Tripadvisor rolled out AI-generated summaries for hotel reviews earlier this year. The feature promised to distill millions of guest opinions into quick, digestible overviews. But a consumer investigation has exposed a troubling flaw. Those summaries often paint glowing pictures of properties plagued by serious safety problems.
Which? tested the new tool. It found the AI downplaying or outright ignoring reports of food poisoning, sexual harassment and dire hygiene failures. Euronews reported the findings. The results raise fresh questions about reliance on artificial intelligence for travel decisions. Holidaymakers could book stays at risky places without realizing the full picture.
Consider the Riu Palace Santa Maria, a five-star all-inclusive resort in Cape Verde. Tripadvisor’s AI described it as popular with travelers. It highlighted spacious rooms, diverse restaurants earning rave reviews and spotless cleanliness. Reality told a different story. Guests reported raw chicken served at buffets. Flies and birds landed on food. One reviewer spotted dead roasted mice near seating areas. “This place will destroy holidays, and has the potential to take lives,” wrote one guest whose family fell ill.
Numbers back the horror. When Which? checked in March, 102 mentions of food poisoning appeared in reviews. Between December 2025 and April 2026, 32 one- and two-star reviews surfaced. Fourteen described serious illness. Some guests ended up hospitalized. Others cut trips short. One died this year. The resort now faces a group legal action from at least 412 holidaymakers. Seven deaths linked to the chain have been reported since 2023.
Tripadvisor’s AI chatbot, named Ollie, performed no better. Asked directly about food poisoning risk at the property, it called the chance quite unlikely. It claimed the resort held a strong reputation for high hygiene standards. The summary has since been removed. Yet the case shows how the system can sanitize harsh feedback.
Other examples followed a similar pattern. At Garza Blanca Resort in Cancun, multiple guests including an entire wedding party reported falling ill over the past year. The AI summary instead praised immaculate cleanliness and positive feedback on dining options. In the Dominican Republic, the Occidental Caribe drew scathing reviews. Guests described sewage smells in rooms. Half a 68-person wedding party got sick. The entire hotel reeked of mould at times. Some lacked running water and showered from bottles. Tripadvisor’s AI mentioned abundant amenities. It offered only vague references to inconsistent cleanliness and maintenance issues.
Sexual harassment complaints met the same softening treatment. Guests at Kaia Coracesium on Turkey’s Antalya coast described feeling unsafe. Male staff made inappropriate jokes and gestures. They repeatedly asked for social media connections. In two cases, workers followed young daughters or followed a guest up stairs toward her room. The AI summary called the service friendly. It noted lapses in service by just a few guests. The language minimized what reviewers called repeated and alarming behavior.
Rory Boland, editor of Which? Travel, didn’t hold back. “The platform has a responsibility to revisit the accuracy of its AI summaries and AI chatbot,” he said. “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.” His warning carries weight. Which? has long scrutinized travel platforms. This time the stakes feel higher.
Tripadvisor pushed back. The company said its systems prioritize transparency and impartiality. Summaries surface both positive and negative feedback. They draw from the previous 12 months of reviews, updated monthly. Reviews receive equal weight regardless of star rating. The AI uses large language models and natural language processing to identify common themes. It turns them into plain-English overviews. “Our AI Summaries have been designed to uphold the integrity and transparency that has made Tripadvisor trusted by millions of travellers for over 25 years,” a spokesperson told investigators.
The company added that it automatically suppresses AI summaries when reviews mention serious safety incidents like deaths, drugging or sexual assault. That content stays highly visible. It confirmed it is looking into the specific examples Which? raised. The chatbot Ollie remains a product in development. Yet critics see a pattern. Generative AI often leans agreeable. It smooths sharp edges. Training data filled with bland observations may explain the politeness.
Duncan Brumby, professor of human-computer interaction at University College London, connected the dots. His research on AI in academic peer review showed similar sanitizing effects. “It’s as if it’s being polite,” he observed. Other studies confirm summarization tools frequently reduce rich consumer feedback to shallower sentiments. The tendency appears baked in.
Comparisons to rivals highlighted the gap. Google’s AI overview for the same Riu Palace warned of potential for illness. It flagged outbreaks and concerns over birds in buffet areas. For Britannia hotels — repeatedly ranked among the UK’s worst chains by Which? — Google noted frequent poor ratings, filthy conditions and horrendous service. Tripadvisor’s version claimed guests often praise clean rooms and described a charming atmosphere. The contrast stands out.
Travel experts echoed the concerns. Martyn Slack of Three Tickets Anywhere noted that many AI overviews favor fluency over accuracy. Most users don’t bother checking underlying reviews. Oli Huggins pointed to how generative models train toward agreeable, cliché outputs with risks of hallucination. Nupur Khurana warned of the danger when systems present information with confidence regardless of facts. These observations align with broader worries about AI in consumer platforms.
The rollout of such features reflects pressure to simplify vast data troves. Tripadvisor handles over a billion reviews and contributions. Travelers want fast answers. But speed comes at a cost when nuance disappears. One-star reviews often hold the most critical details. Scrolling past polished summaries to read them directly remains essential. Cross-checking with sites like Google adds another layer of protection.
Recent coverage has amplified the Which? findings. The Guardian detailed how the AI glossed over sexual harassment allegations while calling a sued hotel spotless. It quoted Boland urging caution. Metro highlighted the potentially life-threatening omissions and expert views on AI agreeability. The Telegraph and The Independent ran similar reports. No major updates from Tripadvisor have emerged since the initial response. The story continues to spark discussion on social platforms.
Industry insiders already know review systems face manipulation. Fake entries persist despite moderation efforts. Now AI introduces a new variable. It doesn’t invent complaints. Instead it filters and frames them. The result can mislead. A summary that sounds balanced may still bury warnings that matter most to safety-conscious travelers.
Boland’s advice holds. Don’t trust the top-line AI summary alone. Read the actual guest accounts. Pay special attention to recent low ratings. Check multiple platforms. The convenience of artificial intelligence shouldn’t replace due diligence. In travel, bad choices carry real consequences. Illness, trauma or worse await those who skip the fine print.
Tripadvisor built its reputation on community feedback. That foundation now faces a test. Can the company refine its AI to capture genuine risks without losing the helpful brevity? Or will users learn to treat these summaries as starting points rather than trusted verdicts? The coming months will tell. For now, caution prevails.
WebProNews is an iEntry Publication