Tripadvisor’s AI gives glowing reviews to ‘dangerous’ hotels, new investigation finds

AI summaries may look useful, but in some cases, scrolling down to guest reviews tells a different story (Picture: Metro/Getty Images)

Tripadvisor’s new AI review summaries are masking reports of food poisoning, sexual harassment and serious hygiene failures, according to a new investigation by consumer watchdog Which?.

When browsing Tripadvisor, holidaymakers now find an AI summary at the top of the page for the hotel they are looking at.

These summaries may look useful in giving quick and easy information about the hotel, but in some cases, scrolling down to guest reviews tells another story. 

Tripadvisor’s AI summary describes the five-star, all-inclusive, Riu Palace Santa Maria in Cape Verde as ‘popular with many travellers’, with ‘spacious rooms’, ‘diverse restaurants’ that earn ‘rave reviews’ and cleanliness summarised as ‘spotless’. 

However, guest reviews on Tripadvisor paint a very different picture.

Recent guests at the Riu Palace reported ‘exceptionally poor hygiene’, ‘no basic cleaning or hygiene standards’ and food that was ‘awful, bland, unsafe and inedible’.

One guest said she was served raw chicken.

Another shared photographs of flies and birds in the buffet food and another spotted ‘dead little roasted mice by the sitting area’ on her ‘nightmare’ holiday.

One guest whose whole family fell ill wrote: ‘This place will destroy holidays, and [has the] potential to take lives.’

Negative reviews are not concealed, but experts warn that the vast majority are unlikely to click through to find them (Picture: Getty Images)

When Which? checked in March this year, there were 102 mentions of food poisoning at the Riu Palace.

The resort also had 32 one and two-star reviews posted between December 2025 and April 2026 alone, 14 of which say at least one member of the party fell seriously ill with some form of food poisoning.

Many were hospitalised, some flew home early and one guest died this year. 

The hotel is now involved in a group legal action representing at least 412 holidaymakers who say they became ill after staying at the property, with seven deaths reported since 2023. 

Negative reviews are not concealed. They can be read in the review section, but experts warn that the vast majority are unlikely to click through to find them.

Martyn Slack, who runs family travel YouTube channel Three Tickets Anywhere, researches hotels hands-on for a living and hits the ‘same misleading AI overviews’ that travellers do.

The problem, he says, is that most people ‘don’t take that extra step’ to check reviews direct on the hotel’s site.

Martyn adds: ‘For budget-conscious parents especially, an inaccurate overview isn’t just annoying, it can mean an unrefundable non-cancellable booking that turns out wrong for their kid.’

Experts tell Metro that AI overviews prioritise ‘fluency over accuracy’ (Picture: Getty Images)

Another of Tripadvisor’s AI tools, an interactive trip planning bot called Ollie, also failed to warn holidaymakers about poor hygiene.

When asked directly about the risk of contracting food poisoning at the Riu Palace, Ollie said food poisoning was ‘quite unlikely’, and that the resort had a ‘strong reputation for high hygiene standards’.

When the consumer champion asked Tripadvisor about its AI summaries, it said it prioritises ‘transparency and impartiality’.

It added that its summaries ‘surface a range of both positive and negative community feedback associated with listings’. 

Tripadvisor also told Which? that AI chat assistant tool Ollie ‘draws from a selection of reviews based on detail and recency, and matches by language and context’, but it added that it is a ‘product in development’, and that it is now actively looking into several examples the consumer champion provided where reviews did not match. 

The Riu Palace wasn’t the only hotel that Which? found serious reports of food poisoning missing from its Tripadvisor AI summary.

Several guests who stayed at Garza Blanca resort in Cancun in the past 12 months also left reviews saying they fell ill, including a wedding party.

Yet Tripadvisor’s AI overview is once again glowing, describing ‘immaculate cleanliness’, adding that its dining options ‘earn [it] positive feedback’. 

Another example was the Occidental Caribe in the Dominican Republic. Recent reviewers called it a ‘disaster and disturbing’ as recently as March 2026. Another called it the ‘worst place imaginable’. 

One guest said her room smelled of sewage and that half of the 68-person wedding party she’d travelled with fell ill.

Someone who visited in January reported that the whole hotel smelled of mould.

Several mentioned the lack of access to running water – one guest resorted to showering with bottled water.

In contrast, the AI review summary talks about the hotel’s ‘abundant’ amenities, with only a vague nod to ‘inconsistent’ cleanliness and ‘maintenance issues’. 

There are other dangers Tripadvisor’s AI summaries don’t share.

At Kaia Coracesium on the Antalya coast in Turkey, several reviewers who visited last summer wrote they felt unsafe due to repeated sexual harassment from male hotel staff, including inappropriate jokes and gestures, and repeated requests to connect on social media. 

Two different guests reported that a male member of staff followed their daughters to request their social media details.

In one of these cases, the reviewer says a restaurant worker followed her up the stairs to her room.

The Tripadvisor AI review summarises the service as ‘friendly’. The closest it comes to referring to these serious allegations is: ‘Lapses [in service] noted by a few’. 

This suggests that the company’s AI systems are capable of identifying and referencing these allegations, recognising that concerns raised by even a minority of reviewers may warrant inclusion in its summaries.

However, questions remain as to why such references do not appear consistently and, when they do, why the language used appears to minimise their significance. 

Tripadvisor said its summaries ‘use large language models and natural language processing to read recent reviews, identify the most common themes and then turn those themes into short, plain-English overviews’.

It said its goal was to ‘make content from Tripadvisor’s reviews and opinions as easy as possible to digest, but also to capture and highlight the broad spectrum of positive and negative opinion without favouring one sentiment or the other’.

Tripadvisor told Which? that summaries are updated on a monthly basis ‘and are rooted in the previous 12 months of reviews at the time of each update’.

It said reviews are treated equally regardless of rating, highlighting what reviewers mention most often.

For comparison, Which? looked at rival platform’s approach to AI summaries for hotels.

Google manages context in their AI summaries far better.

Google’s overview for the Riu Palace warned of ‘potential for illness’ and flagged ‘outbreaks of illness’ and ‘concerns over birds in the buffet areas’. 

The consumer champion also compared Google’s AI overview to Tripadvisor for Britannia hotels, the worst-rated chain in Which?’s hotel survey for more than 10 years.

For the Britannia International hotel in London, Google accurately shared that the hotel was ‘frequently rated as one of the worst hotel chains in the UK’ and further highlighted guest reviews of ‘filthy’ conditions and ‘horrendous’ service.

Tripadvisor’s own summary of the same hotel said guests ‘often praise the clean rooms’ and described the atmosphere as ‘charming’. 

What the experts say

Oli Huggins, CEO of ExpertEdge and VP of Partnerships at Packt Publishing, is ‘an AI advocate and a generative AI sceptic’.

What he distrusts, he says, is generative AI that is asked to produce fluent prose with nothing underneath it.

And because a hotel overview is exactly that, it fails ‘in three predictable ways’.

‘The first is that these systems are trained to be agreeable, so they lean positive by default,’ Oli tells Metro.

‘Ask for a summary of almost any hotel and you get a warm one, because upbeat, accommodating prose is what the model has been rewarded for producing, not because the place earned it.

‘The second is cliché. A model reaching for the most probable next word lands on “charming,” “convenient location” and “perfect for families” precisely because those are the most worn phrases available, which means the overview reads as reassuringly familiar while telling you almost nothing specific.

‘The third, and the dangerous one, is hallucination.

‘When the model does not know something it does not stop, it fills the gap with a plausible invention delivered in the same confident tone as everything else, so a pool drained two years ago is described as a highlight, and nothing in the prose signals which parts are real.’

Put those together and Oli says you get fluency standing in for accuracy.

Nupur Khurana, a former AI consultant who works with QXP India Travel, a boutique luxury destination management company, agrees.

‘The danger isn’t that [AI summaries] are wrong, outright, it’s that they present with total confidence regardless of accuracy,’ she tells Metro.

‘A summary can carry an outdated fact, a review taken out of context, or a detail invented altogether, and it reads with exactly the same authority as everything else on the page. Most travellers have no way of knowing which parts to question, so they don’t question any of it.’

Others say Tripadvisor’s response is missing the point.

Rory Boland, Editor of Which? Travel said: ‘Tripadvisor may insist users can still fact-check its summaries against real reviews, but this ignores the fact that it made the decision to push these summaries to the very top of the page.

‘This failure to surface critical safety information is unacceptable and potentially life-threatening.

‘The platform has a responsibility to revisit the accuracy of its AI summaries and AI chatbot. In the meantime, users should scroll past these summaries and look at guest reviews, particularly one-star ratings, and at reviews on other sites, to make sure their next stay is a safe one.’

Comments Add as preferred source