Product data for AI shopping: a practical readiness review

AI shopping creates a new place for customers to compare products. For a commerce team, the immediate question is practical: can an external experience understand what you sell, distinguish the right variant and present an offer that still holds when the shopper arrives? Start with those conditions before commissioning another integration. A connection can distribute a catalog; it cannot make missing commercial facts reliable.

What changed, and what it means for a merchant

On March 24, 2026, OpenAI announced an expansion of the Agentic Commerce Protocol to support product discovery in ChatGPT. The announcement describes product feeds, promotions and different delivery paths, while emphasizing discovery and merchants' own checkout experiences. Treat it as a dated account of that product direction, not proof that a particular store or country is eligible today. OpenAI announcement.

Our editorial interpretation is that catalog readiness deserves a place beside search and conversion work. Discovery, recommendation and checkout are separate steps. A product appearing in one experience does not demonstrate that its stock is current, that every variant is represented or that the visit produces a sale. Ask each platform which integration applies to your account and geography before assigning implementation work.

Review five kinds of product truth

We propose a five-part review: identity, suitability, offer, fulfillment and continuity. It is an operating framework for a merchant team, not an official platform certification. Choose a representative product family and inspect the facts a buyer needs to select and purchase one item. Include a popular product, a variant-heavy product and a product with a meaningful restriction.

Identity covers the exact item and variant. Suitability explains what it fits or does, using supported information. Offer covers price, currency and applicable conditions. Fulfillment covers where and when the order can be supplied. Continuity asks whether these facts agree across the product page, source catalog, distributed feed and checkout. Mark each fact as verified, conflicting or unavailable; do not convert unknowns into reassuring copy.

Identity and suitability: make the comparison possible

An attractive title may still conceal an important difference. Imagine two replacement filters with similar photographs but different compatible devices. If compatibility exists only in an image or an internal spreadsheet, a shopper has little basis for choosing confidently. Put the approved compatibility information where the shopper can read it and where the relevant catalog fields can carry it. Never infer a fit from visual similarity.

The same method applies to package size in CPG, dimensions in furniture and vehicle fitment in automotive. Keep parent products and purchasable variants distinguishable. Assign an owner to resolve missing identifiers or disputed attributes. Where the manufacturer has not documented a claim, the useful action is to obtain evidence or state the limitation. Generating a longer description does not resolve an unknown specification.

Offer and fulfillment: reconcile the moving facts

Price and availability can change faster than descriptive copy. Record which system is authoritative for each field, how updates reach each destination and what happens after a failed update. Compare the displayed currency, variant price and stock status against the actual offer. Review promotion start and end conditions, particularly when a discount applies only to a segment, quantity or location.

A delivery statement also needs context. A general promise on a category page may not apply to a particular postcode or item. For an illustrative inspection, follow one eligible and one ineligible delivery scenario from discovery to checkout. Document where the shopper learns about the restriction. The purpose is to find inconsistent promises, not to assume that every inconsistency has caused lost revenue.

Structured data is one evidence layer

Google documents product structured data and Merchant Center feeds as ways to provide product information. Its guidance also covers variants and merchant policies. Eligibility for richer presentation is different from guaranteed display. Google's product guidance.

Inspect the information itself as well as the syntax. A valid record can contain an outdated price or a description that omits a decisive constraint. Have the technical owner check supported fields and validation results, and have the commercial owner verify the represented offer. Keep a record of the inspected URL, variant, timestamp and destination so that a later reviewer can reproduce the finding.

Run a bounded readiness sprint

Begin with one product family, one target market and one discovery channel. Choose a small sample that includes normal cases and known exceptions; its findings apply to that sample. Capture the current state before editing, then assign corrections to catalog, commerce, content or engineering owners. Do not ask one writer to resolve stock systems, product policy and data delivery alone.

Copy this audit record into your working sheet: one row per SKU and variant, with columns for field, authoritative source, page value, feed value, checkout value, timestamp, mismatch and owner. Add a verification date when the correction is confirmed.

After corrections, repeat the same scenarios and record unresolved dependencies. A successful review ends with verified facts and named next actions, including cases that remain unavailable. Expanding the sample is a separate decision based on catalog diversity and operational capacity, not an automatic claim that the full store is ready.

Can each sampled offer be tied to the correct purchasable variant?
Are the attributes required to choose it supported and readable?
Do price, currency, availability and promotion conditions agree?
Are delivery and return conditions clear for the target market?
Can an owner explain when the distributed record was last refreshed?
Does the destination preserve the product and offer the shopper selected?

Measure readiness and commercial results separately

Track missing attributes, conflicting offers, failed updates and resolution time as operational measures. Track identifiable referral visits, product engagement and orders as separate commercial measures, using consistent definitions and privacy-appropriate instrumentation. A better catalog score does not establish incremental revenue. Seasonality, pricing and campaign activity can change sales during the same period.

For an ARS discussion with Nextriad, bring one product family, its source systems and the inconsistencies your team has documented. That creates a concrete starting point for assessing commerce and search work. The valuable output is a scoped improvement plan with evidence owners and verification steps. No checklist or integration can promise recommendation by an AI platform; the merchant can control the clarity and reliability of the facts it supplies.