Google is rolling out a set of changes that shift product discovery from keyword matching to conversational, intent-driven experiences. AI will directly suggest your products to potential buyers.
Two parts matter most for retailers and agencies managing Google Shopping at scale:
- Business Agent on Google Search
Business Agent is a new conversational experience on Google Search where shoppers can “chat” with a brand, like a virtual store associate. Google positions it as an evolution of the earlier brand-profile Q&A experience, with more customisation and deeper data integration. It can use Merchant Center shopping data plus information from your website to answer questions, and Google notes additional capabilities like training the agent with business data and performance insights are coming. - New Merchant Center data attributes designed for conversational commerce
Google also announced dozens of new Merchant Center attributes aimed at improving discovery in conversational surfaces like AI Mode, Gemini, and Business Agent. These attributes “complement” existing feeds and go beyond traditional keywords, including things like answers to common product questions, compatible accessories, and substitutes.
The practical takeaway is simple: if a shopper can ask it, you should be able to represent it in structured product data within your Google Shopping feed.
Why this makes feeds more important (and more operational)
Historically, Google Shopping feeds were a blend of compliance (meet spec, avoid disapprovals) and performance tuning (titles, product_type, labels, GTIN coverage, pricing accuracy). That still matters a lot, but conversational shopping adds a new requirement:
- The “answer” often needs to be constructed from product data, not inferred from a thin title.
- Eligibility and relevance will increasingly depend on attribute coverage, consistency, and specificity.
- Your feed becomes the system of record for what an AI assistant can safely and confidently say about your catalog.
Google’s own Merchant Center documentation already frames correct formatting and completeness as essential, since inaccurate or missing product information can prevent ads and listings from showing. With agentic interfaces, the penalty for weak data is not only reduced reach, it is also weaker answers, fewer qualified clicks, and more “wrong product” conversations.
What “conversational attributes” really mean in practice
From Google’s announcement, the direction is clear: move beyond keywords and expand structured information that maps to natural questions:
- Product specifications (structured specs that typically live on PDPs)
- Q&A lists (common questions and answers that shoppers ask)
- Feature lists (bullet highlights, differentiators)
- Compatibility information (accessories, spare parts, substitutes)
- Additional descriptive facets like shapes, flavors, themes (category-dependent)
This changes how you should think about feed optimisation. It is less “make the title longer” and more “make the product understandable to an assistant”.
- Examples of shopper questions your data should answer“Is this case compatible with iPhone 15 Pro Max?”
- “Does this sofa fit through a 76 cm doorway?”
- “Is this supplement vegan and gluten-free?”
- “What’s the difference between Model A and Model B?”
- “What spare filter fits this air purifier?”
If you cannot represent those answers in clean attributes (or you represent them inconsistently), the assistant either will not show your item, or it will give a generic answer that might not convert that well.
Feed readiness checklist for Business Agent and AI surfaces
Here is a practical checklist you can use to assess your current Google Shopping feed and catalog data.
1) Get the basics right
This is still non-negotiable because Merchant Center eligibility depends on it.
- IDs, prices, availability, links, image_link consistency
- GTIN/MPN/brand coverage where relevant
- category mapping (google_product_category), product_type hygiene
- shipping and tax configuration accuracy
If these are unstable, any higher-order enrichment becomes fragile.
2) Treat PDP content as a data source, not just copy
Most of the “conversational” value already exists somewhere on your site, but it is unstructured:
- specification tables
- bullet lists
- FAQs
- size guides
- compatibility lists
- manuals, ingredient lists, care instructions
- product images
The task is to extract, normalise, and publish it into dedicated fields that can travel through Google Merchant Center and into AI experiences.
3) Build a compatibility layer (accessories, parts, substitutes)
Google explicitly called out compatibility and substitutes as examples of the new attribute direction. For many verticals, compatibility is where conversion is won or lost:
- electronics (cases, chargers, mounts)
- appliances (filters, spare parts)
- automotive (fitment)
- fashion (styling, matching sets)
- beauty (shade equivalents, refills)
Even a lightweight version helps: parent-child relationships, “works_with” groupings, or curated accessory mappings.
4) Create Q&A that matches real shopping language
Business Agent can answer product questions in your brand’s voice and use Merchant Center plus website information. That implies you should:
- standardise “common questions” per category
- ensure answers are consistent with policy and PDP truth
- avoid ambiguous claims (especially regulated categories)
5) Put governance and QA around enrichment
As soon as you start generating or extracting attributes at scale, you need controls:
- validation rules (allowed values, regex, length limits)
- confidence thresholds for extracted attributes
- audit trails for what changed and why
- sampling workflows (spot-check by category, brand, price band)
- monitoring for drift when PDP templates change
Where Feedoptimise fits: enrichment, extraction, and feed operations
Feedoptimise is built around the idea that feed management is ongoing data work, not a one-off export. On the platform side, that shows up in three capabilities that are directly relevant to Google’s agentic shift:
1) Scale feed modifications and channel-specific rules
Feedoptimise supports bulk data modification using rules and formulas for converting, extracting, validating, merging from remote files, and more, with changes reviewable in real time. This is the foundation for building “conversational attribute” outputs that differ by category or market without rewriting your source catalog.
2) AI-powered attribute extraction and completion
Feedoptimise explicitly supports AI-powered feed content generation and AI-assisted extraction to fill missing attributes (examples given include color, material, gender), alongside bulk generation and rewriting of titles, descriptions, categories, and other attributes.
That’s the key unlock for conversational commerce: you can turn messy PDP text into structured fields at scale, without waiting for engineering to remodel your product database. You can also extract missing PDP information from product images, where it’s present.
Practical use cases for the new Google attributes include:
- extracting spec pairs from description and spec tables into a normalized spec block
- extracting specs or missing informations from images using vision AI
- generating a clean feature list from long-form descriptions
- drafting Q&A answers grounded in your PDP and policy constraints
- deriving compatibility tags from model names and fitment notes
- inferring missing attributes (material, pattern, finish, use-case) to improve match quality
3) QA, reporting, and controlled experimentation
Feedoptimise includes built-in quality assurance reporting and monitoring for feed attributes, plus the ability to A/B test data variants by toggling modifications. For conversational attributes, this matters because you will want to test questions like:
- Do generated feature lists improve CTR versus manufacturer bullets?
- Does stricter compatibility data reduce returns?
- Which spec formats lead to better visibility in AI surfaces?
The new optimization baseline: “Can an assistant sell this item correctly?”
Google is aligning Search, Shopping, and Gemini toward conversational discovery, and it is explicitly investing in Merchant Center attributes and Business Agent as the data and interaction layer. That raises the bar for feed quality. A good feed is no longer just compliant and keyword-rich. It is complete, structured, and answerable.
If you want Feedoptimise to support this, the most valuable starting point is usually an enrichment audit focused on:
- which high-intent questions your products should answer
- which attributes you already have vs. can extract
- what validation and QA you need before scaling AI enrichment across the full catalog
if you need any help, make sure to get in touch using our contact page.