Most talks about AI NPCs ask: How do we make characters more intelligent? We suggest a more interesting question: When players are the intelligent agents, how can the game keep up? Meaning Machine partnered with researchers at the University of Bristol to independently study how players interacted with AI characters in the conversational detective games Blood Will Out and Dead Meat. The findings revealed something more fundamental: the future of conversational AI isn’t making NPCs more intelligent. It’s building games that recognise the game players are trying to play. Given open-ended conversation, players don’t simply talk to NPCs – they invent ways to play. They develop distinct interrogation strategies, test theories about character psychology, and even create their own rules for how the game works. Faced with total freedom, players fill the gaps and become co-designers of the experience. More surprisingly, players assume the game understands the game they’re trying to play. They expect it to recognise what they’re trying to do and respond accordingly – even when it is doing nothing of the sort. For us, this exposed a missing layer in conversational game design. The challenge isn’t generating better dialogue. It’s recognising the meaningful gameplay players are already creating. We needed to make the confidence trick of AI real. That insight fundamentally changed how Meaning Machine approaches AI-powered gameplay. Rather than focusing on character roleplay, they developed gameplay systems that “listen” to player strategies and give those strategies meaningful consequences before generating dialogue. Ultimately, this talk argues that the future of conversational AI lies not in giving players more freedom, but in giving that freedom meaning in gameplay.