min read20 sources, checked 20 November 2026
On this page
  1. What changes when a shopper asks an assistant
  2. Google's six conversational attributes
  3. A worked example: Northpeak Trail Runner 2
  4. The same facts in three formats
  5. Where the AI shopping surfaces stand
  6. How to write answers an assistant can quote
  7. What not to put into question_and_answer
  8. Where Central helps

In short

  • An AI shopping assistant answers a shopper's question from product facts, so it needs each fact as its own labeled value and not as keywords in a title.
  • Google documents six conversational attributes: question_and_answer, document_link, related_product, item_group_title, variant_option and popularity_rank.
  • As of November 2026, all six are optional, can be used in all countries and do not affect product approval; Google recommends sending them in a supplemental data source.
  • question_and_answer takes up to 30 pairs per product, with up to 1,000 characters per question and per answer and 10,000 characters in total.
  • OpenAI's product feed has no field for questions and answers, related products or documents; schema.org Product has properties for related products but none for questions or documents.
  • Google's checkout in AI Mode and Gemini is available to select merchants, for products eligible in the United States, Canada and Australia.
Product FAQ writerFree tool5 free runs a day

What changes when a shopper asks an assistant#

A keyword search matches the words in your title. An assistant gets a question, such as "Which trail running shoe has a rock plate and comes in a wide fit?", and has to answer it from facts. Google's Business Agent page gives examples of what shoppers ask: "What is this rug made of?" and "Do these shoes run true to size?". Your product can only be in the answer if each of those facts exists as its own value, in a field the assistant reads, and says the same as your product page.

This page covers that data. How assistants find and sell products at all is in AI shopping for merchants, and the term itself in agentic commerce.

Google's six conversational attributes#

Google's page How to use conversational attributes lists six attributes that "help AI systems and conversational agents better understand your products". Their status as of November 2026:

  • Optional. Google calls them "completely optional", and including them does not change the approval status of your products.
  • All countries. Each attribute page says the attribute can be used in all countries.
  • How to send them. In a supplemental data source (Google's recommendation), in the primary data source or through the Merchant API. Google offers a TSV template.
AttributeWhat it holdsLimitsFormat example
question_and_answerQuestions and answers about the product, written by users, the merchant or the manufacturerUp to 30 pairs; question and answer up to 1,000 characters each; 10,000 characters in total"Does it support Bluetooth?":"It has full Bluetooth 6.0 support."
document_linkPDF documents such as user guides, assembly instructions and package insertsUp to 5 URLs of up to 2,000 characters; PDF only; 50 MB per file; public and crawlablehttps://example.com/manual.pdf
related_productOther products in your data and how they relate to this oneUp to 30; three parts: relationship type, identifier type, identifieraccessory:gtin:811571013579
item_group_titleOne common name for all variants of a product1 to 150 charactersOrganic Cotton Men's T-Shirt
variant_optionThe options that define this variantUp to 30 name and value pairs of up to 250 characters each; 5,000 characters in totalShoe width:narrow,size:8
popularity_rankHow well the product sells within your own inventoryA number from 0 to 100 with at most one decimal and no percent sign95.5

related_product knows six relationship types: part_of_set, required_part, often_bought_with, substitute, different_brand and accessory. The identifier type is gtin or id, the product ID in your own data source. Send one value per related product, and use only letters, numbers, underscores and dashes in the identifier.

popularity_rank is your own assessment, based on recent sales against products in the same general category. It is not a customer rating.

All six sit in the "Detailed product description" group of the specification. The full attribute list is in our Merchant Center product data specification reference.

A worked example: Northpeak Trail Runner 2#

The facts on record for the fictional Northpeak Trail Runner 2, a men's trail running shoe:

  • 6 mm drop: 30 mm at the heel, 24 mm at the forefoot
  • 4 mm lugs and a rock plate under the forefoot
  • 280 g (9.9 oz) per shoe in men's US 9
  • Recycled polyester mesh upper, not waterproof; a waterproof version exists (Trail Runner 2 WP)
  • Regular and wide widths; fits true to size; removable insole
  • No animal materials

Written as feed values for one variant, one attribute per line:

id: NP-TR2-GRN-9-R
item_group_id: NP-TR2
item_group_title: Northpeak Trail Runner 2 Men's Trail Running Shoe
title: Northpeak Trail Runner 2 Men's Trail Running Shoe, Forest Green, US 9
variant_option: color:Forest Green,size:US 9,width:regular

question_and_answer: "What is the heel-to-toe drop?":"The drop is 6 mm, with a 30 mm stack at the heel and 24 mm at the forefoot."
question_and_answer: "Is the Trail Runner 2 waterproof?":"No. The mesh upper is not waterproof and drains quickly. The Trail Runner 2 WP is the waterproof version."
question_and_answer: "Does it fit true to size?":"Yes. It fits true to size and comes in regular and wide widths."
question_and_answer: "Does it have a rock plate?":"Yes. A rock plate under the forefoot protects the foot from sharp stones."
question_and_answer: "How deep are the lugs?":"The outsole lugs are 4 mm deep."
question_and_answer: "Can I remove the insole?":"Yes. The insole is removable and can be replaced with an orthotic insole."
question_and_answer: "How much does one shoe weigh?":"One shoe weighs 280 g (9.9 oz) in men's US 9."
question_and_answer: "Does the shoe contain animal materials?":"No. The upper, the lining and the glue contain no animal materials."

related_product: substitute:id:NP-TR2WP-GRN-9-R
related_product: accessory:id:NP-GAITER-LOW
related_product: often_bought_with:id:NP-SOCK-TRAIL-M

document_link: https://example.com/docs/trail-runner-2-care-guide.pdf
popularity_rank: 88.5

How to read it:

  • In a file, each attribute name is a column. A repeated attribute becomes several columns with the same name, or one cell with the values separated by commas. For question_and_answer, Google recommends tab-separated files over CSV because of the commas.
  • Quoting. In question_and_answer, a value that contains a comma or a colon goes in double quotes, and a quote inside the value is doubled.
  • Variants. Every variant in the group carries the same item_group_id and item_group_title and the same three option names (color, size, width), with a value combination no other variant has. The group title leaves the variant details out; the title carries them.
  • No duplicates. The care instructions are in the PDF, so no question repeats them. Google extracts FAQ information from documents sent with document_link and asks you not to send the same information twice.

The same facts in three formats#

Google is not the only reader. The table maps each kind of fact to Google Merchant Center, OpenAI's product feed for ChatGPT and the schema.org Product markup on your own product page, as of November 2026.

FactGoogle Merchant CenterOpenAI product feedschema.org Product
Titletitle or structured_title, 150 characterstitle, aim for 150 or fewername
Descriptiondescription or structured_description, 5,000 charactersdescription, aim for 5,000 or fewerdescription
Priceprice, sale_priceprice, sale_priceoffers.price with offers.priceCurrency
Availabilityavailability: in_stock, out_of_stock, preorder, backorderavailability: the same, written pre_order, plus unknownoffers.availability, such as InStock, OutOfStock, PreOrder, BackOrder
Identifiersgtin, mpn, brand, identifier_existsgtin, mpn, brandgtin (or gtin8, gtin12, gtin13, gtin14), mpn, sku, brand
Imagesimage_link, additional_image_linkimage_url, additional_image_urlsimage
Variantsitem_group_id, item_group_title, variant_optiongroup_id, listing_has_variations, variant_dictProductGroup with productGroupID, variesBy and hasVariant
Questions and answersquestion_and_answerno fieldno field on Product; FAQPage is a separate page type
Related productsrelated_productno fieldisRelatedTo, isSimilarTo, isAccessoryOrSparePartFor, isConsumableFor
Documentsdocument_linkno fieldno dedicated field; subjectOf can point to a document about the product

OpenAI's schema says it plainly: "Raw review entries and question-and-answer lists are not part of this discovery contract." For ChatGPT, the facts behind your answers therefore have to sit in the description and attribute fields of the feed and on the public product page. The feed itself is covered in how to get your products into ChatGPT Shopping, and the three feeds side by side in Google vs Microsoft vs ChatGPT product feeds.

Google reads the schema.org column from your pages: its variant documentation uses ProductGroup, and its merchant listing documentation requires name, image and offers on a Product. To see what an assistant's crawler can read on one of your product pages today, run this check:

AI shopping readiness checkerFree tool10 free checks a day

Where the AI shopping surfaces stand#

What each official page says, read on November 20, 2026:

SurfaceStatus according to the official page
AI Mode in Google SearchWorks from Merchant Center product data. Four of the six attribute pages say the attribute is "primarily intended for use in conversational experiences" such as AI Mode. The AI performance insights report covers AI Mode and AI Overviews for English-language queries in Australia, Canada, India, New Zealand and the United States
GeminiFree listings "may show" in Gemini, next to Search, Images, Lens, YouTube and the Shopping tab
Business AgentA chat with your brand on Google Search that answers from Merchant Center data and your website. The rollout begins in the US. Customizing it needs a US-based store, at least 50 approved free listings and a claimed brand profile. Agentic checkout is "coming soon"
Checkout through the Universal Commerce ProtocolA Buy button in AI Mode and Gemini, available for select merchants and for products eligible in the United States, Canada and Australia; you apply with an interest form. Only listings with native_commerce(checkout_eligibility) show the button
ChatGPTA product feed uploaded through OpenAI's Business Hub after an identity check. The standard upload targets the US. The feed carries no questions and answers, and checkout needs a separately enabled integration

Google's pages are not fully aligned on the checkout rollout. The help page above covers products eligible in the US, Canada and Australia. The integration hub page says onboarding in Merchant Center is rolling out in the US, "with Australia and Canada to follow next year". Google's blog post of May 20, 2026 announced Canada and Australia "in the coming months, and later to the U.K.", without a date. The protocol is explained in Universal Commerce Protocol.

How to write answers an assistant can quote#

  1. Put the fact first. Start with "Yes", "No" or the number, then add the detail.
  2. One fact per answer. A question about weight gets the weight. A second fact gets its own question.
  3. Use numbers with units. "The lugs are 4 mm deep" can be quoted and compared. "Aggressive grip" cannot.
  4. Name the variant when the fact depends on it. "280 g in men's US 9" is true; "280 g" is not true for every size.
  5. Leave promotion out. Google asks for "correct and informative answers". An assistant has no use for "you will love it".
  6. Answer only from the record. If no source supports a fact, leave the question out. A wrong answer is repeated to every shopper who asks.
  7. Cover different topics. Google suggests specifications, ingredients, package contents, and facts about the occasions, sports or activities the product is made for.

The same question, weak and strong:

Answer
Weak"Our Trail Runner 2 keeps you going in any weather, and it is on sale this week."
Strong"No. The mesh upper is not waterproof and drains quickly. The Trail Runner 2 WP is the waterproof version."

The weak answer avoids the question, makes a claim nobody can check and adds an offer, which Google does not allow in this attribute.

What not to put into question_and_answer#

Google's attribute page and the specification rule out:

  • Prices, dates and other time-related information. Offer data belongs in the price and availability attributes.
  • Shipping, delivery dates and your company's name.
  • Keyword lists or search terms, promotional text, all capital letters and gimmicky characters.
  • Information you already send in title, description, product_detail or product_highlight.
  • Information that is in a PDF you send with document_link.
  • A question without an answer. Both sub-attributes are required.

Three more checks before you send the file: every answer matches the product page, no answer is true for only some variants in the group unless it says so, and the 10,000 characters per product are not exceeded.

Where Central helps#

Central researches a product's facts in your own data, on the open web and in verified databases, and backs every value with sources you can check. Questions, answers, titles and attributes are written from that verified record within each channel's limits, and what the sources do not support is held back. When a fact behind an answer changes, Central flags the text.

See it on your own products

Questions

What are Google's conversational attributes?

Six optional Merchant Center attributes: question_and_answer, document_link, related_product, item_group_title, variant_option and popularity_rank. Google says they help AI systems and conversational agents understand a product's details, on surfaces such as AI Mode in Google Search.

Do conversational attributes affect product approval?

No. Google states that including them does not change the approval status of your existing products. As of November 2026 they are optional and can be used in all countries.

How do I send conversational attributes without changing my main feed?

Google recommends a supplemental data source: a second file or sheet that carries the product id and the new attributes. You can also add them to the primary data source or send them through the Merchant API.

Does ChatGPT's product feed accept questions and answers?

No. OpenAI's product feed schema says raw review entries and question-and-answer lists are not part of its discovery contract. Product facts reach ChatGPT through the feed's own fields and through your public product pages.

How many questions and answers can one product have on Google?

Up to 30 pairs. Each question and each answer can have up to 1,000 characters, and all pairs of one product together up to 10,000 characters.

Sources

  1. How to use conversational attributesGoogle Merchant Center Helpchecked 20 November 2026
  2. Question and answer [question_and_answer]Google Merchant Center Helpchecked 20 November 2026
  3. Document link [document_link]Google Merchant Center Helpchecked 20 November 2026
  4. Related product [related_product]Google Merchant Center Helpchecked 20 November 2026
  5. Item group title [item_group_title]Google Merchant Center Helpchecked 20 November 2026
  6. Variant option [variant_option]Google Merchant Center Helpchecked 20 November 2026
  7. Popularity rank [popularity_rank]Google Merchant Center Helpchecked 20 November 2026
  8. Product data specificationGoogle Merchant Center Helpchecked 20 November 2026
  9. Free listings for productsGoogle Merchant Center Helpchecked 20 November 2026
  10. About AI performance insightsGoogle Merchant Center Helpchecked 20 November 2026
  11. Get started with Business AgentGoogle Merchant Center Helpchecked 20 November 2026
  12. About the Universal Commerce Protocol (UCP) and UCP-powered checkout feature on GoogleGoogle Merchant Center Helpchecked 20 November 2026
  13. How to onboard to the UCP integration hub in Merchant CenterGoogle Merchant Center Helpchecked 20 November 2026
  14. Shopping updates from Google Marketing Live (May 20, 2026)Googlechecked 20 November 2026
  15. Product feed schemaOpenAI developer docschecked 20 November 2026
  16. Submit product feedsOpenAI developer docschecked 20 November 2026
  17. ProductSchema.orgchecked 20 November 2026
  18. FAQPageSchema.orgchecked 20 November 2026
  19. Merchant listing (Product, Offer) structured dataGoogle Search Centralchecked 20 November 2026
  20. Product variant structured dataGoogle Search Centralchecked 20 November 2026

You choose what goes in and where it goes out.

Central AI does everything in between. Your first 3 products are free.

  1. Your inputYour shop, your files or your product pages.
  2. Central AIResearches, verifies and writes everything in between. Every value is backed by sources you can check; what the sources do not support is held back.
  3. Your channelsProduct data and listings for the channels you choose.