Why we don't fully trust AI when we shop
People trust AI to help them discover things, not to buy them. I built my own shopping assistant design on that distinction.
Before I started designing my own shopping assistant, I read every piece of research I could find on shopping with AI. The more I read, the more the picture split in two.
On one side, people use AI a lot when they shop. In BCG's research, more than 60 percent of participants say they trust the results AI gives them.1 But when it comes to actually buying, the same people pull back.
Discovery yes, decisions no
In Gartner's survey from early 2026, at most 11 percent say they'd leave the purchase decision to AI. Thirty-one percent would let it narrow down the options.2 "Show me a few options" is fine. "Buy it for me" isn't.
Cover Genius's survey across seven countries shows the same picture. Among people open to AI help, the biggest group says "suggest, then wait for my approval." Five percent are willing to let it act completely on its own.3
A wrong answer gets billed to the brand
For brands this has a heavy side, because they're the ones who pay for the mistake. According to Rithum's survey in the US and UK, when a language model gives wrong information, 58 percent of shoppers lose trust in the brand. Only 5 percent check the information on the brand's own site.4
So if AI quotes the wrong price, the culprit isn't the AI. It's the brand. In Gartner's survey, 54 percent of participants say they've had to double-check what AI told them, and 62 percent say that ate up their time.2
Nobody notices the on-site assistant
Then there's visibility. In Nielsen Norman Group's study of Rufus, Amazon's assistant, users simply didn't notice it. It sits in an unusual spot, and a made-up name plus a sparkle icon don't tell you what it is.5 In another of their studies, on on-site chatbots, users said search and filters were faster.6
On most sites the assistant is a window that pops open in a corner. It doesn't tell you what it's for, and once you open it, it's slower than the filters.
The design rules that came out of the research
The rules of the assistant I designed came largely out of this research:
- Information like price, stock and size comes only from the brand's own data. The AI doesn't invent a number; if it doesn't know, it says so.
- It suggests and stops. Adding to the cart and paying are the user's decisions.
- Every suggestion comes with a line saying why it was suggested.
- Its name says what it is, and it doesn't hide in a corner.
- It doesn't remove the filters or take their place. Anyone can go back to the classic site at any moment.
The way I see it, the real obstacle to shopping with AI is trust.