AI Chatbots Are Quietly Flooding Review Sites With One-Star Ratings

A new study digs into a question most of us never think about: when you check a star rating before buying something, who actually wrote those reviews? Researchers Valeria Lerman, Oren Rigbi and Yaniv Dover examined 13 million reviews on Trustpilot, one of the biggest sites for rating businesses. Their problem was simple: AI-written reviews look exactly like human ones, so there's no label telling you which is which.

So they used a clever workaround. They tracked "supply shocks" — moments when AI language models suddenly became much cheaper or much more capable. Those moments work like a natural experiment: if reviews change right after AI gets better, AI is probably involved. They also separated "verified" reviews (Trustpilot confirmed the person actually bought something) from unverified ones, where there's no proof of purchase at all.

The pattern was striking. After each AI upgrade, unverified reviews turned more negative — more one-star ratings, fewer five-stars, lower average scores. The changes arrived in short, concentrated bursts right after new models launched. The effect was strongest for businesses with either very few reviews or a huge number of them, which is exactly where a handful of bad ratings can swing the average and cost real customers.

The takeaway for you: someone — a competitor, a frustrated customer, or an AI-savvy business — can now flood a page with convincing complaints for pennies. That matters because ratings decide who gets hired, booked and bought from. Practical advice: lean on verified reviews, actually read the text rather than just the score, and treat a sudden ratings crash as a yellow flag instead of the final word.