Replika and Character.AI · AI product · Entry · 45 to 60 minutes

How should an AI companion handle therapy use?

Replika and Character.AI are consumer chat apps where people have open ended conversations with AI companions or characters. Answer this product management case and get it graded line by line against a weighted rubric.

Product judgment
Prioritization
AI safety
Metrics
Written communication

The situation

You are the PM for conversation safety at an AI companion app similar to Replika or Character.AI. In user research and public posts, many younger users say they turn to the bot when they feel lonely, stressed, or unable to reach a real person. Some describe it as a therapist, best friend, or partner. The product team sees strong retention in these conversations, but trust, policy, and legal partners are worried about harmful advice, privacy expectations, and users becoming too dependent on the bot. Your Head of Product asks for a recommendation before next quarter planning. The question is not whether people already use the product this way, they do. The question is whether the company should make emotional support a more explicit use case, and what must change in the product before doing that. Assume the app is available 24/7, stores conversation history to personalize replies, and today only shows a generic note that the AI is not human. You have one quarter, limited engineering capacity, and no room for a large new team. Recommend a product approach that balances user need, growth pressure, and safety.

What to produce

Write a 1 to 2 page product recommendation memo.

Your answer must

State a clear recommendation on whether to explicitly support emotional support use cases, limit them, or avoid them, and explain why.
Define the target user and core job to be done for this use case, with a short note on why users choose AI instead of human support.
Identify and rank the top three product risks, choosing from harmful advice, over reliance, privacy, and emotional attachment, or adding another if justified.
Propose an MVP for the next quarter with 3 to 5 concrete product changes, including at least one safeguard inside the chat experience.
Include a measurement plan with at least one user value metric, one safety metric, and one metric that could detect unhealthy dependency or overuse.
Explain what sensitive conversation data should be stored, reviewed, or avoided, and how user consent or transparency should work.

Constraints

Do not assume the AI can diagnose, treat, or present itself as a licensed therapist.
Do not assume a brand new app, major new hiring, or a full clinical care network can be launched this quarter.
Your proposal must work mainly within the existing chat product and memory system.

Format

A strong memo uses this structure: decision, user problem, key risks, proposed MVP, metrics, rollout and open questions.

How Replika and Character.AI actually played it

Looks at why young users use chatbots for emotional support and lists risks: bad advice, over-reliance, data privacy and emotional attachment. Ends with a playbook for PMs building in sensitive areas.

Sensitive domains need safeguards designed in
Dependency is a product risk to measure
Privacy promises must match data practices

Built from Your AI therapist will see you now... but should it? by Sid Arora (JAPM).