AI summaries of Tripadvisor hotel reviews downplay serious complaints, investigation finds | Travel & leisure
A hotel sued over mass food poisoning has been described as “clean” and a resort where guests complained about sexual harassment by staff has been praised for its “friendly” service by an artificial intelligence aimed at summarizing millions of Tripadvisor reviews.
Consumer campaign organization Which?
AI-generated reviews appear on the travel website’s hotel web pages to help holidaymakers decide where to book. In one case, the AI brief described the Riu Palace Santa Maria in Cape Verde as popular for its spacious rooms, “a variety of restaurants” that received “rave reviews,” and impeccable cleanliness.
But customers reported being served raw chicken, sharing photos of flies and birds at the buffet, and “little fried mice dying near the seating area.” One guest whose entire family fell ill wrote: “This place will ruin the holidays.”
The hotel chain is being sued in the high court by hundreds of guests alleging illnesses linked to poor hygiene standards and lapses in food safety.
The AI-generated review for the Cape Verde hotel is no longer available. Its operator, RIU Hotels & Resorts, said: “We operate with the highest standards of professionalism and service, keeping hygienic-sanitary safety as our top priority.”
The AI praised a separate hotel in the Dominican Republic for its “abundant” amenities, citing only “inconsistent” cleaning and “maintenance issues.” However, guests reported showering with bottled water because the mains taps ran dry and everyone at a large wedding got sick.
Meanwhile, guests at a hotel in Türkiye wrote on social media that they felt unsafe due to constant sexual harassment from male hotel staff, including requests to connect, but the AI summary described the service as “friendly” and said only “timed out”. [in service] noted by several people”, Which? was found.
Tripadvisor said it monitors and improves its AI tool and is “looking for instances where reviews do not match the intended business.” But “we’re confident these features deliver exactly what they were designed to do: help travelers quickly understand the breadth of feedback while making it easier to fully discover underlying reviews.”
He said AI-generated summaries do not replace travellers’ reviews, and customers have the good sense to check any AI recommendation against the billions of reviews and contributions it collects.
Rory Boland, editor of Which? Travel said: “The platform has a responsibility to revisit the accuracy of the AI summaries and the AI chatbot. In the meantime, users should scroll through these summaries and look at guest reviews, particularly one-star reviews and comments on other sites, to ensure their next stay is safe.”
After the newsletter launch
Tripadvisor said its systems automatically suppress AI summaries when travelers warn of serious safety incidents such as death, drug use or sexual assault, which “helps ensure this content is highly visible to our community.”
Duncan Brumby, professor of human-computer interaction at University College London, said the case was consistent with his own research into academics using artificial intelligence in the peer review process. He found that the AI tended to “scrub and clean up the edges” of harsher criticisms, possibly because much of the training data contained far more boring observations.
“Here you have guests describing a really negative experience, but the AI decided to tone it down,” he said. “As if it were polite.”
Other studies found Technologies designed to summarize ideas often suffer from a blind spot, meaning they reduce the wealth of consumer feedback to more superficial emotions.
This year, Google removed some of its AI health summaries after Guardian research found people were being put at risk of harm by false and misleading information.
