ChatGPT product cards vs offer cards: what should sellers audit first?

By Nexscope Team · Published and reviewed October 13, 2026

ChatGPT product cards help shoppers choose a product; offer cards help them choose a seller for that product. For a merchant, these are two different checks. A product can be easy to find but have the wrong image or variant, or it can look convincing in a product card while a competing offer wins on price, availability, or delivery. Start by checking one SKU across both steps, then compare what the shopper sees with the merchant’s current catalog and landing page.

Key takeaway: A ReFiBuy-commissioned study published October 6, 2026 found that 84% of final product choices in its observed ChatGPT tasks came from product cards. That is a selection result from a limited sample, not a claim that 84% of shoppers purchased through cards or that card position causes sales. The practical response is to verify product and offer accuracy, not to promise a ranking.

What did the 2026 product-card study actually measure?

It observed shopping choices, not transactions. ReFiBuy commissioned Clickstream Solutions to record 40 US participants completing 224 desktop shopping tasks across six product categories in ChatGPT and Google AI Mode. The headline 84% finding uses 104 ChatGPT tasks in watches, consumer electronics, and bags in which participants were not instructed to choose a card. The broader study also reported that participants often engaged with cards before the written answer, and that first-position cards attracted a disproportionate share of selections. The researchers’ methods and limitations matter as much as the headline number.

These sessions do not establish how every shopper, category, device, or future version of ChatGPT will behave. Card position was observed rather than randomized, so a first-card selection cannot be attributed to position alone; product relevance, price, image, and other factors may differ. The research was commissioned by a company that sells AI-commerce optimization services. It is useful directional evidence, not a universal conversion benchmark.

OpenAI describes ChatGPT shopping as a visual comparison experience with product details, prices, reviews, and features. Its March 2026 update focuses on discovery and supports merchant-owned checkout paths. That is why a seller should keep being selected, receiving a visit, and making a sale as separate outcomes.

What is the difference between a product card and an offer card?

The product card answers “Which item fits?”; the offer card answers “Where can I buy it on acceptable terms?” A shopper can select the right model and still reject an inaccurate or unattractive seller offer. In the study, 76% of observed offer choices went to the first offer card, but the same non-causal and sample-size cautions apply to this finding.

This is a diagnostic framework, not a description of a fixed ChatGPT interface. Cards, fields, and eligible sellers can vary by query, market, account, and platform update. Do not turn one screenshot into a permanent ranking report.

How can a seller audit one SKU in ChatGPT shopping?

Use a short, repeatable set of buyer questions and preserve the evidence. For an illustrative 24-ounce insulated bottle, a buyer might ask for “a leak-resistant bottle under $35 that fits a car cup holder.” Before testing, identify the exact SKU, variant, current price, and verified dimensions; the example is not a claim about a real product or a live Nexscope test.

1. Write three constrained questions. Vary the buyer need, not just the wording of your brand name: capacity and price; cup-holder fit and leak resistance; cleaning or replacement-part requirements. Record the exact prompt, market, date, and account context.
2. Capture the product-selection step. Record which products appear, their order in that run, displayed image, title, and specifications. Note whether the exact SKU appears or whether a similar but incompatible variant takes its place. Repeat the questions later because one AI response is not a stable rank.
3. Capture the seller-offer step separately. If offers are shown, note the seller, price, currency, stock, shipping claim, link, and whether the selected variant remains the same. An absent offer is a finding to investigate, not proof of a technical error.
4. Reconcile against sources of truth. Compare the card and offer with the merchant’s current catalog or approved feed and the public selling page. Correct factual inconsistencies at their source; do not invent specifications or reviews to fill gaps.
5. Change one thing and retest. Save the before-and-after evidence and state the time gap. A later card change may coincide with your edit without being caused by it. Track visits and orders in the merchant’s own analytics rather than inferring revenue from a screenshot.

Use a compact worksheet so a team can repeat the check:

Which product-data fixes are worth making first?

Fix discrepancies a buyer can verify before adding more marketing copy. OpenAI’s current product-feed specification calls for stable item or variant IDs, a factual title and description, a public product URL and image URL, brand, seller name, availability, and price for its OpenAI-format discovery feed. It also documents variant grouping and optional shipping details. Sellers should follow the specification for the integration they actually use; these fields are not a formula for first-card placement.

Prioritize in this order:

1. Identity and variant: the image, model, size, color, and purchasable URL refer to the same item.
2. Current offer: price, currency, availability, and any sale or shipping claim agree across the feed, page, and checkout.
3. Decision-making facts: publish measured dimensions, materials, compatibility, limitations, and relevant policies in clear language on the public page.
4. Evidence: substantiate claims such as “leakproof,” “waterproof,” or “fits most cup holders” with the applicable test conditions or exact measurements.

OpenAI says Shopify product data is integrated through Shopify Catalog, while direct product-feed onboarding is currently for approved partners. A supported catalog route makes data available for discovery; it does not guarantee a card, a position, a click, or a sale. For onboarding and catalog fundamentals, read our broader ChatGPT product-discovery merchant guide. This article focuses on the narrower product-card-to-offer audit.

Where can Nexscope help, and where can it not?

Nexscope can help identify a starting point for investigation; it does not control ChatGPT’s shopping interface. Its AI Product Visibility Tool reports sampled product mentions, recommendation positions, citations, and competing products across supported AI shopping answers. Use that report to decide which buyer questions and competitors to inspect manually. It is not a live product-card or offer-card eligibility test, a complete ranking tracker, or a sales attribution report.

For one public product page, the Nexscope Website SEO Auditor can inspect available metadata, readable content, indexing signals, sampled action links, and optional mobile Lighthouse evidence. It does not validate a private catalog feed, check live ChatGPT cards, or verify the final checkout price. The merchant should manually compare the page with its catalog and order system.

Frequently asked questions

Did 84% of ChatGPT shoppers buy from product cards?

No. The study reported product choices, not completed purchases. Its headline percentage came from a subset of 104 observed ChatGPT tasks, not from all ChatGPT users or all 224 tasks across both platforms.

Does the first product card get a guaranteed sales advantage?

No. The study observed more selections of first-position cards, but it did not randomize position or measure purchases. The observation does not establish a ranking factor or a causal sales lift.

Is a product card the same as a seller offer?

No. The first helps a shopper compare products; the second presents a place and terms for buying one. Product identity and merchant-offer accuracy should be checked separately.

Will a product feed or a Nexscope report put my SKU into a card?

No. OpenAI determines its own discovery results and controls merchant-feed eligibility. Nexscope’s sampled visibility report can reveal questions worth testing, but it neither submits a product feed to OpenAI nor guarantees card display.

Start with one verifiable mismatch

Choose one commercially important SKU, test three buyer questions, and record the product and offer evidence shown in each session. Fix the first verified mismatch—wrong variant, stale price, missing dimension, or unavailable offer—at the merchant-controlled source. Then retest and measure the later visit and order stages separately. That is a more defensible response to the product-card research than claiming a new “AI ranking trick.”

First check: sample whether AI shopping answers mention your product. If you need a purchase-stage framework after the product and offer checks, see our agentic checkout seller checklist.

Sources and scope

• ReFiBuy and Clickstream Solutions: product-card study announcement and methods, October 6, 2026. Vendor-commissioned observational research; measured selections, not completed purchases.
• ReFiBuy: study summary, including the 40-participant, 224-task scope and product-card versus offer-card findings.
• OpenAI: Powering Product Discovery in ChatGPT, March 24, 2026. Platform description of visual comparison and merchant-owned checkout direction.
• OpenAI: product-feed field specification and integration onboarding guide. Confirm current requirements for the seller’s actual catalog path.

Scope note: Nexscope did not participate in or independently replicate the ReFiBuy/Clickstream study. The bottle example and audit worksheet are illustrative; no live product-card test or measured conversion uplift is claimed here.

Last reviewed: 2026-10-13 · Maintained by the official Nexscope team · Editorial & evidence policy.