Using a 'Grill Me' Prompt for AI Knowledge Extraction

Nate Herk | AI Automation2026-06-04go watch the original →

the gist

Improve AI skill accuracy by using a recursive 'grill me' prompt that forces the model to interview you, checkpointing every answer into a markdown knowledge document to ensure no context is lost.

The Breakthrough

By implementing a recursive interview loop that checkpoints every interaction into a persistent markdown document, you can front-load context into your AI system, moving from a 70% success rate on the first iteration to approximately 90%.

What Actually Worked

Adopt a recursive interview prompt: Use a system prompt that instructs the AI to interview you relentlessly, resolve dependencies, ask one question at a time, and explore the codebase before asking for manual input.
Automate checkpointing: Modify the prompt to automatically save every Q&A pair, key decision, and open flag into a markdown file within a brainstorms/ directory at the project root.
Maintain a living knowledge base: Use the generated markdown files as a source of truth that the AI can reference in future sessions, allowing you to update specific processes as your business or project evolves.
Identify knowledge gaps: Treat the AI's inability to answer a question as a signal to flag that specific area for external research or stakeholder consultation, rather than guessing.

Context

Building effective AI skills often fails because of incomplete context transfer from the human to the model. Manual brain dumps are rarely sufficient, leading to iterative cycles of trial and error. This technique treats the AI as an interviewer that forces the user to articulate processes, constraints, and decisions, which are then codified into structured documentation that the AI uses to perform tasks more accurately.