Hypothesis generation prompt builder
Good research starts with good hypotheses — testable, specific explanations you can actually go and check. Asked casually, an LLM tends to produce mushy “it might be because…” answers that are impossible to act on. The fix is structure: tell the model to frame each idea as a hypothesis, attach a rationale, name the evidence that would support it, and state how it could be falsified. This builder assembles that scaffold around your research question.
How it works
You provide the domain and the research question, then choose the evidence type the model should reason about and the output format. The tool writes a prompt that asks for a set number of distinct hypotheses, each returned with four parts: the hypothesis itself, a short rationale, the kind of evidence or study that would support it, and a falsifiability test describing what observation would prove it wrong. The result is a structured, comparable list rather than a wall of speculation — easy to scan, rank, and pursue.
Tips and example
- Keep the question narrow. “Why did D7 retention drop in Brazil after the March release?” produces far sharper hypotheses than “why is retention bad?”.
- Treat named citations with suspicion. The scaffold deliberately asks for evidence types rather than specific papers, because models fabricate references. If the model names a study anyway, verify it before relying on it.
- Use the falsifiability line as your to-do list. Each test tells you the cheapest experiment or query that would confirm or kill the hypothesis.
- Generate in batches. Ask for several, discard the overlapping ones, and re-run on the survivors to deepen the strongest candidates.