How to Prepare Your Product Catalog for ChatGPT Shopping and OpenAI ACP

ChatGPT shopping now spans discovery, comparison, product and offer selection, and merchant conversion. That broader journey changes the readiness question. A merchant is not only connecting a checkout action; it is supplying the facts that determine whether a product becomes a candidate in the first place.

ChatGPT shopping is no longer just checkout

Think about the journey as a sequence:

query → discovery → comparison → product and offer selection → merchant conversion

Each stage introduces a different failure. An incomplete record can be invisible during discovery. An ambiguous variant can survive comparison but select the wrong SKU. A stale offer can look perfect until conversion. Readiness means preserving the same identity and facts across the entire sequence.

How product information can reach ChatGPT

There is no single mandatory merchant architecture. Product information may come from:

• a merchant-provided catalog or product feed;
• a commerce platform or catalog provider;
• ACP feed and commerce capabilities;
• a crawlable product page and structured markup;
• the merchant’s checkout experience; or
• a deeper app or backend integration.

These paths should converge on one governed product record. Creating separate “AI catalog” content causes price drift, duplicate identities, and claims that cannot be reconciled with the product page or system of record.

Product is not variant is not offer

The identity model must distinguish:

Conflating those entities causes parent-level images on the wrong variant, duplicate feed rows, expired promotions, and recommendations that cannot resolve to a purchasable offer.

What the discovery record needs

At minimum, provide stable IDs, brand, title, description, valid product URL, useful images, structured category attributes, variant dimensions, price, availability, and a merchant offer identity. “Minimum” is not “sufficient”: the winning fields are the facts buyers use to constrain a query in that category.

For a replacement component, compatibility and model relationships matter. For furniture, dimensions, materials, finish, assembly, and load capacity may matter. For electrical equipment, rating, poles, breaking capacity, certification, and mounting can decide eligibility.

Discovery data vs transaction data

Neither layer can compensate for the other. A live price does not explain compatibility, and a complete specification does not guarantee inventory.

Run a ChatGPT Catalog Audit

Make freshness an explicit contract

Price and availability need timestamps, update expectations, expiry behavior, and monitoring. Decide how old each value may be before the system refreshes, suppresses, or refuses to use it. Promotions need effective windows; local inventory needs a location; customer prices need authenticated scope.

Do not label a daily batch “real time.” Publish an honest observation time and test the lag from the source system to every discovery surface.

Supply facts that support real comparisons

Titles and prose help interpretation, but structured attributes support constraint matching. For each high-value category:

• Collect the attributes buyers use to include or eliminate a product.
• Normalize units and value types while preserving the source value.
• Model compatibility, replacement, accessory, bundle, and supersession relationships.
• Separate hard constraints from preferences and marketing claims.
• Validate that images and URLs resolve to the selected variant.

Preserve product provenance

When a manufacturer page, supplier PDF, PIM, and ERP disagree, the shopping surface should not receive an arbitrary merge. Record the source, source authority, extraction or update time, validation status, and confidence for each consequential value. Define precedence by field: the ERP may own sellable status while a manufacturer document owns a technical rating.

Provenance supports two outcomes. It lets automated workflows choose safe values, and it gives catalog teams evidence when a recommendation or claim is challenged.

Test realistic failure scenarios

ChatGPT shopping readiness checklist

• Resolve duplicate supplier, platform, and internal records to canonical identities.
• Validate product, variant, SKU, listing, and offer mappings.
• Measure decision-attribute coverage by category.
• Confirm canonical URLs and variant-correct images.
• Publish current price, availability, promotion windows, and markets.
• Trace technical and compatibility claims to authoritative sources.
• Test constrained discovery, comparison, offer selection, and unavailable-item behavior.
• Monitor drift between sources, feeds, pages, and checkout.

Make one catalog ready for every agent channel

Claro resolves duplicate identities, fills and validates missing attributes, preserves provenance, and keeps records current across PIM, ERP, and supplier sources—so trusted product data can serve ACP, UCP, Claude, or the next commerce interface without rebuilding the catalog each time.

Send a representative category or product export. Claro evaluates identity, attribute coverage, conflicting values, freshness, and agent-search readiness, then returns a prioritized remediation plan.

Run an Agentic Commerce Readiness Audit

Related resources

Further reading

• Agentic Commerce Protocol specification
• OpenAI: Agentic commerce

FAQ

Does every merchant need the same ChatGPT shopping integration? +

No. Product information may reach ChatGPT through a merchant feed, platform or catalog provider, web pages, structured data, ACP capabilities, or a combination. The right architecture depends on the merchant, but every path needs consistent product truth.

What is the difference between discovery data and transaction data? +

Discovery data explains what a product is and why it satisfies a request. Transaction data confirms whether an exact offer can be purchased now, including current price, availability, eligibility, fulfillment, and checkout state.

How can a merchant test whether its catalog is ChatGPT-ready? +

Test constrained queries that require specifications, compatibility, variant and offer resolution, then verify that recommendations use the right SKU, current commercial facts, valid URLs, and supportable claims.