Where Do ChatGPT Product Cards Come From?
A Study of AI Shopping in Korea and the US
Across 410,000 Answers

When you ask an AI search engine to “recommend products,” product cards showing prices and retailers appear alongside the answer. Do they appear just as often for Korean-language queries? And if they do, where do the products come from? From October 19 to November 1, 2026, Anymorph asked the questions from the same measurement projects every day on ChatGPT, Google AI Overviews, and Naver AI Tab, then analyzed 410,088 answers and 544,067 product cards. The study focused on Korean-language queries and compared them with US English-language queries collected in the same way.

Summary of Findings

Research Method

What We Measured

All figures combine multiple measurement projects; results for any specific company are not included. The data consists of AI search answers collected by Anymorph.

Measurement Approach

• Product card: We counted each product shown in an answer with its name, price, and seller as one card. This includes groups of cards, horizontally scrollable product lists, and products within the body of the answer.
• Share of answers with product cards: The number of answers with at least one product card divided by the total number of answers.
• Comparison across the same commerce projects: To reduce differences in industries and question topics, we separately aggregated only the 5 commerce projects that measure Korean and English questions together. Questions in both languages cover the same product categories within each project.
• ChatGPT product source: We recorded the catalog ChatGPT used to find each product on every card. The three sources were Google Shopping, product feeds sent to ChatGPT by sellers, and Shopify. For Google Shopping products, we also recorded the search terms ChatGPT used and the country it searched in.
• ChatGPT model input: We recorded the product information passed to the model that writes the answer for each card (product name, description, rating, number of reviews, and product page).
• Seller type: For each engine and language, we manually classified the 150 sellers with the most cards and grouped the rest as “Other.” We did not include seller names. The share of top sellers was calculated by consolidating different spellings of the same seller name.

The share of Korean answers with product cards was less than half the US share

These results are from 5 commerce projects that measure Korean and English questions together. When the same product categories were asked about in the same way, 81.0% of US English answers included ChatGPT product cards, compared with 28.6% of Korean answers. The gap was even larger for Google AI Overviews: product cards appeared in 72.6% of English answers and 6.5% of Korean answers. Even when cards were included, Korean answers had fewer cards.

The same gap appeared across broader measurement scopes

Across all commerce projects, as well as all projects including software and services, US English answers had product cards more than three times as often. Naver AI Tab was the engine that most often included product cards in response to Korean queries, with products appearing in 53.6% of its answers to commerce project questions.

Naver AI Tab showed 21 products per answer

For each answer with products, Naver AI Tab showed an average of 21.1 products selected from 14.4 sellers (median: 10 products). All 142,316 products linked to Naver Shopping product pages, and there were 6,919 sellers. The top 5 sellers accounted for 8.4%, so no single seller dominated. ChatGPT and Google AI Overviews showed 2–3 products per answer. Ratings appeared on 63.1% of ChatGPT’s Korean product cards, and the median number of reviews for rated products was 1,500.

Product cards in ChatGPT Korea came from Google Shopping

All but two of the product cards in Korean-language ChatGPT responses featured Google Shopping products. Of the products whose search country was recorded, 99.9% were found on Google Shopping in Korea. In US English responses, product feeds sent directly to ChatGPT by retailers accounted for 93.2% of cards, Shopify products for 4.3%, and Google Shopping products for 2.4%. Even though it was the same ChatGPT, the product cards in Korean-language responses came from products listed on Google Shopping.

Korean products were passed to ChatGPT without descriptions

ChatGPT passes product information to the model that writes the response alongside the product cards. More than 99.9% of products from US retailer feeds came with descriptions written by the retailers, and 97.0% also came with ratings and review counts. Fewer than 0.1% of Korean Google Shopping products came with descriptions, while 63.1% came with ratings and review counts. When ChatGPT describes product features in Korean-language responses, it seems to rely more on the web pages referenced by its response than on catalog descriptions.

ChatGPT also searched in English for products asked about in Korean

We classified 13,414 search queries used by ChatGPT for 30,285 Google Shopping products in Korean-language responses. One in three products was found using an English query with no Hangul at all. A review of the sample showed that English queries often listed the product type, ingredients, skin concerns, and intended use. For 24.9% of products found using English queries, the query included "korea" or "korean." Queries averaged 6.2 words. Even when asked in Korean, ChatGPT often searched for products in English.

The top 5 retailers accounted for 88% of US ChatGPT cards, versus 43% in Korea

US ChatGPT cards were concentrated among a small number of large retailers. A single retailer accounted for 44.7% of cards, the top 5 retailers accounted for 88.4%, and category-specialty retailers made up 70.4%. In Korea, the top 5 retailers accounted for 43.2% of ChatGPT cards, which were spread across large general merchandise retailers, specialty retailers, official brand stores, and cross-border shopping sites. Official brand stores accounted for 17.6% in Korea and 6.2% in the US.

The results were the same when we looked at just five commerce projects. The top 5 retailers accounted for 46.4% of ChatGPT cards in Korea and 89.3% in the US, while official brand stores accounted for 17.3% and 6.9%, respectively. In Google AI Overviews, official brand stores accounted for about 20% of cards in both countries, meaning retailers were more broadly distributed in the US than on ChatGPT as well.

How to take action

Korean ChatGPT Shopping starts with Google Shopping product information

Of the 30,943 product cards in Korean-language ChatGPT responses, 30,941 were Google Shopping products. If you want your products to appear in Korean ChatGPT cards, they first need to be on Google Shopping.

• List your products in Google Merchant Center and keep prices, inventory, and images up to date.
• Add product structured data (product name, price, inventory, and rating) to your product pages so search engines can read product information accurately.

Include English text in product names and information

Of the search queries ChatGPT used to find products asked about in Korean, 33.8% were in English, and typically listed the product type, ingredients, and intended use.

• Include key ingredients, functions, and uses in both Korean and English in the product name and product information.
• Include the English generic name for the product category (e.g., sunscreen, cleansing oil) in the product information.

Write product descriptions as text on web pages

Products from Google Shopping in Korea were passed to ChatGPT models without descriptions. To include product features and differences in its answers, the engine needs web pages it can read.

• Write the key description on each product detail page as text, not as an image.
• Publish pages on your own site that directly answer questions, such as product comparisons, how to choose, and how to use.

Keep product information consistent across your official store and sales channels

ChatGPT cards in Korea were spread across a wide range of sellers, and brand official stores accounted for 17.6%—about three times the U.S. share.

• Include your official store as one of your sales channels and list it on Google Shopping as well.
• Keep the name, size, and price formatting for the same product consistent across your official store, online marketplaces, and specialty stores.

Link ratings and reviews to product information

Ratings were displayed on 63.1% of ChatGPT cards in Korea and 79.0% of products in the Naver AI Tab. Ratings and review counts appear together on cards and are also passed to ChatGPT models.

• Link reviews on your own store to product information and structured data.
• If you are targeting the Naver AI Tab, manage product information and reviews on Naver Shopping. Every product in the Naver AI Tab linked to a Naver Shopping product page.

If you sell in the U.S., check product feeds from retailers

In U.S. English answers, 93.2% of ChatGPT cards came from retailer product feeds, and the top 5 retailers accounted for 88.4%.

• Check the names, descriptions, and ratings of products listed with major U.S. retailers. For products from retailer feeds, descriptions and ratings are also passed to ChatGPT models.
• For your own store, check the process for sending product feeds to ChatGPT and whether it is integrated with Shopify.

Limitations

• This is a 14-day observation from October 19–November 1, 2026. Results may change if the way engines display products changes.
• The questions are general questions Anymorph asks for its measurement project, which focuses on commerce. They do not fully represent the distribution of questions across the market as a whole. Differences in industry composition were mitigated by comparing five commerce projects.
• The figures come from answers collected under Anymorph's measurement conditions (with web search enabled). Product cards may appear differently depending on a user's account history and settings.
• The Naver AI Tab sample is smaller than those for other engines, with 10 measurement projects and 1,087 questions.
• We manually classified the top 150 sellers for each engine and language by seller type, and grouped the rest as “other.” Because the Naver AI Tab has many small sellers, we looked only at the number of sellers and the share accounted for by top sellers, rather than classifying seller types.
• Product cards indicate that products appeared in an answer. This study did not determine whether that led to clicks or purchases.
• The relationships between ratings, seller type, and card appearances reflect observed distributions; the causes were not established through experiments.

Conclusion

1. Few product cards appear with Korean-language questions. In the same commerce projects, ChatGPT included product cards in 28.6% of Korean-language answers and 81.0% of English-language answers, while Google AI Overviews did so in 6.5% and 72.6%, respectively. The engine that showed the most products for Korean-language questions was the Naver AI Tab.
2. Products on ChatGPT in Korea come from Google Shopping. In the U.S., retailer product feeds accounted for 93.2% of cards, but in Korea, nearly all cards were for Google Shopping products, which were passed to the model without product descriptions.
3. Sellers are more fragmented in Korea. The top 5 sellers accounted for about half the share they did in the U.S., and brand official stores also made up 17.6% of cards.

To get product cards in AI search in Korea, list accurate product information on Google Shopping and Naver Shopping, and write product features as text on web pages that engines can read. Since products from the same brand can reach cards through different routes in Korea and the U.S., you need to measure and prepare separately for each country.

Appendix: Research Data