Hypothesis Generator

The Prompt

Help me form testable hypotheses from these observations:

Observations: [WHAT I'VE NOTICED - patterns, anomalies, correlations]
Domain: [FIELD/CONTEXT]
Resources available: [WHAT I COULD REALISTICALLY TEST]

For each hypothesis (generate 3-5):

1. **Statement:** "If [independent variable], then [dependent variable], because [mechanism]"
2. **Variables:**
   - Independent (what you manipulate)
   - Dependent (what you measure)
   - Controls (what you hold constant)
   - Confounds (what could invalidate results)
3. **Prediction:** Specific, measurable outcome if hypothesis is true
4. **Falsification:** What result would DISPROVE this? (If nothing could disprove it, it's not scientific)
5. **Alternative explanations:** 2-3 other reasons the predicted outcome might occur
6. **Test design:** Simplest experiment that would generate evidence for or against
7. **Required sample/data:** What's the minimum to be meaningful?

Rank hypotheses by:
- Testability (easiest to test with available resources)
- Impact (most useful to know)
- Novelty (most surprising if true)

How To Use It

Describe your observations — things you’ve noticed that seem like they might be connected or patterns that seem too consistent to be random. The AI will formalize these into testable hypotheses.

Why It Works

The gap between “interesting observation” and “testable hypothesis” is where most informal research stalls. This prompt bridges it by forcing the specificity needed for actual testing: measurable variables, falsification criteria, and awareness of confounds.

Variations

Business hypothesis:

I think [BUSINESS ASSUMPTION]. Turn this into a testable hypothesis with: metric to track, minimum sample size, test duration, and success/failure criteria.

Competing hypotheses:

This outcome could have multiple causes. Generate 5 competing hypotheses and design a single experiment that distinguishes between them.

Null hypothesis framing:

My hypothesis is [X]. State the null hypothesis clearly, identify what effect size would be meaningful (not just statistically significant), and estimate the sample size needed.