Prepare for statistical reviewer questions with AI

Use AI as a skeptical reviewer before the real one arrives. It finds gaps fast; you decide which ones are real.

Facts on this page are checked against the sources linked here. The prompt has not yet been run on a live assistant. How we check.

Step 1: Generate the questions

Act as a skeptical statistical reviewer at a regulatory agency.
Review the analysis plan summary below. Produce the 15 most likely
questions, grouped by: estimand, missing data, multiplicity,
sensitivity analyses, subgroups, model assumptions, data integrity.
Rank within each group by seriousness. For each question, say in one
line why a reviewer would ask it.

SUMMARY: [paste - public or simulated only]

Step 2: Draft response outlines, not answers

For the 5 most serious questions, draft a response OUTLINE:
- the position to take
- the evidence or analysis I would need to point to
- what I must verify before answering
Do not cite any guidance document, paper or regulation unless you can name
it exactly. Mark every citation [VERIFY CITATION]. Never invent numbers.

Review checklist

• Each question is something a real reviewer could ask (cut the generic ones)
• Every citation was verified against the original source
• No invented results or numbers in any draft response
• Responses reflect decisions your team actually made
• Gaps that are real become action items, not just talking points

Worked example (public data)

For the public record NCT02906930 (oral semaglutide, three doses vs placebo, primary outcome HbA1c change at week 26), the questions an AI reviewer should raise include: how treatment discontinuation and rescue medication are handled in the estimand, how missing week-26 data is treated, how multiplicity across three doses is controlled, and whether the "ratio to baseline" endpoints were analyzed on a log scale. A good run lists these as questions to answer, not answers.

Get the next workflow by email →