The AI-Assisted Audit Testing Playbook

You ran the prompt. The AI said it tested all 25 samples. It didn't.

Or worse — it tested them, but the comments are so vague your reviewer can't tell what was actually evaluated.

Now you're re-doing work that was supposed to save you time.

This playbook exists because I hit every one of those walls during a real Test of Effectiveness (TOE) where I used AI to manually review over 400 screenshots and follow 2 different business desktop procedures (DTPs). Throughout the process I documented where AI failed, how I changed the prompts, and what actually ended up working to ultimately produce reviewer-ready output.

📦 What's Inside

→ An 8-phase operational workflow from Desktop Procedure analysis through workpaper finalization

→ A Controlled Execution Prompt with built-in guardrails that force the AI to stop, reconcile, and prove its work before moving forward

→ An evidence tracker structure designed to catch gaps before testing begins — not after

→ 4 field-tested prompts covering DTP decomposition, evidence reconciliation, attribute testing, and manager communication

→ A lessons learned table from real fieldwork failures

→ The full prompt evolution story showing how a naive "run fast" prompt failed and what replaced it

✅ This Is

• An operational field guide for auditors already using AI who need structure around their testing
• Built from a real audit test, not theory
• Designed to make AI failures loud and early instead of silent and late

🎯 Who It's For

Internal auditors running TOE testing with AI assistance who are tired of:

• Re-running prompts because the output wasn't specific enough
• Discovering skipped samples at the workpaper stage
• Getting vague AI-generated comments that don't cite evidence
• Spending more time fixing AI output than it would've taken to do it manually

🔧 Works with any LLM — ChatGPT, Claude, Gemini, Perplexity. No API key needed. No special setup.