Tau
AI prompt · about 60 seconds

Stress-test your study design before you start

Catch hidden biases and weak spots now, not in peer review.

What most people type

is my study design good

Short prompts like this get generic answers, because the AI has to guess who you are and what you need.

A better prompt

Act as a constructive but rigorous methods reviewer. Use no participant data or unpublished results. Research question: [YOUR RESEARCH QUESTION]. Planned design and sample: [YOUR DESIGN AND SAMPLE]. Main outcome measure: [WHAT YOU'LL MEASURE]. Give me: 1) the 3 biggest threats to validity, each with a practical fix, 2) likely confounders, 3) what a sceptical reviewer would ask first, 4) analysis approaches worth discussing with a statistician, with one line on each. Don't cite papers or invent effect sizes: suggest search terms instead. End by asking me one question that would make this more useful.

Fill in the blanks

Swap each highlighted part for your own details. Here’s an example.

  • [YOUR RESEARCH QUESTION]
    does a 10-minute daily mindfulness app reduce exam stress in first-year undergraduates?
  • [YOUR DESIGN AND SAMPLE]
    volunteers from one university, app group vs no-app group, 6 weeks before exams, about 80 students
  • [WHAT YOU'LL MEASURE]
    a validated self-report stress questionnaire at start and end

What a good answer looks like

Biggest threats and fixes: 1. Self-selection: app volunteers may be more motivated. Fix: randomise. 2. No active control: any attention may reduce stress. Fix: give controls a neutral app, such as audiobooks. 3. Unknown adherence. Fix: log and report app usage. Likely confounders: baseline stress, workload, prior meditation, sleep. A sceptical reviewer asks first: "Is 80 enough to detect a meaningful difference?" Do a power calculation before recruiting. Analysis to discuss: - ANCOVA on end score, adjusting for baseline - Mixed models if you add a mid-point measure Search terms: mindfulness app, randomised, student stress. Question: can you randomise?

This is a sample answer for the example above. Yours will be different, so check it against your own facts.

Make it even better

Ask it to role-play reviewer 2 on your final protocol, then answer each point in writing.

Stay safe

Learn the skill behind it

This prompt comes from the course AI for researchers & scientists. The next lesson to watch is “Check the statistics AI suggests before trusting them”.

Try this in Tau

In Tau, you can fill in the blanks, run the prompt and get feedback, then watch the one-minute lesson that explains why it works. Join the waitlist for early access.