Business and Product Catalog. The two layers AI agents read.

When a consumer asks ChatGPT “where can I find a good bike shop in Oakland,” the agent is not looking at a single signal. It is reading a stack of them. Is there a business here? Is it real? What does it sell? Is it trusted? Can I recommend it confidently?

Each question is answered by a different layer of merchant data, stored in a different place, governed by different rules. A merchant who shows up cleanly in every layer gets cited. A merchant missing any layer gets skipped — not because the agent dislikes them, but because the answer does not hold together.

A-Comm evaluates two layers of merchant data, because that is what AI agents actually read. We call them the business identity layer and the product catalog layer. Both matter. Neither replaces the other. They are parallel pipelines into the agent’s retrieval.

Business identity. AI agents answer business-level queries — “best [type of business] near me,” “is [brand] legitimate,” “what does [company] do” — by pulling from sources that describe the merchant as an entity. The sources are different per platform, but the question is the same: can the agent resolve this merchant cleanly as a real, identifiable business?

Your website’s Organization and LocalBusiness schema tells agents what kind of business you run, where, and for whom. Partner directories extend that identity signal. OpenAI has a confirmed Foursquare partnership for place data, which informs ChatGPT’s local business responses. Gemini is influenced by Google Business Profile and the broader Maps and Search ecosystem. Copilot draws on Bing Places and the Bing local index for entity data. Perplexity’s local and travel surfaces appear to leverage Yelp, Tripadvisor, and publisher partnerships — the exact ingestion mechanics are not fully documented by Perplexity. Observed trust and authority signals — review density, business verification, domain age, NAP consistency — appear to weight entity credibility in industry practice, though no platform publishes a deterministic ruleset for how they are scored. Long-tail content on your domain — FAQs, policies, how-to articles, blog posts — establishes subject-matter authority so the agent cites you in conversational responses and can surface your products downstream.

A merchant with a strong product catalog but a thin, misaligned, or unverified business identity is invisible for entity-level queries. They never come up when a consumer asks “who makes the best [category],” because the agent cannot resolve the entity in the first place. The products exist. The business does not register.

Product catalog. AI agents answer purchase-intent queries — “best [product] for [use case],” “compare X vs Y,” “where to buy Z,” “what’s the price of [item]” — by pulling from structured product data. This is where SKUs, prices, availability, variants, and transactional fields live.

ChatGPT’s Commerce feed is a typed JSON schema merchants submit to make products eligible for inline citation in ChatGPT shopping responses — OpenAI publishes the spec. Google Merchant Center feeds Google’s shopping surfaces; Google’s own Merchant Center documentation lists Gemini alongside Search, Maps, YouTube, the Shopping tab, Images, and Lens as places where free product listings may appear. Microsoft Merchant Center is a core product-catalog input for Bing and Microsoft shopping surfaces that feed Copilot commerce experiences. Perplexity Shopping surfaces merchant products via its Buy with Pro integration; Shopify has published guidance for merchants optimizing for Perplexity, though Perplexity has not fully documented its ingestion mechanics. Schema.org Product and Offer markup on your product pages carries catalog signal even before a merchant has submitted any feed.

A merchant with a strong business identity but a missing, sparse, or malformed catalog is cited for brand queries but skipped during purchase decisions. The agent knows the brand exists. It just cannot select one of the merchant’s products to recommend with confidence — so it recommends someone else’s.

Each layer answers a distinct class of consumer question. Neither layer can substitute for the other.

Catalog strong, identity weak. The merchant shows up in product comparisons but loses trust-weighted selection. The agent says “there are three options — Brand A, Brand B, and [merchant]. Brand A is well-reviewed and has been making these for 20 years. I’d go with Brand A.” The merchant gets cited by name, then recommended past.

Identity strong, catalog missing. Consumers find the brand. They cannot find the SKUs. The agent says “check their website directly — I don’t have current pricing or availability.” Every such answer is a lost conversion.

Both weak. The merchant is not in the response at all. The agent recommends competitors by default.

Both strong. The agent names the merchant, cites a specific product, quotes the price, confirms availability, and hands off a purchase-ready link. That is what full discoverability looks like.

Not every merchant has a catalog. A plumbing company, a law firm, a dentist, a hair salon, a consultancy — these businesses have no product catalog. Their discoverability surface is entirely the business-identity layer: website, partner directories, reviews, authority content, NAP consistency. For them, a catalog scanner returns nothing. A business scanner returns everything they need to show up in AI responses to “best [service] near me” or “recommended [specialist] in [neighborhood].” A-Comm’s Business Scanner evaluates the identity layer in full, independent of any catalog. A merchant without products is not a merchant without discoverability — the identity layer is often all they need.

Each platform weights the two layers differently and draws from different sources. The map below mixes publicly documented merchant integrations with reported partnerships and observed ingestion behavior — no AI platform publishes an exhaustive ingestion graph for its shopping and entity surfaces. A merchant visible across both layers on all four platforms is reaching the agent through multiple parallel inputs.

Mix of publicly documented merchant integrations (OpenAI Commerce, Google Merchant Center, Microsoft Merchant Center), reported partnerships (OpenAI–Foursquare for place data), and observed ingestion behavior. Platforms do not disclose exhaustive ingestion graphs; specifics evolve as platforms iterate.

The input sets for the two layers are almost entirely disjoint. Foursquare does not carry SKUs. Google Merchant Center does not resolve whether a merchant is trusted. Bing Places does not know what is in stock. These are parallel pipelines into agent retrieval, not a hierarchy.

AI agents have already changed how consumers discover, evaluate, and buy from merchants. The surface area of discoverability has fractured across four platforms, eight pipelines, dozens of source-specific requirements, and two distinct identity layers. A merchant trying to manage this by hand is running a full-time data-operations job they did not sign up for.

A-Comm translates the merchant’s reality — their business, their catalog, their brand voice — into the formats each AI agent expects, and publishes that translation to every surface those agents read. One canonical model. Eight publication targets. Two identity layers. Continuous sync.

Both layers, read together, in every language each agent speaks.

A note on sources. The platform attributions in this piece mix publicly documented merchant integrations (OpenAI’s Commerce feed spec, Google’s Merchant Center documentation, Microsoft Merchant Center) with reported partnerships (OpenAI’s Foursquare partnership for place data) and observed ingestion behavior (Perplexity’s apparent use of Yelp and Tripadvisor, Copilot commerce surfaces drawing on Bing Places and Microsoft Merchant Center). No AI platform publishes an exhaustive ingestion graph for its shopping and entity surfaces, and the exact internal weightings shift as platforms iterate. A-Comm maintains an internal research registry tracking each platform’s documented spec, observed behavior, and confirmed partnerships; the specifics here are a snapshot as of April 2026.