ChatGPT Merchant Center: why your product feed now decides your AI visibility

This week, OpenAI added product feeds to its Ads Manager. Retailers can now send their catalog straight to ChatGPT through a structured feed, the same way they have fed Google Merchant Center for years. Until now, ChatGPT pieced products together by crawling the web or through a handful of partner integrations. Moving to a proprietary feed shifts the question entirely: your visibility in ChatGPT will no longer depend on what the model can guess, but on what your product feed gives it to read.

What changes when OpenAI opens product feeds

For a long time, ChatGPT rebuilt product information by hand, so to speak. It crawled pages, cross-checked descriptions, leaned on a few partner integrations to fill the gaps. Clever, but fragile. One ambiguous title, a stale price or a wrong availability flag, and the product dropped out of the model's reasoning.

With product feeds open inside the Ads Manager, the logic flips. You no longer let the assistant guess what you sell, you declare it. The feed becomes the source of truth, structured and kept current, that ChatGPT refers to when answering shopping queries. That is exactly the model behind Google Merchant Center and its role in product distribution: a normalized catalog, explicit attributes, regular syncing. OpenAI is taking the same path, and moving fast.

Why Google keeps an edge

One factor still weighs heavily: data. Google spent more than fifteen years building proprietary databases whose depth is hard to catch up with. Shopping, Maps, Flights, Travel, all layers of structured, verified, cross-referenced data that now feed its AI answers.

That lead is not just about history. It shapes how reliable the answers are. An assistant that already knows a product's availability, price and location reasons better than one discovering them. By opening its own feeds, OpenAI is chasing exactly that: a native product base rather than dependence on what it scrapes from the web.

The takeaway is simple. The AI race runs on data, and product data is your feed. That same data battle shows up in how ChatGPT Shopping positions itself against Google, each one trying to become the entry point for buying intent.

What a ChatGPT product feed demands on the data side

A feed that performs in an AI environment does not look like a feed built for search ranking alone. AI does not read your catalog like a classic engine. It looks for explicit, unambiguous signals it can reuse in an answer without risk of being wrong. Three families of signals live inside a feed.

Signals AI reads effortlessly

• Normalized identifiers (GTIN, MPN, brand) that tie the product to a known entity.
• Descriptive, structured titles where the product type comes before the variants.
• Category attributes (size, color, material, gender) filled in consistently.
• Price and availability synced in near real time.

Signals AI misreads

• Marketing titles loaded with superlatives, where the actual product has to be guessed.
• Attributes left empty or filled with defaults, which the model then fills with assumptions.
• Variants poorly linked together, making one product look like several listings.

Signals AI ignores

• Decorative content that carries no decision-relevant information.
• Keywords stacked without structure, inherited from old SEO habits.
• Descriptions duplicated from one product to the next.

This is where enrichment matters. Filling missing attributes, normalizing identifiers and clarifying titles turns a raw feed into a source of truth. Requirements vary by channel, as the different Google Shopping feed types and their uses show. The same discipline connects to Merchant Center attributes and the UCP protocol, which set the base for a catalog agents can actually read.

Google Merchant Center and ChatGPT Merchant Center: what transfers, what differs

Good news for teams already fluent in Google Merchant Center: most of your work carries over. Feed discipline, attribute normalization, rigor on identifiers and availability all hold in both environments.

What transfers directly:

• catalog structure and attribute logic,
• product identifiers (GTIN, MPN),
• price and availability management,
• the freshness requirement on data.

What clearly differs:

• ChatGPT reasons in natural language and intent, not keyword queries. Your titles and attributes must answer questions, not just contain terms.
• the agentic environment adds a direct transactional layer, with instant checkout and the Agentic Commerce Protocol, where the purchase sometimes completes without leaving the assistant.
• ranking logic stays opaque and shifting, where Merchant Center exposes established rules.

What this means for e-commerce teams

Three concrete consequences are taking shape for brands.

First, the product feed stops being a purely technical topic parked inside acquisition. It becomes a visibility asset in its own right, on par with editorial content. Attribute quality decides your presence in AI answers.

Second, the growing list of destinations (Google, ChatGPT, and others to come) makes cross-channel consistency critical. The same product described differently across platforms blurs the assistants' reasoning. That mechanic already appears in agent-driven product discovery in e-commerce, where the smallest inconsistency pushes a product out of the recommendation.

Third, freshness becomes a measurable competitive advantage. Unsynced stock or a stale price no longer just hurts your campaigns: it removes you from AI recommendations, which favor verifiable data.

Key takeaways

• OpenAI opening product feeds brings ChatGPT closer to the Google Merchant Center model.
• Visibility in ChatGPT depends on product feed quality, not page volume.
• Normalized identifiers, explicit attributes and synced availability are the signals AI reads most.
• Most of the work done for Google Merchant Center transfers to AI-bound feeds.
• Cross-channel consistency and data freshness become ranking factors.

The shift to agentic commerce will not reward the biggest catalogs, but the best described ones. If your attributes are clean, your identifiers normalized and your stock current, you start with a head start. That is exactly the ground Feed Enrich works on to enrich and clean up your product feed, so it stays readable by engines and AI assistants alike.

FAQ

What is ChatGPT Merchant Center?

ChatGPT Merchant Center refers to the ability, opened by OpenAI inside its Ads Manager, to submit a structured product feed to ChatGPT. Retailers send their catalog directly to OpenAI, the way they do with Google Merchant Center, instead of letting the model reconstruct products by crawling the web.

How does a ChatGPT product feed differ from Google Merchant Center?

Feed structure and attributes are close, but ChatGPT reasons in natural language and intent rather than keyword queries. The agentic environment also adds a direct transactional layer through instant checkout and the Agentic Commerce Protocol.

Which attributes does a ChatGPT product feed require?

The most readable signals are normalized identifiers (GTIN, MPN, brand), structured descriptive titles, consistent category attributes, and price plus availability synced in near real time.

Should you drop Google Merchant Center for OpenAI?

No. Google keeps an edge tied to its proprietary databases. The point is not to choose, but to maintain a single clean product source distributed consistently to both environments.

How do you prepare a product catalog for AI search?

By fixing identifiers, enriching missing attributes, removing ambiguous marketing titles and keeping stock and prices fresh.

Will agentic commerce replace Google Shopping?

Nothing points to a near-term replacement. Agentic commerce adds a discovery and purchase channel. Google Shopping and AI feeds coexist and rely on the same raw material: structured product data.

Why does feed quality matter more than volume for AI?

An AI only reuses a product in an answer if it can do so without risk of error. One explicit, verifiable attribute is worth more than thousands of poorly structured pages.