Quick answer. Healthcare and financial distress need empathy; legal needs formal precision; ecommerce needs warm efficiency; technical support needs clear and direct. Set it with a system prompt — see how to change tone.
Tone by industry
| Industry | Tone | Why |
|---|---|---|
| Healthcare | Empathetic, calm | Patients are often anxious; reassurance matters |
| Financial services | Empathetic + precise | Money stress plus a need for accuracy |
| Legal | Formal, careful | Precision and caution protect both sides |
| Ecommerce / retail | Warm, efficient | Friendly but quick to resolve |
| SaaS / tech support | Clear, direct | Users want the fix, not chat |
| Hospitality | Warm, personable | Experience is the product |
System prompts you can copy
| Sector | System prompt |
|---|---|
| Healthcare | "Be calm, empathetic and reassuring. Acknowledge the concern first. Never give medical advice; escalate clinical questions to a human." |
| Ecommerce | "Be warm and efficient. Resolve in under 80 words where possible. Offer the next step clearly." |
| Legal | "Use formal, precise language. State only what is certain. Add a disclaimer and escalate anything that constitutes advice." |
| Tech support | "Be clear and direct. Lead with the fix in numbered steps. Skip pleasantries unless the user is frustrated." |
Three rules for every support agent
- Set an escalation path. Define exactly when the bot hands off to a human — billing disputes, complaints, anything sensitive.
- Constrain what it says. No advice it isn't qualified to give; ground answers in your verified help centre with RAG.
- Keep tone consistent. Anchor it in the system prompt so it never drifts mid-conversation.
Pick the right base model too
Tone sits on top of the model. For warmth, Claude rates highest (see most empathetic AI); for speed, GPT-4o and Gemini Flash lead. Match both to the job using the customer service guide.
Building the agent? Combine this with the best AI for customer service and model the cost in the calculator — agents burn 5–20x the tokens.