ChatGPT Shopping is pulling far more product recommendations from connected feeds, giving eCommerce teams a technical problem that sits underneath the storefront.
Feed-connected sources accounted for 61.54% of ChatGPT Shopping recommendations around July 10, up from 8.26% earlier in the month, according to Search Engine Journal, citing Profound data.
The analysis covered more than 1.75 million tracked prompt runs in July, and Profound says the figures describe its own tracked prompts, not all ChatGPT shopping activity.
The data does not show how ChatGPT ranks products or establish that feed retrieval caused every change in visibility.
But the rise in feed-sourced recommendations points to a larger role for product feeds in AI shopping, putting more pressure on eCommerce teams to keep catalog data accurate and current.
For Isadora Marlow-Morgan, president of Isadora Agency, the issue reaches into the website’s underlying architecture.
“When product data lives in several places, every new shopping channel creates another synchronization problem,” Marlow-Morgan says.
“The website, structured data, merchant feeds, and AI systems need the same product facts, or inconsistencies will surface exactly where shoppers are making decisions.”
Why are product feeds becoming important for ChatGPT Shopping?
The answer starts with how buyers are already using AI.
Thirty-eight percent of consumers in France, Germany, and the United Kingdom reported using AI tools to find or decide on products and services, according to McKinsey’s December 2025 consumer survey.
That puts AI inside the shopping journey before a transaction ever reaches a retailer’s checkout.
Search Engine Journal’s report suggests the technical side is moving with that behavior.
Profound’s sample showed feed-connected retrieval rising from a small share of tracked recommendations to a majority after July 10.
By Sept. 3, the report said feed retrieval represented about 65% of the product recommendations it tracked.
The jump also came with a tighter merchant pool.
Search Engine Journal reported that references to the top 10 merchants rose from 22.5% to 41.8%, while unique merchants referenced fell from 13,524 to 10,607.
That makes product-data quality more consequential: a clean catalog gives AI systems a clearer set of fields to read, compare, and return.
“The job of an eCommerce website is expanding,” Marlow-Morgan says.
“Your catalog has to serve people browsing pages and systems reading product attributes, prices, availability, variants, and policies. That requires product information to be structured before it reaches the channel.”
McKinsey puts a much larger number behind the stakes.
Global B2C retail could see $3 trillion to $5 trillion in orchestrated revenue flowing through agentic commerce models by 2030, according to a June 2026 McKinsey article.
The projection covers goods and reflects scenarios for commerce mediated by AI agents.
Most retailers don’t need to replace their platform tomorrow. But their product catalog now has to keep up with more places reading the same data, from the storefront and structured data to merchant feeds and AI shopping systems.
What an AI-ready eCommerce product data system needs
Start with one authoritative product record.
That record should hold attributes, descriptions, identifiers, images, prices, inventory status, shipping information, availability, and variant relationships in a format that downstream systems can consume consistently.
Variant management is an easy place for this model to crack. A shirt with six sizes and four colors can produce dozens of purchasable combinations.
If the storefront says one size is available while a feed reports it as out of stock, an AI recommendation can become wrong before a shopper ever reaches the site.
Pricing has the same problem.
Promotional prices, regional pricing, sale windows, and inventory changes need reliable synchronization across the commerce platform, structured data, and connected feeds.
Otherwise, the machine may recommend an item at a price the merchant can no longer honor.
“A product feed should be treated as a live extension of the catalog. When prices, availability, or variants change, those changes have to travel through the system cleanly.
“A delayed feed can make an otherwise accurate storefront look unreliable to an AI shopper,” Marlow-Morgan says.
Validation needs to happen before data leaves the source system.
Teams can check required fields, identifier consistency, variant relationships, image availability, price ranges, and inventory status, then flag failures before those records reach an AI channel.
That discipline also addresses a trust problem.
Fifty-four percent of U.S. consumers who used AI while shopping for a recent purchase said they had to double-check the accuracy of all information GenAI tools provided, according to a Gartner survey of 846 consumers conducted in November and December 2025.
“A missing attribute or stale price can change the recommendation itself. Validation therefore has to be part of the commerce system, with clear ownership for fixing errors when they appear,” Marlow-Morgan says.
The architecture also needs reliable connections between systems.
Product information may originate in a PIM, ERP, eCommerce platform, or other catalog system, while inventory and pricing can change elsewhere.
Isadora Agency saw a version of this challenge with Popcornopolis, connecting its eCommerce platform to the brand’s ERP and unit-level inventory tracking so product availability could be updated in real time.
The work also reorganized inventory around individual SKUs, giving the team more precise product and sales data.
APIs, feed generation, structured data, and monitoring have to keep those systems aligned without creating separate versions of the truth.
The final requirement is testing.
Engineering teams should validate what a connected feed actually sends, how variants are represented, how updates propagate, and what happens when a record is incomplete.
OpenAI calls for structured product feeds, required fields, validation rules, and ongoing updates, with file and API delivery paths available to approved partners.
“AI shopping is exposing the parts of eCommerce architecture that were easy to ignore when the website was the main destination,” Marlow-Morgan says.
“Brands need product data that stays accurate across every place it appears, because machines can only recommend what they can reliably read.”
That puts catalog owners and developers on the same operational path: keeping every product field current wherever it is consumed.
