1. Discovery
A shopper describes a need in natural language, often with several constraints: budget, location, use case, compatibility, or delivery timing.
Ecommerce visibility guide
The new product shelf is a conversation
AI shopping systems do not just match a keyword to a product title. They interpret a need, compare constraints, and explain why an item fits. Your feed, product pages, reviews, policies, and availability data all become part of that decision.
How the journey works
From a natural-language request to a merchant action
Discovery, comparison, and checkout are related but different jobs. A feed makes a catalog legible; product pages and independent sources add context; an agentic commerce protocol connects approved actions to merchant systems.
2. Catalog matching
The system matches the request against product data from feeds, merchant pages, structured metadata, and other sources.
3. Comparison
The answer may compare products or merchants using relevance, availability, price, quality, reviews, and fit for the stated request.
4. Recommendation
The product appears with a generated explanation, simplified attributes, and links or a shopping action. The wording is not necessarily copied from the product page.
5. Conversion
Where supported, an agentic commerce flow can hand the user to the merchant’s checkout or complete an approved action while the merchant remains responsible for fulfillment and support.
Catalog completeness
The fields that determine whether a product is understandable
A product feed is not a second homepage. It is a structured source of truth that should agree with the pages customers and agents can inspect. More attributes are useful only when they are accurate, current, and meaningful for the category.
Identity
Stable product IDs, brand, GTIN/SKU where applicable, canonical URLs, and clear variant relationships.
Prevents an AI system from merging two products or treating a variant as a separate offer.
Conversational product language
Specific titles and descriptions that explain who the product is for, what it does, and which constraints it satisfies.
Helps a product match natural-language shopping questions instead of only exact product names.
Price and availability
Current price, currency, stock status, sale dates, and variant-level availability.
A stale price or unavailable recommendation damages trust at the moment of decision.
Images and attributes
Useful product images plus material, size, color, compatibility, dimensions, and other category-specific attributes.
Gives comparison systems enough detail to distinguish similar products.
Policies and fulfillment
Shipping, returns, pickup, delivery area, and merchant support information where relevant.
Makes local and constraint-based recommendations more useful and less speculative.
Proof outside the feed
Product pages, reviews, editorial coverage, and other sources that corroborate important claims.
Feed data helps discovery; independent evidence helps an AI answer explain why a product fits.
Readiness audit
Six checks before you blame the algorithm
Most product visibility failures are boring and fixable: stale inventory, variant confusion, blocked pages, missing attributes, or claims that cannot be verified. Start with the source system and work outward.
01
Catalog truth
Can the same price, inventory status, title, and variant be reconciled across the feed, product page, structured data, and merchant systems?
02
Intent coverage
Does the catalog answer the questions buyers actually ask, including “best for”, “under”, “near”, “compatible with”, and “alternative to”?
03
Merchant identity
Is it obvious who sells the item, who fulfills it, how to contact support, and which policies apply?
04
Crawlability
Can relevant product pages be fetched and rendered without robots, authentication, JavaScript, or performance blockers?
05
Evidence quality
Do reviews, comparisons, expert pages, and first-party facts support the claims an AI assistant might make?
06
Change monitoring
Will the team know when an answer shows an outdated price, wrong availability, missing product, or inaccurate comparison?
Prompt coverage
Test the questions shoppers actually ask
A feed can be technically valid and still miss the language of demand. Build a test set that covers discovery, constraints, compatibility, local availability, comparisons, and alternatives.
Category discovery
“What are the best [category] products for [audience] with [constraint]?”
Budget
“Which [product type] under [price] is best for [use case], and what are the trade-offs?”
Compatibility
“What should I buy if I need [feature] to work with [system or product]?”
Local intent
“Where can I find [product] near [city] with [availability or delivery requirement]?”
Comparison
“Compare [product A] and [product B] for [specific buyer profile].”
Alternative
“What are reliable alternatives to [known product] if I care about [constraint]?”
Official references
Keep the protocol and the promise separate
The Agentic Commerce Protocol is an infrastructure layer. It does not replace product truth, merchant support, or a useful shopping experience. Read the primary documentation for current eligibility and integration details; product feeds and commerce capabilities evolve quickly.
Monitor the outcome
Track which products appear, which attributes are wrong, which competitors are recommended, and which sources support the answer.
Explore ecommerce intelligence →FAQ
Questions merchants ask about AI shopping
What is a ChatGPT product feed?
It is a structured catalog of merchant and product information that can help ChatGPT understand what a business sells, including product identity, descriptions, attributes, price, availability, and links.
What is the Agentic Commerce Protocol?
The Agentic Commerce Protocol, or ACP, is an open protocol designed to let AI agents, people, and businesses work together during commerce flows. Product discovery can use merchant feeds, while approved purchase actions can connect to merchant systems.
Does a product feed guarantee that ChatGPT will recommend a product?
No. Product data improves completeness and freshness, but recommendations still depend on the shopper’s intent, relevance, availability, quality, evidence, and the system’s own selection process.
Is ChatGPT shopping the same as Google Shopping?
No. They are different discovery environments with different interfaces, data flows, and ranking behavior. A merchant should keep product data accurate across its own site, feeds, merchant systems, and major search surfaces.