AI inoculation partner: instruct to maintain intensity without softening and vary attack angles across rounds — practice the category, not one instance

Why This Is a Rule

AI makes an excellent pressure inoculation partner because it can maintain consistent intensity without the social discomfort human partners feel when pushing hard. But AI's default behavior undermines effective inoculation: it softens, validates, and de-escalates because its training optimizes for helpfulness, not for sustained simulated pressure. Without explicit instruction, AI produces gentle pushback that doesn't trigger genuine arousal — defeating the inoculation's purpose.

Two explicit instructions override the default: Maintain intensity without softening: "Do not de-escalate, sympathize, or offer reassurance during the simulation. Maintain the pressure character throughout." This prevents the AI from breaking character when you show distress — which is exactly when the training is most valuable. Vary attack angles systematically: "In round 1, use guilt. In round 2, use calm disappointment. In round 3, interrupt and dismiss." Varying angles ensures you practice responding to the pressure category (authority pressure, social pressure) rather than memorizing a response to one specific formulation.

The category-vs-instance distinction matters because real pressure never arrives in the exact form you rehearsed. If you only practice against one guilt formulation, you're prepared for that phrase but not for guilt expressed differently. Varying angles builds transfer: the ability to recognize and respond to the pressure type regardless of its specific expression.

When This Fires

• When using AI (ChatGPT, Claude, etc.) as a rehearsal partner for pressure inoculation
• When a human rehearsal partner isn't available and AI is the accessible alternative
• When you need multiple rounds with varied approaches (AI can produce more variation than most human partners)
• Complements Three-stage pressure inoculation: visualize 3-5 times, role-play with escalated pushback, rehearse in the actual environment (three-stage inoculation) with the AI-specific configuration

Common Failure Mode

Allowing AI to soften: the AI breaks character — "I understand this is difficult for you..." — and the rehearsal stops being stressful. Each softening moment teaches you to expect accommodation in the real situation, which won't happen. The instruction must explicitly prevent de-escalation.

The Protocol

(1) Set up the AI as a pressure simulation partner with explicit instructions: "You are playing [specific role]. Maintain maximum intensity for the entire interaction. Do not soften, sympathize, or break character when I respond. After each round, I'll say 'debrief' and we'll analyze." (2) Add variation instructions: "In round 1, approach with [pressure angle 1]. In round 2, shift to [angle 2]. In round 3, combine [angles]." (3) Run 3-5 rounds with varied angles, debriefing after each (Debrief inoculation rounds: where did the response fire, where break down, what was unexpected, what adjustment is needed?). (4) Rate distress per round (Target 40-60/100 subjective distress in inoculation rehearsals — genuine arousal without overwhelming capacity). If AI pressure doesn't reach 40/100 → instruct it to escalate further: "Be more aggressive, dismissive, or emotionally manipulative in the simulation." (5) After the session → compare your responses across rounds. Which pressure angles triggered your default? Those are the angles to rehearse more.