ChatGPT Shopping runs on two different indexes in the US and Europe
A debug API inspected by Metehan Yeşilyurt's team shows OpenAI's own product index powering US results while Europe still pulls from Google Shopping.
Metehan Yeşilyurt, writing on LinkedIn on 19 October 2026, reported that his team found a ChatGPT debug API returning around 150,000 lines of retrieval data per prompt. The data shows how ChatGPT selects each product shown and the score assigned to it.
In the US, products are drawn from OpenAI's own index, called Labrador, built from product feeds. In Europe, ChatGPT still pulls products from Google Shopping. That split means European visibility in ChatGPT Shopping today is largely a reflection of Google Shopping rank.
Other findings from the debug data
US shops can carry a "verified seller" flag. In one test, a shop held 25 of 32 candidate products and placed none in the results. ChatGPT reads only the first 50 tokens of a product description, roughly 40 words, and ignores the rest. Listing each product variant separately gives more chances at a carousel slot. OpenAI is also running an A/B test labelled "No Amazon".
Yeşilyurt credits the underlying research to Malte Landwehr, who presented it at brightonSEO.
How to read this if you sell into Europe
Strong ChatGPT Shopping performance in Europe right now may mean nothing more than strong Google Shopping performance. If Labrador rolls out to Europe, that correlation breaks, and feed quality into OpenAI's own index becomes the thing to measure.
Before acting, pull your current ChatGPT Shopping appearances for a set of product queries and compare them to your Google Shopping rank for the same queries in the same market. If they move together, you are seeing the Google dependency Yeşilyurt describes. The 50-token description limit is one team's measurement rather than a documented spec. Check it against your own product data before rewriting feeds around it, and treat the variant-splitting observation the same way.