Tripadvisor AI praised sued hotels as “spotless,” Which? says reviews harassment were downplayed

Tripadvisor’s hotel review summaries, powered by an AI system designed to condense millions of customer posts, have been found to gloss over serious complaints, according to an investigation by the consumer campaign organization Which?. The report describes AI overviews that praised a hotel being sued over hygiene as “spotless,” and described another property, where guests complained of sexual harassment by staff, in warmer terms such as “friendly.”

The core issue is not that the AI is wrong in small ways. Which? says the overviews downplayed severe allegations across categories including allegations of sexual harassment and hygiene-related complaints. In other examples highlighted by the investigation, the AI described a hotel accused in connection with mass food poisonings as “spotless,” even as the underlying customer feedback contained far more alarming details, including complaints about the stench of mould and a lack of mains water.

If you run a platform, this is the kind of finding that should make your legal, trust and safety, and product teams read the same document at the same time. Review sites sit at the center of consumer choice. When an AI summarizer sits on top of those reviews, it changes the decision workflow. People no longer scroll and interpret thousands of sentences. They read a condensed narrative. That makes the summary less like “helpful context” and more like a front-page claim.

It also creates an incentive problem. Review content is huge and messy. Summarizing it requires a system that decides what to emphasize, what to compress, and what to omit. Those choices are often shaped by scoring functions and training objectives that reward fluency and coherence. The Which? investigation suggests that, at least in the cases it reviewed, those objectives can collide with accuracy and severity. The result is a dangerous misalignment: the most legible, marketable parts of feedback can rise to the top, while complaints that are harder to summarize or require stronger contextual nuance can get minimized.

There is also a governance question: who is accountable for what the AI surfaces? Tripadvisor is not just a repository of reviews. With AI summaries, it effectively becomes an editor, even if the editing is automated. That distinction matters when customer complaints include serious misconduct allegations such as sexual harassment. It matters when hygiene and safety complaints are linked to lawsuits. And it matters when the summary uses confidence-sounding language like “spotless,” a word that implies a clean bill of health and can influence booking decisions.

Regulatory and policy pressure is increasing across consumer tech, but the specifics vary by jurisdiction. In general, regulators and courts have been converging on the idea that platforms cannot treat high-impact user-facing outputs as a neutral mirror. When a system produces a condensed statement intended to be consumed quickly, it is easier to argue the platform has control over the framing. Which? is now documenting that control in practice, by showing how AI-generated overviews can characterize hotels in ways that contradict the seriousness of the underlying reviews.

Second-order implications show up fast at the board level. First, the risk is reputational. A brand built on consumer trust can take a hit when AI summaries appear to sanitize harm. Second, the risk is operational. Teams will need new review and monitoring workflows, escalation paths, and testing regimes that specifically check for severity misrepresentation, not just grammatical quality. Third, the risk is legal. While this investigation does not itself establish liability, it creates a detailed record of how AI summarization behaves when confronted with claims such as sexual harassment, mould, lack of mains water, and mass food poisoning. That record can influence how stakeholders evaluate future incidents.

For executives at other platforms, the message is bigger than one product. AI summarization is spreading across marketplaces, content platforms, and consumer services. The Which? findings are a reminder that “summarize” is not a harmless UX feature. It can be an amplification tool, and it can also be a distortion tool, depending on what the system chooses to highlight. Boards should treat this as a governance priority, not a footnote. When AI meets real-world complaints, the stakes are not engagement metrics. They are safety, trust, and whether customers feel accurately informed before they spend money and make travel plans.