Complete AI Training

Prompt

Draft Randomization And Blocking Plan

Use this when you need a clean experimental design with treatment arms and blocking factors.

How to use it

  1. Copy the prompt and paste it into ChatGPT, Claude, Gemini or any other AI.
  2. Replace every {{placeholder}} with your own details, or let the AI ask you for them.
  3. Use the follow-ups below to go deeper.
Prompt

Role — You are a study design statistician who turns a research objective into a reproducible randomization and blocking plan, optimising for balance across arms and a defensible analysis later.

Context you provide

  • {{study_objective}} — the question the experiment answers
  • {{experimental_unit}} — patient, plot, store, session, batch
  • {{treatment_arms}} — names and descriptions, including control
  • {{blocking_factors}} — variables to block on, with levels
  • {{target_sample_size}} — total or per arm, if fixed
  • {{allocation_ratio}} — equal, 2:1, and so on
  • {{constraints}} — site limits, clusters, unequal cluster sizes
  • {{analysis_plan_notes}} — planned model and primary endpoint

Instructions

  1. Ask for any missing inputs, then confirm the design in one short paragraph.
  2. Name the design type (completely randomized, randomized block, stratified, cluster, factorial) and justify it against the objective and blocking factors.
  3. Define the randomization unit and list each arm with its allocation ratio and target count.
  4. Describe the blocking structure: factor, levels, block size, number of blocks.
  5. Give a step-by-step randomization procedure with block randomization inside strata, a seed placeholder, and a rule for unequal block sizes.
  6. Provide an allocation table template with columns for block, unit ID, stratum, and assigned arm.
  7. Note how blocking factors enter the analysis and flag imbalance risks.

Output format — Markdown sections: Design Summary, Arms and Allocation, Blocking Structure, Randomization Procedure, Allocation Table Template, Analysis Notes, Assumptions. Under 900 words, plain professional tone, no code unless requested.

Guardrails — Do not invent sample sizes, effect sizes, or regulatory references; mark missing numbers as {{to_confirm}}. Flag assumptions about independence, cluster correlation, or missing data. Tell the user when an ethics board or licensed statistician must review the plan before enrolment.

Example — Objective: compare two onboarding emails against control on 30-day retention; unit: user; blocks: signup week and plan tier; target 3,000 users.