Exclusive: ChatGPT’s new commerce product chief opens up about the hard lessons of building shopping in AI

Just four months ago, OpenAI hosted a small group of journalists, including yours truly, for a press briefing on the rooftop of a swanky boutique hotel in downtown Manhattan.

The occasion was the launch of a dedicated AI model for shopping queries on ChatGPT, called Shopping Research. I wrote about the launch here.

One detail that has stuck with me since: a giant, untouched plate of bacon sitting in the middle of the table during the presentation. The angle of my photo below does not actually do the portion size justice. It was a sight to behold.

Jokes aside, the OpenAI staff members who presented the model’s use cases that day were sharp, as were those who demoed the new Shopping Research experience afterward.

But beyond the pitch, the product reality of the model became problematic. In many cases, ChatGPT would take several minutes to respond to users’ shopping queries with final recommendations when powered by this shopping model. That kind of latency might be acceptable for deep research, but is a much tougher sell for everyday shopping decisions. I should have been more skeptical about that gap at the time.

But it seems OpenAI leaders now have come to a similar conclusion. In an interview with The Aisle yesterday, OpenAI’s new product lead for commerce initiatives, Neel Ajjarapu, admitted that the Shopping Research model that launched that November day is not being prioritized as a standalone feature any longer.

“[M]ost users aren't willing to wait more than a couple of seconds,” Ajjarapu told me, “and so core ChatGPT ends up being the best experience for them.”

Ajjarapu and I also discussed the new shopping experience ChatGPT rolled out this week; how the company thinks about the risk of losing trust with would-be shoppers as it flip-flops on shopping features; and why it pulled back from its Instant Checkout rollout only six months after launching it.