The Six Feed Fields Google Built for AI Shopping

The six feed fields Google built for AI shopping

In May 2026, Google announced a set of new Merchant Center feed attributes designed specifically for conversational AI surfaces like AI Mode in Search and the Gemini app. Google’s own documentation calls them “conversational attributes” and marks each one with an explicit note: “This attribute is primarily intended for use in conversational experiences such as AI Mode in Google Search.”

These are not the standard attributes you already know — brand, GTIN, price, shipping. They are entirely new fields created to answer the types of questions shoppers now ask AI assistants: Does this have a headphone jack? What else do I need to use this? Which version is your most popular?

The six attributes are:

• Question and answer (question_and_answer): Up to 30 pre-written FAQ pairs per product. Google uses them to answer detailed shopper questions inside AI Mode responses. Base them on the questions in your support tickets and reviews.
• Document link (document_link): URLs to PDF manuals, assembly guides, and spec sheets. Google crawls these and uses the content to answer technical questions in AI conversations.
• Related product (related_product): Declares relationships between products — accessory, required part, often bought with, substitute. When a shopper asks AI Mode what else they need, this attribute drives the answer.
• Item group title (item_group_title): A shared group-level name for all your variants. Helps AI understand that twelve feed entries are one product in six colors and two sizes, not twelve separate products.
• Variant option (variant_option): Names the dimension that distinguishes variants beyond the standard color, size, and material — for example, Graphics Card: GeForce 5070. Used alongside item group ID.
• Popularity rank (popularity_rank): A number between 0 and 100 showing how popular a product is relative to the rest of your inventory. AI uses it when a shopper asks for your most popular model.

All six are optional. Adding them does not affect the approval status of your existing products. You deploy them through a supplemental data source in Merchant Center — a separate table that Google joins with your primary feed automatically.

Why feed data matters more than your product page

AI shopping inside Google does not start with your product page. It starts with the Shopping Graph — Google’s index of more than 60 billion product listings, with over 2 billion refreshed every hour. When a shopper asks AI Mode a question, the system reads the Shopping Graph, not your website.

This means the attributes in your Merchant Center feed are the primary input Google’s AI uses to decide whether your product matches the query. A well-structured product page still matters for traditional search and for the AI engines that crawl the web directly — ChatGPT, Perplexity, and Claude read your page differently than Google reads your feed. But inside Google’s ecosystem, the feed is the source of truth.

The shift is structural. Queries in AI Mode are, on average, three times longer than traditional searches. A classic feed built around a title, price, and description cannot answer “which headphones under 100 dollars have a headphone jack and 40 hours of battery?” The conversational attributes exist because the old feed vocabulary was not built for that question.

The evidence: most sellers have not added them

An anonymized FirstShelf audit of 24 marketplace listings over 90 days scored entity authority — the structured-data layer AI shopping engines read to match products to queries — at just 32.5 out of 100. That score reflects standard attributes being incomplete. The new conversational attributes are almost entirely absent from the feeds we audit.

Practitioner evidence reinforces the gap. FeedArmy, a Google Shopping consultancy and Diamond Product Expert, reported that one client saw an 80% improvement in Google Ads performance after improving product data alone — without changing bids or creative assets. That case study predates the conversational attributes, but it underscores the broader point: data quality, not campaign mechanics, is the bottleneck for merchants competing on AI-driven surfaces.

Meanwhile, the competitive window is wide open. Research published in July 2026 found that 73% of brands ranking on page one of Google still have zero AI mentions. For sellers who add conversational attributes now, the advantage is being legible to AI shopping before competitors are.

Google also launched the measurement layer

Announced alongside the conversational attributes at Google Marketing Live on May 20, 2026, the AI Performance Insights pilot went live for select US Merchant Center accounts on July 13. It is the first native report showing how products surface inside AI Mode, AI Overviews, and the Gemini app.

The report tracks four metrics, all impression-based: Share of Voice (your AI impressions relative to competitors), competitors’ average share, Query Frequency, and Query Type. It covers organic AI traffic only — no paid ads, no clicks, no conversions. Google describes it as helping merchants understand how their brands show up in conversational results.

The pilot is US-only, with Australia, Canada, India, and New Zealand named as next in line, but without a firm date. For sellers outside those markets, the report is a preview of what is coming — and a reason to start optimizing feed data before the measurement arrives.

How to deploy conversational attributes

Google recommends deploying the six attributes through a supplemental data source. You create a table with product IDs and the new attributes, and Merchant Center joins it with your primary feed. The primary feed stays untouched.

Three practical points matter more than the rest:

Formatting breaks more feeds than content. If you use CSV files, commas inside attribute text will split values. A product highlight reading “titanium case, and a green nylon band” becomes two separate entries. Use tab-separated values instead, or escape commas with a backslash in Google Sheets.

Do not duplicate content. Google explicitly warns against copying text from your description, product highlights, or product details into conversational attributes. The question and answer pairs should answer questions your description does not — specific compatibility, use-case, and what-is-included queries.

Map related products with correct identifiers. The related product attribute requires a relationship type (accessory, required part, often bought with), an identifier type (GTIN, MPN, or feed ID), and the identifier itself. An incorrect GTIN silently breaks the link.

For marketplace sellers on Shopify or Etsy, the built-in Google integration may not support these advanced attributes yet. Check whether your feed app allows custom attributes or supplemental feeds. If not, you will need a dedicated feed management tool or a manual supplemental upload through Merchant Center.

What this means for GEO for marketplace sellers

Generative Engine Optimization has mostly focused on website content — atomic passages, entity consistency, structured data on the page. Conversational attributes add a parallel layer: feed-side GEO. You optimize the data source AI shopping reads first, before it ever reaches your product page.

This matters differently depending on where you sell. If you run a Shopify store with a Merchant Center feed, you control the feed directly. If you sell primarily on Etsy, your listings may not flow into Merchant Center at all — but the same principle applies to any AI shopping surface that reads structured product data. The fields an AI assistant uses to recommend your product are not the same fields your marketplace displays to human shoppers. You need to think about both surfaces.

The broader pattern connects to what we found in our analysis of the attribute gap: AI shopping engines filter and compare by structured fields, not keyword prose. Conversational attributes are the next generation of those fields — the ones designed specifically for the questions buyers ask in conversation.

How FirstShelf can help

FirstShelf audits your product data and scores the exact attributes AI shopping engines read — including the entity authority layer where most marketplace listings score below 35 out of 100. We identify which conversational attributes you are missing, which standard attributes are incomplete, and which fields contradict each other between your feed and your visible page. The result is a prioritized fix list, not generic SEO advice.