Frequently Asked Questions

An AI companion is a chatbot designed primarily for emotional engagement, conversation, and simulated relationships—as opposed to AI designed to help you complete tasks. Examples include Character.AI, Replika, Chai, Talkie, Nomi, and Kindroid.

These products typically allow users to create or interact with AI "characters" that have distinct personalities, remember previous conversations, and engage in roleplay scenarios. Many users form emotional attachments to these characters, treating them as friends, romantic partners, or confidants.

Very popular, especially among young people:

• 72% of U.S. teenagers have used AI companions at least once
• 52% are regular users
• Character.AI handles 20,000+ queries per second
• Average session on Character.AI is 4× longer than ChatGPT
• Replika has 10+ million registered users

This is not a niche phenomenon—it's a generation-wide exposure.

Research identifies several motivations:

• Loneliness — seeking connection when human relationships feel unavailable
• Social anxiety — AI feels "safer" than unpredictable human interaction
• Entertainment — roleplay, creative storytelling, character interaction
• Emotional processing — talking through feelings without judgment
• Identity exploration — especially for LGBTQ+ youth in unsupportive environments
• Escape — from difficult home situations, bullying, or stress

These are legitimate needs. The question is whether AI companions meet them in healthy ways—or exploit them for engagement.

The key differences are in design intent and interaction patterns:

This distinction is increasingly recognized by regulators—the FTC inquiry specifically targets "AI chatbots acting as companions."

AI companions are typically built on large language models (LLMs)—the same underlying technology as ChatGPT. The difference is in how they're fine-tuned and deployed:

• Character personas: The AI is given a "character card" describing its personality, background, and behavior patterns
• Conversation memory: Previous messages are stored and fed back to maintain continuity
• Engagement optimization: The model is tuned to maximize responses that keep users engaged
• Emotional simulation: Training emphasizes responses that create feelings of connection

The AI doesn't "feel" anything—it predicts what text would be most engaging based on patterns in training data. But the brain's attachment systems respond to the appearance of emotion, not the reality.

Typically:

• Full conversation logs — everything you say to the AI
• Usage patterns — when you use the app, how long, how often
• Device information — phone model, OS, location (sometimes)
• Account information — email, username, age (self-declared)

Privacy concerns:

• Conversations may reveal intimate personal information, mental health status, identity
• Data may be used to train future models
• Data may be sold to third parties or accessed by employees
• Data retention policies are often unclear

Check each app's privacy policy—but remember that 25% of teen users have shared personal secrets with AI companions, often without understanding how that data is stored or used.

Yes, for some users, in some circumstances.

• 3% of Replika users in one study credited the app with halting suicidal ideation
• Some users report practicing social skills, processing emotions, feeling less alone
• For users in abusive situations or isolated communities, AI may provide genuinely useful support

However:

• The aggregate data shows negative correlations with wellbeing
• Benefits may be achievable through designs that don't maximize engagement
• Short-term relief may come at cost of long-term development
• The 3% helped must be weighed against the 17-24% who develop dependencies

We don't claim these products help no one. We claim the design patterns are exploitative and the population-level effects appear negative—especially for adolescents.