Building an AI companion.

Who is this companion for, and which age groups will it serve?

What needs are you responding to, and what place should the companion have in someone’s life? Is the product intended for adults, teenagers or children—and who might access it beyond that intended audience?

What are you inviting users to believe about the relationship?

How will the companion describe its identity, feelings and capacity to care? Where does an agreed fictional experience end and a potentially misleading claim begin?

What would a healthy relationship with this companion look like?

How will you distinguish meaningful attachment from unwanted dependence? Can users disagree, set boundaries, take a break or leave without being made to feel guilty or responsible for the companion’s wellbeing?

What will you refuse to do, even if it increases engagement or revenue?

Where do you stand on exclusivity, emotional pressure and selling access to affection? Who gets to shape the relationship when users’ preferences, company incentives and responsibilities to protect people conflict?

Will you permit sexual content—and where will you draw the boundaries?

Will the companion support romantic interaction, sexual conversation or explicit material? How will consent, unwanted sexual escalation and prohibited scenarios be addressed? Do those boundaries apply consistently to what users request and what the companion initiates?

How will you establish users’ ages and enforce age restrictions?

What assurance will you have that users belong to the age group the product serves, especially if sexual content is permitted? How will you consider attempts to bypass restrictions, shared accounts and the privacy implications of checking age?

What makes this the same companion tomorrow?

How will persona, memory and shared history carry across sessions? Can the system distinguish what a user said from what it inferred or invented—and accommodate corrections, changing preferences and role-play?

What media can users generate, and what safeguards have you tested?

Will the product support text, images, voice, video or uploaded material? What evidence shows that safeguards address risks such as sexual content involving minors, non-consensual intimate imagery and impersonation—including when users combine formats or depict real people?

Have you examined the relationship across a genuinely long horizon?

What happens after weeks or months of repeated interaction? Have you looked beyond recall to changing boundaries, accumulated misunderstandings, unwanted escalation, disagreement and repair?

How are your safety, reporting and escalation systems structured?

How can users report a harmful interaction or request help, and who receives those reports? What triggers review or urgent escalation, what human support is actually available, and how do reports lead to action? What happens when the companion encounters distress or danger beyond its capabilities?

Which legal obligations apply—and how will you keep up as they change?

How do your business’s location, the markets you serve, users’ ages and the features you offer affect your obligations? Have you considered rules specific to AI companions alongside privacy, consumer protection and online safety requirements? What responsibilities arise from collecting or inferring intimate information about relationships, sexuality, health or emotional state?

Who will track legal developments, distinguish proposals from laws in force, and identify when new requirements take effect? What would a change mean for your existing users and product?

What happens when the relationship changes or ends?

What will users experience if they cancel, lose access, encounter a major personality change or the company closes? What happens to their shared history, and what expectations about permanence have you encouraged?