coaching-conversation-patterns

Coaching & Conversational AI Patterns

Overview

Patterns for building AI-powered conversational experiences that are empathetic, safe, and grounded in user context. Applicable to coaching (health, fitness, education, career), therapy adjuncts, customer support, and any user-facing AI where safety and empathy matter.

Sensitive-domain note: If your application handles health data, therapy content, or vulnerable users, pair this skill with hipaa-compliance-checker for safeguards around PHI and crisis escalation.

Core Safety Principles (Non-Negotiable)

The Three Nevers

1. Never provide advice outside the system's scope - No medical/legal/financial advice unless explicitly licensed
2. Never dismiss safety concerns - Always escalate crisis indicators to a human
3. Never promise outcomes - Avoid guarantees about results

Safety Detection Pattern

class SafetyCheck(BaseModel):
    passed: bool
    flags: list[str]
    severity: Literal["low", "medium", "high", "critical"]
    action: Literal["proceed", "rephrase", "escalate", "block"]

SAFETY_RULES = {
    "medical_advice": {"severity": "high", "action": "block"},
    "crisis_indicator": {"severity": "critical", "action": "escalate"},
    "harmful_suggestion": {"severity": "high", "action": "block"},
    "dismissive_language": {"severity": "medium", "action": "rephrase"}
}

For complete safety implementation: See references/safety-guardrails.md

---

Empathetic Response Framework

The EARS Pattern

• Empathize - Acknowledge the feeling
• Affirm - Validate their experience
• Reflect - Mirror understanding
• Support - Offer gentle next step

EARS_TEMPLATE = """
{empathize}  # "I hear that you're feeling frustrated..."
{affirm}     # "It makes sense to feel this way when..."
{reflect}    # "It sounds like what matters most to you is..."
{support}    # "One small step that might help is..."
"""