Seller Preparation for ChatGPT Shopping: The Key Lies in Product Data

OpenAI provided a figure when launching Instant Checkout on September 29, 2025: over 700 million people use ChatGPT weekly. The Shopify official blog added two more figures this June: by Q1 2026, AI will drive traffic to Shopify stores up to eight times compared to the same period last year, with orders sourced through AI approaching thirteen times.

Many people interpret preparation as researching a new backend; this article won’t discuss that. When ChatGPT recommends products, the data it reads comes from the Shopify Catalog. Stores that meet the criteria are automatically included, with no application entry or required apps to install. Therefore, the preparation largely revolves around the data itself: whether the title and description are accurate, whether all fields are complete, whether inventory is accurate, and whether reviews are connected.

The Pool is Already Open

First, clarify the position. According to the Shopify official blog, qualified products are automatically included in the Catalog, and merchants have nothing to configure. Changes to prices, inventory, or descriptions in the backend will automatically sync to the connected AI channels. The condition for ChatGPT as a channel is to sell to buyers in the U.S., regardless of where the store is located.

There are thresholds for entering the pool, but the limiting factor is not the richness of data. Among the qualification requirements listed in the Shopify Help Center, besides the product being able to enter the Shopify Catalog, there are two easily overlooked points: you must agree to the additional service terms of Agentic Storefronts; and the policy pages for Terms of Service, Privacy Policy, and Return and Refund Policy in Settings must be fully completed. Stores with empty policy pages are blocked at this step.

To confirm the store’s status, the backend path is Sales channels > Agentic, and you can also disable this channel there if you don’t want it.

Review Product Pages According to Feed Specifications

The OpenAI developer documentation includes a product feed specification, with nine required fields: item_id, title, description, url, brand, seller_name, image_url, availability, and price. Stores using the Shopify Catalog do not need to submit this feed themselves, but it serves as the most direct reference for ChatGPT when reading product data, making it suitable as a checklist for product pages.

Several specific standards can be directly adjusted. It is recommended that titles do not exceed 150 characters, and descriptions should be plain text not exceeding 5000 characters, with variant information such as color and size included in the title. The price format should be written as “79.99 USD” including the currency. For the barcode field, the specification requires that the GTIN be a valid number of 8, 12, 13, or 14 digits, and it specifically states: if there is no GTIN, leave it blank; do not use fabricated numbers. If there is a manufacturer number, it can be filled in with mpn alongside brand. Among these nine fields, the barcode and brand are often overlooked, with many stores having these two fields empty in the backend.

Stores with many variants should check one more thing: each variant should have its own image and link; do not let the black version display the beige image.

Price and Inventory: Synchronization is Not That Fast

There is a noteworthy statement in the OpenAI Help Center: after merchants update prices or shipping fees, the information seen on ChatGPT may be delayed; they are working on faster update methods. Meanwhile, the Shopify blog states that the Catalog verifies prices and inventory in real-time. The two sources do not align, so I interpret conservatively: after changing the price, it is normal for buyers to see the update a few hours later; do not make price changes just an hour before a major promotion.

Inventory also needs to be managed properly. The legal values for availability in the feed specification are in_stock, out_of_stock, pre_order, and backorder; mark out-of-stock items as such. For stores using the Catalog, this value comes automatically; all you need to do is avoid chaos in the backend: do not continue promoting out-of-stock variants with “in stock” images, and remember to handle seasonal items that are taken off the shelves, ensuring they do not linger in the pool with outdated prices.

Review Data: Limited Methods for Verification

ChatGPT’s product results will include review summaries, and the Help Center specifies that these summaries come from reviews on public websites, which OpenAI does not verify. It will also label products with tags like “Budget-friendly” and “Most popular,” based on various third-party data obtained by the model. The reputation you build on review sites and in your store is accessible to the model.

For structured data on-site, the verifiable method is as follows: official themes like Dawn output Product schema using the product | structured_data filter, which by default does not include aggregateRating. Rating data exists in the reviews.rating and reviews.rating_count metafields; the review app must sync this data, and then the theme code must integrate it into JSON-LD. Stores with review apps can check if the aggregateRating field exists in the product page source code; if not, it indicates a break in the chain.

Shopify also offers a free tool at shopify.com/agentic-readiness; by pasting a product page link into it, it will report whether the structured data and robots.txt of that page are friendly for AI crawling.

It is unclear whether ChatGPT uses your site’s schema; the official documentation does not clarify this, but it can be confirmed that it follows the Catalog line. Just ensure the schema is done according to basic SEO principles; do not expect immediate results.

Ranking Weight: No Official Figures Provided

What products rank higher? OpenAI has listed factors such as inventory status, price, quality, whether the product is from the original manufacturer or a major seller, and whether Instant Checkout is enabled. No weight has been provided, and the official documentation does not give specific values; moreover, the Help Center explicitly states that this sorting will continue to evolve towards user preference personalization.

Anyone claiming to have a solid grasp of the ChatGPT ranking algorithm, including service providers selling “AI search optimization” packages, should be listened to with caution; there is currently no reproducible official method.

For example, a standalone site selling camping lights with 120 SKUs needs to prepare by: filling in the missing barcodes for over sixty variants, stopping eight out-of-stock variants, checking whether the review app has synced ratings into the metafields, and finally using agentic-readiness to scan the three best-selling product pages. This can be done in half a day without spending a dime.

Stores selling to U.S. buyers should complete this round of checks now; the pool is already open, and the competition is about who has cleaner data first. For stores targeting other markets, first confirm that the three policy pages are filled out, and address other matters when the channels open.

References:

OpenAI Help Center: Shopping with ChatGPT Search https://help.openai.com/en/articles/11128490-shopping-with-chatgpt-search

OpenAI: Buy it in ChatGPT https://openai.com/index/buy-it-in-chatgpt/

Shopify Blog: Agentic Commerce on Shopify: How It Works https://www.shopify.com/blog/how-agentic-commerce-works

Shopify Help Center: Selling on ChatGPT https://help.shopify.com/en/manual/online-sales-channels/agentic-storefronts/chatgpt

OpenAI Developers: Product feed spec https://developers.openai.com/commerce/specs/file-upload/products/