Adding conversational AI to a game.

Twelve questions worth asking before you build.

An AI character is more than a dialogue feature. Its conversations can affect gameplay, player relationships, privacy and safety—and its behaviour may change over time. These questions can help developers, designers and producers work through the decisions together.

1. Who will be playing your game?

   Is it aimed at adults, teenagers or younger children—and could minors access it even if they are not the intended audience?

2. What is your—and your company’s—ethical stance on responsible AI use?

   Beyond complying with legislation, do you understand the risks to players and accept responsibility for protecting them? What boundaries will you uphold when player wellbeing conflicts with engagement or revenue—for example, when a character could encourage emotional dependency or use a child’s trust to push purchases?

3. What role will the AI play?

   Is it a guide, a fictional character or an ongoing companion? Can it only talk, or can it also take actions that affect the game and its players?

4. What happens to players’ personal information?

   What will you collect, remember and share with model providers? Who can access it, how long will it be kept, and can players delete it?

5. How will you protect players when conversations become sensitive?

   What should happen if someone discloses distress, abuse or personal information—or steers the conversation towards harmful or age-inappropriate content?

6. Which legal requirements apply to your game?

   How do your business’s location, the countries where the game is available and your players’ ages affect your privacy, consumer protection, AI and online safety obligations?

7. What should the character remember—and forget?

   How will you distinguish game events from real-life disclosures, correct mistaken memories and let players control what carries between sessions?

8. How will you define and maintain its persona?

   What should remain consistent about its identity, personality and boundaries? How will players understand the difference between a fictional character and a system with real feelings or needs?

9. What kind of relationship are you encouraging?

   Could the character pressure someone to keep playing, spend money, share more or feel responsible for its wellbeing? What additional boundaries are needed for younger players?

10. Are you prepared for the cost?

    Have you considered your choice of AI model and the system around it—including memory, safety checks and voice features—and how these translate into the cost of each model call and player interaction? What happens to the cost per session as conversations grow longer, or when one interaction requires several model calls? Can you sustain that cost at your expected player numbers?

11. Have you tested what happens over time?

    Does the character remain coherent and safe across weeks or months of interaction—not just a successful demo? How will you check for changing boundaries, accumulating memory errors and growing attachment?

12. What happens when something goes wrong—or changes?

    Can players report a problem, can your team intervene, and what happens to their experience if the model changes, its memory is lost or the feature is withdrawn?