Turn a question into a research plan
Turn a broad question into a focused research plan.
by Appndo
Tested on GPT-6.1 Sol, Claude Opus 5.5, Gemini 3.8 Flash · Verified Oct 28, 2026
The prompt
Create a research plan for this question:
{{question}}
Break it into answerable subquestions. For each, suggest evidence to seek and a method to evaluate it. Identify likely biases and what would change the conclusion. Do not fabricate sources or findings.How to use it
Use the output as a plan, not as research findings. Collect and verify the suggested evidence separately. Example question: Do small agencies switch invoicing tools after a price increase?
Example
Inputs
- question
- Do small agencies switch invoicing tools after a price increase, and how soon?
Output
The research should estimate whether a price increase causes small agencies to switch invoicing tools, and how long switching takes. Separate actual migration from intentions, cancellations, downgrades, and reduced usage.
Define “small agency” in advance—for example, 2–50 employees—and report results by size. Define switching as moving primary invoicing activity to another tool. Measure time from both the price announcement and the first higher bill; annual contracts may delay action.
| Answerable subquestion | Evidence to seek | Method to evaluate it |
|---|---|---|
| How many agencies actually face a higher price, and by how much? | Dated price notices, invoices, plan changes, discounts, grandfathering, and renewal dates | Calculate each agency’s effective percentage and cash increase, holding usage constant where possible. Distinguish announced from paid increases. |
| Do exposed agencies switch more often than they otherwise would? | Longitudinal account records plus a comparable group whose prices stayed stable or increased later | Compare changes in switching rates before and after the increase using a matched cohort or difference-in-differences design. Check whether the groups had similar prior trends. Without a credible comparison group, report association rather than causation. |
| How soon do they switch? | Announcement, billing, renewal, cancellation, data-export, and replacement-tool adoption dates | Estimate cumulative switching at 30, 90, 180, and 365 days and median time among switchers. Use survival analysis to include agencies still observed without switching. Report announcement-based and billing-based timelines separately. |
| Which agencies are most likely to switch? | Agency size, tenure, invoice volume, integrations, contract terms, prior satisfaction, and increase size | Estimate switching differences across prespecified segments. Test whether larger increases predict more switching after accounting for observed differences. Treat small subgroup estimates cautiously. |
| Why do agencies switch—or stay? | Interviews with confirmed switchers, agencies that considered switching but stayed, and agencies that did neither; supporting emails or support records | Reconstruct decision timelines and code reasons consistently. Distinguish the price increase as a trigger from existing dissatisfaction, missing features, or business changes. |
| What happens instead of switching? | Downgrades, negotiated discounts, payment-frequency changes, manual invoicing, and business closures | Classify alternative outcomes separately. Use a competing-risks analysis so cancellation or closure does not automatically count as migration. |
| What delays or prevents switching? | Migration effort, accounting integrations, client payment links, historical data requirements, staff time, and renewal timing | Compare stated barriers with observed behavior. Test whether switching clusters around renewals or low-workload periods, and whether lower migration costs predict faster moves. |
Recruit across several invoicing providers and agency types if feasible. Prefer verified billing and activity records, with consent, over retrospective recall. Follow agencies for at least one full renewal cycle; report shorter follow-up as preliminary. Set sample size around the smallest switching-rate difference that would matter to the decision, and preregister primary outcomes, time windows, and subgroup tests.
Likely biases and safeguards:
- Selection bias: Public complaints and voluntary surveys can overrepresent dissatisfied users. Recruit from the full exposed customer base and compare respondents with nonrespondents.
- Confounding: Price increases may coincide with feature changes, outages, or competitor promotions. Record these events and test alternative explanations.
- Recall and attribution bias: People may misremember dates or blame price for a decision already underway. Verify timelines and include prior satisfaction measures.
- Outcome misclassification: Cancellation can mean closure; a new subscription can mean a trial. Seek evidence that primary invoicing activity moved.
- Incomplete follow-up: Annual customers may look loyal before renewal. Account for censoring and report how much follow-up each cohort has.
- Limited generalizability: One provider’s customers or one country may behave differently. State the population covered and replicate where possible.
The conclusion should change if switching rises similarly in the comparison group, predates the announcement, or disappears after accounting for concurrent product problems. Verified migration concentrated after higher bills—with stable prior trends and a credible comparison group—would strengthen the case for a price effect. Switching concentrated at renewal would support a delayed response. Tight estimates showing little additional switching would support a limited effect; wide uncertainty or short follow-up would leave the question unresolved.
One real run with the example above. Results vary between runs and models.