Tripadvisor AI summaries downplay serious complaints, Which? finds
TL;DR:
- A Which? investigation found Tripadvisor’s AI-generated review summaries softened serious complaints about hotels.
- One hotel being sued over alleged mass food poisoning was described as “spotless”; a resort facing harassment complaints was called “friendly.”
- Tripadvisor says the tool works as intended and that summaries do not replace the full reviews.
Generative AI’s habit of smoothing over sharp criticism has surfaced in a consumer-facing setting. An investigation by the UK campaign group Which? found that Tripadvisor’s AI summaries of hotel reviews repeatedly downplayed serious complaints — from food-safety failures to sexual harassment — that appeared plainly in the underlying reviews.
When the summary sands off the edges
In one case, a Cape Verde hotel being sued in the High Court by hundreds of guests over alleged illness was summarised as popular and “spotless,” despite reviews describing raw chicken, flies on the buffet and a family left seriously ill. A Turkish hotel where guests reported harassment by male staff was described as offering “friendly” service, with problems reduced to “lapses noted by a few.”
Tripadvisor said it was “monitoring and refining” the tool and looking into cases where summaries did not match the property, but insisted the features were “delivering exactly what they were designed to do,” and that systems suppress summaries when travellers flag serious safety incidents. Which? Travel editor Rory Boland urged users to scroll past the summaries to guest reviews, “particularly one-star ratings.”
The finding chimes with a known failure mode. Duncan Brumby, a professor of human-computer interaction at University College London, said AI tends to “sanitise and rub off the edges” of criticism because its training data skews bland. It is the consumer-trust flip side of reliability worries Resultsense has tracked elsewhere, from frequent chatbot users being likelier to believe vaccine myths to UK shoppers ready to abandon an AI agent after a single mistake. The same investigation notes Google pulled some AI health summaries this year after being found to surface misleading information.
Looking forward
With summarisation now sitting between consumers and the reviews they rely on, the episode sharpens a question for any business bolting AI onto user-generated content: who is accountable when the summary is materially wrong? For UK firms, the reputational and legal exposure of a “polite” AI that buries safety warnings may prove a stronger check than any voluntary guideline.