Commerce on ChatGPT is very different from most other use cases. ChatGPT is not using its massive AI model to answer shopping questions. Instead, it outsources its thinking to the web: it runs a few web searches, combines those results and picks out products from them. 53% of commerce prompts trigger web search, compared to <10% for most other tasks.
When you ask ChatGPT, "Which NBA player has scored the most points?" it can answer that from its learned intelligence. But when you ask "Which running shoes are best for trail running?" it typically searches the web for "best trail running shoes 2026." You can see that in the raw conversation history, which includes its decision probability whether the prompt needs web search.
ChatGPT is not deciding which running shoes are best. It doesn't even try to. It relies on external sources instead and gets to the answer by summarizing their results. This is super important to wrap your head around. In fact, it reminded me of the "Thinking, Fast and Slow" book by Daniel Kahneman. Just like human brains, there are questions AI systems answer quickly from stored intelligence and slowly by requesting additional and up-to-date data from the web.
For shopping specifically, ChatGPT is going to end up building a product catalog so it could build actual commerce intelligence instead of relying on trusted-yet-subjective external sources. In the meantime, the task for everyone else is thinking about optimizing for AI through optimizing for web visibility. (Keep in mind that ChatGPT will most likely end up on "The 12 Best Running Shoes of 2025" on Runner's World rather than go to Nike, Adidas, and ASICS websites individually.)