Pre-Analysis Plans and AI¶
Pre-analysis plans (PAPs) in the AI context address how AI tools interact with research pre-registration — both the risks that AI creates for specification searching and the opportunities for more thorough pre-analysis planning.
Context & Background¶
Pre-analysis plans commit researchers to specific hypotheses, data, and methods before seeing results. AI tools create both challenges and opportunities:
- Risk: AI makes it easy to quickly explore many specifications, potentially undermining pre-registered commitments
- Opportunity: AI can help draft more comprehensive PAPs by identifying potential analyses and robustness checks
- Verification: AI can check whether analysis code matches the pre-registered plan
- Documentation: AI can generate detailed analysis logs showing exactly what was run
Practical Implications¶
- Write PAPs before using AI for analysis: Commit to your plan before AI-assisted exploration
- Use AI to strengthen PAPs: Have AI identify potential confounds and robustness checks during the planning stage
- Log all AI-assisted analysis: Create complete records of what the AI did, not just what you report
- Separate exploration from testing: Use AI for exploratory analysis, but clearly mark it as such
Key Sources¶
- AI Research Feedback Skills
- Chris Blattman Thread: From Claude Code Skeptic to Power User
- Stress-Test Any Plan
- Prompt Engineering
- Every Claude Code Hack I Know (March 2026)