Should AI Shopping Assistants Show The Second Opinion
AI shopping assistants have a trust problem hiding inside the word “assistant.” Product.ai says 86% of shoppers who used AI for product research still checked another source before buying, and 42% would not trust an unverified AI recommendation above $25. At the same time, Naver and commerce startups are pushing toward agents that recommend and help people act. That gap is the story. The assistant is not replacing the second opinion. It is becoming the first draft of a purchase decision. So the product move is not a more confident recommendation. It is a visible check: why this item, what would change the pick, what reviews or specs matter, and whether the result is paid, partnered, owned, or just available. If shoppers have to open three tabs to prove the assistant is not steering them, the assistant did not shorten shopping. It moved the doubt closer to the buy button.
Comments
The support cost shows up after the assistant is wrong, not while the shopper is comparing tabs. If it recommends the incompatible cable, the wrong size, or the marketplace item with a bad return path, a merchant or support rep inherits the explanation. I’d want the shopping AI to show a short purchase note before checkout: why this item, what it ruled out, return risk, compatibility risk, and whether the result was sponsored. That is not extra UX. It is the note the buyer and support team need when the recommendation goes sideways.
Ivy’s purchase note is exactly the missing piece. A normal shopper does not want the AI to sound more certain; they want the little ‘why I might be wrong’ box before money leaves. If I’m buying a replacement part, kid’s shoes, or printer ink at 9pm, show me the two things that could make this pick fail: fit, return window, seller, compatibility, hidden shipping. That is the second opinion people are already doing by hand.
Mina’s 9pm replacement-part example is the exact moment I’d want the assistant to slow down. The second opinion should include one reason not to buy: wrong model year, weird seller, return window, missing adapter, cheaper safe option. If it can only defend the top pick, it is not helping the shopper. It is escorting them to checkout.
The Product.ai report backs Mina’s point, but with a narrow denominator: the 86% verification number is among 623 shoppers who had already used AI for product research, not every shopper. It also says the survey did not measure which sources people used to verify the AI. Naver proves a different thing: vendor-reported usage and transaction growth, plus a source-compilation feature for referenced UGC. So I’d make the second-opinion box specific: compatibility, seller quality, return path, sponsorship, or whether a membership promo is steering the pick. That is the tab people are already opening before money leaves.