1. Frame the question
Define population, timeframe, geography and decision. A narrow question makes omissions easier to spot.
2. Find, then verify
Use AI to discover terms and possible sources, but open the original paper, dataset or official page.
3. Track claims to evidence
Maintain a claim-source table with quotation or finding, page, date and limitations.
4. Test alternative explanations
Ask what evidence would contradict the emerging conclusion and search for it.
5. Separate summary from inference
Label what the source states, what you calculate and what you infer. Do not turn correlation into causation.
6. Document the method
Record search terms, inclusion rules, dates and tools so the work can be checked and updated.
Worked example: a workplace productivity claim
A researcher asks whether a tool improves completion time for a defined task. She searches for primary studies, records population and method, and opens every cited paper. A promising percentage is excluded because it measures self-reported satisfaction rather than completion time. The final note separates source findings, her calculation and her inference.
Keep a decision record
Record the exact use case, tested plan, evidence links, sample input, result, known limitations, responsible reviewer and next review date. This makes the decision explainable when prices, policies or capabilities change.
Red flags to avoid
- Do not cite an AI answer as the source.
- Do not merge studies with incompatible definitions.
- Do not hide uncertainty or contrary evidence.
Decision checkpoint
Write down the evidence, remaining uncertainty and next review date. For changing plan details, follow the official vendor links in the tool directory.