Emmanuel Flossie, a Google Shopping specialist and Google Ads Diamond Product Expert who runs the feed management consultancy FeedArmy, published a video guide on May 24, 2026 walking through eight product feed attributes that Google has explicitly connected to AI-driven surfaces, including AI Mode in Google Search and AI Overviews. Several of the attributes were first spotted by Hana Kobzova, a PPC specialist who publishes daily coverage on the subject via LinkedIn, and two were identified by Flossie himself.

The video arrives at a moment when the commercial stakes of feed quality have climbed sharply. AI Mode, which surpassed one billion monthly active users globally as of May 2026, increasingly serves as the surface through which shoppers discover and compare products. Queries in AI Mode are, according to Google, on average three times longer than traditional searches. That shift creates a structural problem for product feeds built around short keyword matching: the structured data that once served standard Shopping ads may not carry enough depth to surface correctly in a conversational response.

Flossie opened the video with a performance figure drawn from his own client work. According to FeedArmy, one client saw "an 80% improvement in Google ad performance" after improving product data alone - without changing bids or creative assets. The case study underpins the central argument of the guide: that data quality, not campaign mechanics, is the bottleneck for many merchants competing on AI-driven surfaces.

Why these eight attributes matter now

Google has been adding new Merchant Center attributes designed for conversational commerce since at least January 2026, when the company announced dozens of new data attributes aimed at AI Mode, Gemini, and Business Agent. The attributes Flossie covers are a subset of that expansion - specifically the ones Google has publicly tied to AI-driven surfaces in its own documentation.

According to Google Merchant Center's help documentation, several of the eight attributes carry an explicit note: "This attribute is primarily intended for use in conversational experiences such as AI Mode in Google Search." That designation appears on the popularity rank, document link, question and answer, and related product attributes, among others. The phrasing is deliberate. It signals that these fields were not retrofitted from existing Shopping infrastructure but designed specifically for the conversational query patterns that define AI Mode.

The broader context matters for feed managers. Google's Shopping Graph contains more than 60 billion product listings, with over 2 billion receiving updates every hour. Both organic and sponsored AI Mode surfaces draw from that graph. A product with incomplete or poorly structured data does not disappear from the graph - it simply becomes harder for Google's AI systems to represent accurately when constructing an answer to a natural-language shopping query. As PPC Land noted in its coverage of sponsored stores and quick results inside AI Mode, data quality determines visibility regardless of ad spend.

The eight attributes

Product highlight

The product highlight attribute accepts short sentence fragments describing the key selling benefits of a product - not technical specifications, which belong elsewhere. According to Google's Merchant Center documentation, submitting this attribute helps customers discover products on AI-driven surfaces like AI Mode. Each highlight must be between 1 and 150 characters. Google recommends 4 to 6 highlights per product, with a minimum of 2 and a maximum of 100.

Formatting is where many submissions break. According to FeedArmy, the most common mistake is submitting product highlights in a CSV file when the highlight text contains commas. A CSV parser treats commas as column separators, which splits a single highlight into two fragments. A highlight reading "titanium case, and a green nylon band" becomes two separate values - the second being merely "and a green nylon band." The fix is to use TSV (tab-separated values) files instead, or to wrap comma-containing values in double quotes in CSV. In Google Sheets, a backslash placed before each comma prevents the parser from misreading the value.

The attribute is a repeated field, meaning multiple highlights are submitted as separate entries rather than a single concatenated string. Google's documentation is explicit that highlights should focus on selling benefits, not on technical specifications - that distinction separates product highlight from the next attribute.

Product detail

Where product highlight handles benefits, product detail handles technical specifications. According to Google's documentation, submitting this attribute helps customers discover product information across AI-driven surfaces while also enhancing traditional search experiences. The attribute has three sub-attributes that must be submitted together: section name (optional but recommended, such as "General" or "Display"), attribute name (required, such as "Material" or "Resolution"), and attribute value (required, the actual value such as "Titanium" or "432 x 240").

The structure is intentional. Grouping specifications into named sections - Connectivity, Memory, Camera Settings, and so on - gives AI systems a structured schema to draw on when answering queries like "what is a good tactical watch with a long battery life." A flat list of key-value pairs provides less signal than a hierarchically organised specification block.

Google's documentation includes a detailed formatting example for a camera with 14 specification rows across four sections. In TSV files, values containing commas or colons must be enclosed in double quotes. In Google Sheets, commas, colons, and backslashes within attribute values must each be escaped with a backslash.

The documentation warns against duplicating data already present in other attributes, including title, description, question and answer, and product highlight. If a document link attribute is also submitted and the same information is available in the linked PDF, Google will extract the specifications from the document directly and the product detail entry should be omitted for that data.

Variant option

The variant option attribute is, according to FeedArmy, brand new. Google's existing feed specification already covers standard variant dimensions - color, size, material, pattern, age group, and gender. Merchants selling products that differ along dimensions outside those six had no structured way to communicate those differences. A laptop available in two graphics card configurations has no standard attribute for GPU type.

Variant option fills that gap. It has two sub-attributes: name (the type of variant dimension, up to 250 characters, such as "graphic card") and value (the specific option for that product variant, up to 250 characters, such as "GeForce 4070" or "GeForce 5070"). The field is repeatable up to 30 times per product.

According to Google's documentation, the attribute must be submitted alongside item group ID and item group title. All variants sharing the same item group ID must use the same set of name sub-attributes. The product details displayed on the landing page must also match the values submitted in the feed - a requirement that has direct implications for how variant landing pages are structured and what structured data they carry.

The practical effect for AI surfaces is significant. According to FeedArmy, AI services previously had to infer variant differences from product titles - an imprecise method that could misrepresent product configurations in conversational answers. Structured variant option data allows the system to understand differences in explicit, machine-readable terms.

Item group title

The item group title attribute assigns a shared, generic title to all variants within a product group. According to Google's documentation, when item group ID is used to group product variants, Google recommends also submitting item group title to provide a common label that sits above individual variant titles.

The distinction between the two title levels is technical but important. An individual variant title might read "Flip Phone 23rd Generation Ultra Max 128 GB" - specific, with variant-identifying details included. The item group title for that same product would read simply "Flip Phone 23rd Generation" - shorter, without size, color, or configuration details.

Google's documentation sets a character limit of 150 characters for this attribute. Users typically see only the first 70 or fewer characters depending on screen size, making front-loading the most important details a practical requirement. The item group title maps to the Schema.org property ProductGroup.name.

The minimum requirements are strict: the item group title must be identical across all variants sharing the same item group ID, it must differ from the individual variant title, and it must not include variant-specific properties such as size, color, or the values submitted via variant option. Promotional text, pricing, and shipping information are also excluded.

The related product attribute tells Google which other products in the merchant's inventory are connected to a given product, and in what way. According to Google's documentation, this attribute "will be used in conversational AI experiences, for example to suggest other products that a user can purchase together with, or instead of, this product."

Six relationship types are supported: part of set (a chair belonging to a dining table set), required part (a battery for a battery-operated lamp), often bought with (a phone case with a phone), substitute (a comparable printer from the same range), different brand (the same product sold under a different brand name), and accessory (a webcam for a desktop computer).

Each entry requires three sub-attributes submitted together: relationship type, identifier type (either GTIN or the product's own ID from the data source), and identifier (the actual value). The field is repeatable up to 30 times per product. A product can carry multiple relationship types simultaneously - it can have substitutes, required parts, and accessories all declared in the same feed entry. The documentation is explicit on one constraint: identifiers for separate related products must not be comma-separated within a single entry. Each relationship requires its own separate attribute row.

For AI Mode conversations where shoppers ask "what else do I need for this product" or "are there any alternatives," this structured relationship data provides the signal Google's system draws on directly to construct its answer.

Question and answer

The question and answer attribute allows merchants to submit pre-written FAQ content directly in the product feed. According to Google's Merchant Center documentation, this attribute "is primarily intended for use in conversational experiences such as AI Mode in Google Search."

The format uses two required sub-attributes: question and answer, each accepting up to 1,000 characters. Up to 30 question-and-answer pairs can be submitted per product. The total combined character count across all pairs for a single product is capped at 10,000 characters. The attribute can accept user-, merchant-, and manufacturer-authored questions and answers, according to Google's documentation.

Google's documentation sets clear boundaries on what belongs here. Q&A entries should not duplicate data already present in the product title, description, product detail, or product highlight attributes. If a document link attribute is also submitted and the same information can be found in those documents, the Q&A entry for that information should be omitted - Google will extract it from the documents instead. Prices, sale dates, shipping details, and other time-sensitive or offer-related data also do not belong in this attribute.

The same formatting caution that applies to product highlights applies here. Colons and commas within question or answer values can break parsing in CSV files. TSV format or backslash escaping in Google Sheets is the approach Google's documentation recommends.

The document link attribute is, according to FeedArmy, "one of the most powerful attributes in this list" and probably the one most merchants will not have. According to Google's documentation, it is "primarily intended for use in conversational experiences such as AI mode in Google Search."

The mechanism is straightforward: a URL pointing to a publicly accessible PDF file about the product. Google uses the content of that document to answer detailed questions in AI Mode conversations. User manuals, assembly instructions, training guides, package inserts, and product specifications are the recommended document types.

The technical requirements are specific. URLs must start with HTTPS and must be ASCII characters only, RFC 3986 compliant, and up to 2,000 characters in length. The maximum file size is 50 MB. The file must be in PDF format. Googlebot must be able to crawl the document without a login or restrictions in robots.txt. The URL must be stable - not reused for different documents over time. Up to five document links can be submitted per product, with multiple URLs separated by commas in text feeds.

According to Google's documentation, the document should be about the specific product, not the company in general. High-quality, authoritative documents with coherent structure and clear purpose are preferred. Merchants who submit assembly PDFs effectively allow Google to answer "how do I assemble this?" queries in AI Mode directly from their own documentation - a capability that has not previously been available through standard feed attributes.

Popularity rank

The popularity rank attribute communicates how well a product sells relative to the rest of the merchant's inventory, expressed as a number between 0 and 100. According to Google's documentation, the attribute "is primarily intended for use in conversational experiences such as AI Mode in Google Search" and is designed to "help consumers make a more informed buying decision."

The format allows up to one decimal point - 95.5 is a valid value - but the percentage sign must not be included. The ranking must reflect actual sales data. Google's documentation states the value should be "an accurate ranking, reflecting how well your product is selling and correctly ranked against other products in your inventory."

Google's best practice guidance recommends updating the value through a supplemental data source when there is a substantial change in product popularity based on recent sales. This implies the attribute should not be a static value set once and left unchanged - sales patterns shift, and the signal used to surface products in AI Mode queries should reflect current performance rather than a snapshot taken at feed setup.

According to FeedArmy, a practical starting approach for merchants without sophisticated sales ranking infrastructure is to assign scores between 90 and 100 to the top 10 sellers, between 50 and 80 to mid-range performers, and below 50 to slow-moving or older inventory, then refine the values over time as sales data becomes more granular.

Feed completeness beyond the eight

Flossie was direct in the video about the scope of the guidance. According to FeedArmy, the eight attributes covered are the ones Google explicitly tied to AI surfaces - but they are not a substitute for overall feed quality. "Every attribute you improve in your feed contributes. Your titles, your descriptions, your product identifiers, your categories, your images, all of it matters."

That framing aligns with what PPC Land has documented across its coverage of the Google Shopping ecosystem: as the Shopping Graph expands to power more surfaces simultaneously, the quality and completeness of product data in Merchant Center has become a performance variable that extends beyond any individual campaign. Google's AI Max for Shopping campaigns, announced on April 30, 2026, explicitly draws on feed attributes including material, fit, and durability to generate ad copy tailored to specific queries. A sparse feed constrains that system just as it constrains AI Mode visibility.

The eight attributes Google has flagged represent a structured baseline - a minimum set that addresses the specific mechanics of conversational AI surfaces. Whether a merchant starts there and expands, or audits the entire feed simultaneously, the direction is the same: more structured, more complete, and more semantically rich product data.

Timeline

Summary

Who: Emmanuel Flossie, Google Shopping Specialist, Google Ads Diamond Product Expert, and founder of FeedArmy. Several of the newly identified attributes were first spotted by Hana Kobzova, a PPC specialist. The attributes themselves are defined by Google in its Merchant Center product data specification.

What: A video guide published on May 24, 2026, detailing eight Google Merchant Center product feed attributes - product highlight, product detail, variant option, item group title, related product, question and answer, document link, and popularity rank - that Google has explicitly connected to AI-driven surfaces including AI Mode and AI Overviews.

When: The video was published on May 24, 2026. The attributes became part of Google's Merchant Center documentation following the broader conversational attribute rollout that began in January 2026.

Where: The attributes are submitted through Google Merchant Center as part of a product data feed. They apply in all countries, according to Google's documentation. The video was published on the FeedArmy YouTube channel and shared on LinkedIn.

Why: AI Mode and AI Overviews generate product answers from structured Merchant Center data. Products with incomplete or unstructured feed data are harder for Google's AI systems to represent accurately in conversational responses. The eight attributes flagged by Google address the specific data signals those systems use for conversational product discovery - including variant differentiation, product relationships, popularity signals, and document-level product information.

Share this article
The link has been copied!