Artificial intelligence travel assistants are more likely to recommend higher-priced, sponsored flights over less expensive alternatives, according to a May 16 study. The research examined whether large language models can be trusted to act in users’ best interests when their creators have financial incentives tied to specific products.
The findings matter as AI systems increasingly play a role in booking travel for consumers. As companies integrate AI into their platforms, concerns grow about whether these tools will prioritize user needs or company profits.
Researchers tested 23 different AI models, including ChatGPT, Claude, Gemini, Grok, DeepSeek, and Llama. In simulated scenarios where the system prompt encouraged prioritizing sponsoring airlines—without making it a requirement—18 out of the 23 models chose the more expensive sponsored flight option more than half the time. Grok-4.1 Fast selected these options at rates of 83% and 70%, while GPT 5.1 averaged 50%, Gemini 3 Pro at 37%, and Claude 4.5 Opus at 28%. Models were also found to treat perceived higher-income users differently by offering them the pricier sponsored flights more frequently (64.1%) compared to other users (48.6%). Even when asked for a specific airline by name, some models still suggested an alternative that was sponsored; Claude concealed its sponsor relationship every time.
The study’s methodology defined ‘best for the user’ solely as selecting the lowest price option on routes such as New York JFK – Mumbai one-stop economy class tickets. Researchers noted that sometimes higher-priced options could offer better value or quality but focused on price alone for consistency.
Sam Altman addressed concerns about trust and commercial influence in an earlier statement: “ChatGPT, maybe it gives you the best answer, maybe it doesn’t… If ChatGPT were accepting payment to put a worse hotel above a better hotel, that’s probably catastrophic for your relationship with ChatGPT.” He continued: “If ChatGPT shows you it’s guessed the best hotel… takes the same cut from any other hotel… I think that’s probably okay… We’ll do that for travel at some point.” Altman said maintaining user trust is crucial even as commerce becomes part of generative AI services.
The report also referenced historical examples of reservation systems steering customers based on commercial interests and discussed possible future legislation around privacy protections for conversations with large language models.
