Ethics & AI Principles

Why ethics is a product feature, not a footer

Twogle is a couples-wellness app that uses AI. The market is full of relationship apps that use AI as a marketing word and a privacy footnote. We are publishing the page you are reading because we think the ethics of how AI is used in a relationship app should be something a couple can read, evaluate, and disagree with — not something buried inside a 4,000-word privacy policy.

What follows is five commitments. Each is concrete, not aspirational. If we ever break one, we will publish a correction on this page, dated and signed.

1. We do not train AI on user data

Your conversations inside Twogle — your private chat with yourself, your shared chat with your partner, your responses to daily prompts, your relationship context — are never used to train any AI model. Not ours (we do not train our own models), and not our AI providers'.

We process user data through Anthropic's Claude API and OpenAI's API under their respective enterprise agreements, both of which prohibit training use of API customer data. We do not opt into any training-data sharing on either platform.

Practical implication: if you delete your account, your messages disappear from our servers within 30 days (within 90 from backups), and they were never part of any model's training corpus in the first place.

2. Both partners consent separately

Twogle's couples-mode features — the shared chat, the shared insights, the partner-comparison views — require explicit, separate consent from both partners. One partner cannot use Twogle's AI to surveil, analyse, or profile the other without that partner's informed agreement, recorded in our system.

This is a structural product decision, not a setting. It means we cannot offer features that would be useful to one partner only at the cost of the other partner's autonomy. We have turned down feature requests that would violate this commitment.

3. Crisis escalation is built in by design

Twogle's AI is bounded. For specific signals — self-harm, suicidal ideation, domestic abuse, child safety, active addiction without sobriety, sexual assault — the AI does not attempt to coach, mediate, or "solve." Instead it routes the user to local crisis resources and offers a clearly framed handoff to a real human practitioner via our Check-In service.

Crisis resources we surface, by region:

• United States: 988 Suicide & Crisis Lifeline; National Domestic Violence Hotline 1-800-799-7233; RAINN 1-800-656-HOPE; Crisis Text Line (text HOME to 741741).
• United Kingdom: Samaritans 116 123; Refuge 0808 2000 247; Mind 0300 123 3393.
• Canada: Talk Suicide 1-833-456-4566 (text 45645); Hope for Wellness 1-855-242-3310; ShelterSafe shelterssafe.ca.

We do not believe AI is the right tool for any of these conversations. We surface real services. This is not a feature we plan to "soften" with future model updates.

4. The AI is clearly disclosed

When you are talking to a Twogle AI feature, the app tells you it is an AI. We do not use fictional human personas. We do not give the AI a name designed to imply it is a person (no "Dr. Twogle," no first-name personas designed to feel relational in a misleading way). The AI is the AI.

We extend the same standard to our editorial content. Every article on twogle.com has a named human author. We do not publish AI-generated content under pen names or implied human bylines. Our editorial policy describes how AI is used in the writing process — for research and first drafts — and how a human editor substantively rewrites every piece before it ships.

5. We are open about our AI providers

Twogle uses two AI providers in production:

• Anthropic, PBC (Claude) — primary AI coaching responses and relationship guidance. Anthropic's privacy policy.
• OpenAI, LLC (GPT-4o mini and Moderation API) — content safety filtering and intelligent message routing. OpenAI's privacy policy.

If we add or change providers, this page is updated. Our privacy policy contains the full list of what data is sent to which provider and why.

What this isn't

Twogle is not therapy, and our AI features are not a substitute for clinical care. The Check-In service is a single 50-minute structured session with a certified practitioner — it is not ongoing therapy either. If you need ongoing clinical support, please see our guide to couples therapy or contact a licensed therapist directly via the professional directories linked in that article.

Tell us when we get this wrong

Ethics is a moving target. If you think we are failing on any of these commitments — or if you have a question about how AI is used in a feature you are seeing — email ethics@twogle.com. We read every message and respond within 72 hours.