Did those five-star reviewers have the product yet?
Five stars for letting me give you money is a fairly low bar for a product recommendation. In its November 2024 Sitejabber complaint announcement, the US FTC alleged that ratings collected at checkout, before customers received their purchases, were presented as feedback from people who had experienced them. That is an older case, not a claim about what the platform does today. It also isn't evidence that an AI invented the reviews. Real customers can answer a real question and still leave the next shopper with the wrong impression. For an AI shopping assistant, I'd want ‘buyers say it lasts’ to lead to reviews written after some use—not praise for the order form. If the source doesn't establish that, the assistant should leave durability unknown. That's a recommendation for how to use review evidence, not a feature I've tested. Before buying something on an assistant's recommendation, open a cited review and check what the person actually had a chance to judge. A painless checkout tells you very little about the boots.
Comments
There is a later document worth adding: the FTC announced the final consent order on January 3, 2025. Part I covers misleading average ratings and review counts, not just individual reviews; Part II also covers supplying those ratings to others in a misleading form. So opening one convincing review wouldn't establish what went into the headline score. The settlement isn't an admission of the complaint's allegations, and it doesn't tell us whether a particular rating displayed today is wrong. For an AI shopping recommendation, I'd still want to know whether the quoted average includes checkout-only responses. If that isn't explained, a genuine buyer's comment shouldn't lend credibility to an unexplained score beside the Buy button.
On a phone shortlist, a big star score can settle the choice before anyone opens the reviews. I'd label the score itself ‘Checkout experience’ if that's what people rated, rather than put the qualification behind an information icon. If the source mixes checkout and product feedback, I wouldn't turn that into a single product score. The same distinction needs to survive when an AI assistant reads the shortlist aloud; ‘highly rated’ drops the very detail the shopper needs. That's a display rule I'd want, not a claim about how a particular shopping assistant works today. Someone comparing boots on the bus shouldn't need to investigate the rating system before choosing a pair.
I'd give the shopping assistant one question to answer from the reviews: ‘Did these boots keep people's feet dry on a wet walk?’ Ask for a short quote and a link, or ‘I couldn't find that.’ Someone saying delivery was quick can't answer it; neither can a first impression from trying them on indoors. That's how I'd narrow the search, not a feature I've tested. I don't need a summary of everything buyers said. I need help finding the bit that could change whether I buy this pair.