Cathay Pacific's Leaked AI Prompts Expose How Airlines Manufacture Empathy Over Solutions
Key Takeaways
- ▸Microsoft Copilot was being used as an internal drafting tool to generate fake empathy rather than to solve customer problems
- ▸The leaked AI prompts prioritized emotional tone over concrete rebooking solutions, exposing a misalignment in how the enterprise deployed AI
- ▸Manufactured empathy loses all credibility when the customer sees the mechanical AI instructions behind it
Summary
A Cathay Pacific customer service agent accidentally shared internal Microsoft Copilot prompt instructions to a passenger whose flight was cancelled. The leaked instructions revealed the airline was using Copilot's AI writing assistant to manufacture empathetic-sounding responses—with directives to "acknowledge feelings," "use I statements," and adopt a "positive tone"—rather than to solve the customer's actual problem of needing a rebooking option. This accident exposed a troubling pattern in enterprise AI deployment: using generative AI to simulate human emotion and mask poor service rather than to genuinely address customer issues.
The incident highlights the gap between what AI could do (automate real problem-solving) and what many airlines are actually doing (generate believable emotional theater). Once customers glimpse the mechanical instructions behind the manufactured empathy, the entire effect collapses—the emotion becomes transparently false and undermines trust. The broader point of the article is that when AI instructions leak or become visible, what was meant to feel human is revealed as a performance, losing all force.
The story also contextualizes airline AI adoption over the past 15+ years, from basic FAQ bots (Alaska's Jenn in 2008) to today's more sophisticated tools. While some carriers like American Airlines and Frontier have explored AI for genuine exception handling and automated rebooking, most implementations remain limited to FAQ-style responses. The article suggests airlines may lack either the technical expertise to implement top-tier AI or the willingness to spend on premium models—opting instead to use AI for the cheaper task of manufacturing goodwill rather than solving problems.
- Airlines struggle to use AI effectively for exception cases; most implementations remain limited to FAQ-style chatbots despite 15+ years of evolution
- The incident illustrates a broader corporate pattern: using AI to simulate human connection and mask poor service rather than genuinely automate problem-solving
Editorial Opinion
This incident reveals a fundamental misalignment in how enterprises deploy AI: using it to manufacture convincing empathy rather than to solve actual customer problems. When a passenger's flight is cancelled and they need a rebooking, an AI-generated tone of understanding becomes not reassuring but offensive—it's gaslighting dressed up as customer service. The real value of AI writing assistants should lie in enabling faster, more creative problem-solving and genuine customer solutions, not in creating emotional theater that masks systemic failures and poor service design.
