Table of Contents
- What Are Conversational Attributes?
- Attribute 1: question_and_answer
- Attribute 2: document_link
- Attribute 3: related_product
- Attributes 4 to 6: item_group_title, variant_option, popularity_rank
- How to Submit Them: Feed Rules, Supplemental Feeds or API
- Compliance Risks Hidden in Conversational Data
- How to Measure Whether It Helps
- Quick Pre-Launch Checklist
- Frequently Asked Questions
What Are Conversational Attributes?
Google Merchant Center's help center now documents a group of six conversational attributes. Search Engine Land explained them on October 22, 2026. They are optional fields that give AI-driven surfaces, such as AI Mode in Search, richer machine-readable detail about a product than a title, price and image can carry. Google says the attributes are meant to help products show up for nuanced searches and to surface complementary items, while also improving traditional search experiences.
The important mindset shift: these fields are additive metadata. Third-party guides, including a SKU Analyzer explainer, note that submitting them does not change the approval status of existing products, and a badly formatted value generally just fails to be used rather than getting the product disapproved. That makes them a low-risk place to experiment, as long as the content you submit is true. Google's own holiday announcement in September said that when lululemon submitted conversational attributes, they were incorporated in relevant AI Mode recommendations about 50% of the time during testing. That is one brand's test, not a promise for your catalog.
Attribute names, limits and the lululemon figure come from Google's Merchant Center help documentation and Google's September 16, 2026 retail blog post, as reported by Search Engine Land. Re-check the live help page before building feed rules, since specs change.
Attribute 1: question_and_answer
The question_and_answer attribute lets you answer the specific questions customers ask, written in natural conversational language. It takes two sub-attributes, [question] and [answer]. Each has a 1,000-character limit, and you can supply up to 30 question and answer pairs per product.
Your existing FAQ section, support tickets, pre-sale chat logs and manufacturer documentation are the best raw material. Think about user, merchant and manufacturer questions: Does it fit a 15-inch laptop? Is it dishwasher safe? Can I use it with a gas stove? Answers should be literal and verifiable, not marketing copy.
An AI surface may repeat your answer to a shopper as fact. If the answer is wrong about safety, compatibility or materials, you have a misrepresentation problem that lives outside your landing page. Only submit answers that your product page, policies or manufacturer documents support.
Attribute 2: document_link
The document_link attribute attaches PDF documentation: user manuals, assembly instructions, training guides or packing inserts. The value is a URL starting with http:// or https://, limited to 2,000 characters, and you can attach up to five documents per product. Conversational AI can use the document to answer detailed questions about the product.
Practical tips: host PDFs on your own domain over HTTPS, keep file names descriptive, make sure the link does not require a login, and do not point to a PDF that contains promotional pricing that conflicts with your feed. Documents that include prices, shipping promises or competitor comparisons create unnecessary mismatch risk. Review our guide to product feed optimization for the baseline feed hygiene this builds on.
Attribute 3: related_product
The related_product attribute describes how other products connect to this one: frequently bought together items, accessories, substitutes or alternatives. A product can have several relationship types. It uses three required sub-attributes, including a relationship type, and the remaining fields identify the related item. Check the current help page for the exact sub-attribute names and accepted relationship values before you map them.
This is the attribute most closely tied to the 'showcase complementary items' goal Google describes. For a store selling a camera, an accessory relationship could point to a memory card; a substitute relationship could point to a cheaper body. Only relate items you actually sell and that are in stock. A related-product link to a disapproved or out-of-stock item is a bad experience and wastes the slot.
Attributes 4 to 6: item_group_title, variant_option, popularity_rank
The remaining three attributes are structured around variants and demand signals:
- item_group_title gives a clean title for the whole group of variants, as opposed to the title for a single color or size. This is useful when your per-variant titles carry long suffixes.
- variant_option describes the specific option that makes this item different from its siblings, giving AI systems a consistent way to distinguish a blue large from a red small.
- popularity_rank communicates how popular an item is relative to your own catalog. Because it is a claim, base it on real sales data and refresh it on a schedule; do not label everything number one.
If your catalog has messy variants, fix grouping first. Our walkthrough on feed structure and the 2026 specification timeline show where these fields fit among other new attributes.
How to Submit Them: Feed Rules, Supplemental Feeds or API
Google's documentation describes adding the attributes through the same mapping and rules systems you already use, and via either a feed file or the API. For most small and mid-size stores the cleanest path is a supplemental feed keyed on product ID, so you do not touch the primary feed that your platform app controls.
- Pick 20 hero products by revenue. Start with products that have the most customer questions.
- Build the content in a sheet: product ID, up to a dozen Q&A pairs, PDF links, related product IDs.
- Upload as a supplemental feed and confirm processing without item-level warnings.
- Check Diagnostics after processing and compare product approval counts before and after.
- Expand by category and keep the sheet as your source of truth.
If your platform plugin cannot output these fields, see our feed method migration guide for supplemental and API routes, and multi-channel feed management if several tools write to the same feed.
Compliance Risks Hidden in Conversational Data
Because these attributes are written in plain language, they can accidentally contain claims that Google's policies restrict. Health claims in a Q&A answer, guaranteed results, medical-sounding statements for supplements, or unsubstantiated sustainability language will be read just as carefully as the same claims on your page. If you sell in sensitive verticals, treat every answer as ad copy. See eco claims and food and beverage compliance for examples of claims that routinely cause trouble.
Also keep shipping, return and price statements out of Q&A entirely. Those belong in their dedicated attributes and policy pages. A Q&A saying 'free returns for 90 days' when your return policy says 30 is a textbook mismatch. Read return policy requirements to make sure the pages these answers rely on are solid.
How to Measure Whether It Helps
Google has not published a guaranteed uplift. Measurement is therefore your job. Use the new AI performance insights, covered in our AI performance insights guide, to watch your share of voice on AI surfaces for the categories you enrich. Keep a control group: leave similar products without the attributes for four to six weeks and compare impression and click trends. Because AI surface attribution is still murky, avoid making budget decisions on week-one data.
Also watch quality signals: customer service questions that the Q&A should have answered, returns citing 'not as described', and any new Diagnostics warnings. If returns for enriched products rise, your content is misleading someone.
Quick Pre-Launch Checklist
Before you submit conversational attributes
Every answer is verifiable on the product page or manufacturer document
PDF links are public, HTTPS, and free of conflicting prices
Related products are live, approved and in stock
popularity_rank is based on real sales data with a refresh date
No health, income or guarantee claims appear in any answer
A supplemental feed or rule is in place and a rollback plan exists
Run the holiday product data readiness checklist alongside this one, since peak season is when mismatches get punished fastest.
Frequently Asked Questions
Are conversational attributes required? No. They are optional and additive. Not submitting them does not change your eligibility for standard Shopping ads or free listings.
Can bad conversational data get my account suspended? Third-party documentation indicates misformatted values are simply not used. However, false claims in the content can create policy and misrepresentation risk the same way false claims on a landing page can, so accuracy matters more than format.
How many Q&A pairs should I add per product? Google allows up to 30, but quality beats quantity. Start with the five to ten questions your customers really ask.
Does this replace structured data on my site? No. Keep your on-page Product schema and feed in sync. Our guide to Shopping measurement setup covers what to track alongside it.
Make Sure Your Store Passes Before You Add AI Fields
Richer data only helps if your site, policies and feed are clean. A free scan at gmcunbanned.com checks your store for the policy and landing page issues that most often trigger suspensions.
Run Free GMC Scan →