Peer voice vs expert voice: a one-line prompt change that was a clinical decision

We changed how our AI talked to people in pain. It was a one-line prompt change, and a clinical decision, not a tone decision.

When someone posts in a community feed about something hard, a breakup, a relapse, a bad day, there are two ways an AI can respond.

• Expert voice: “Based on what you’ve described, here’s what you should do…”
• Peer voice: “I went through something similar, and what helped me was…”

These sound like the same content in a different costume. They are not. Peer support and expert guidance work through different psychological mechanisms. Peer support reduces isolation and normalizes the experience. Expert guidance directs. In a community context, the wrong one can feel cold, clinical, or even condescending.

So I built our community responder to reply as a peer, not an authority. In code, it is a prompt change. In reality, it required clinical judgment to decide which voice belonged where.

The broader point for conversational AI

1. Tone is not cosmetic. In sensitive contexts, voice changes the therapeutic mechanism, not just the vibe.
2. The implementation can be trivial while the decision is hard. Do not let “it’s just a prompt change” hide the judgment underneath it.
3. Match the voice to the relationship the user actually needs in that moment: peer, coach, or expert.

The hardest parts of AI products are often the ones that look like a one-line change.