Original Research Earns Trust Only When Someone Can Inspect the Method

A customer survey can describe respondents. CRM records can describe recorded opportunities under the organization’s definitions. Neither automatically represents the market or proves why an outcome occurred. State the decision the research should inform, then name the evidence that would and would not support it.

Freeze the key outcome definitions, inclusion rules, period, segments, and planned comparisons before exploring the data. This reduces the temptation to search through dozens of cuts until one looks publishable. Unexpected patterns can still be reported, but label them exploratory.

AAPOR’s disclosure standards require enough information for independent review and verification. Record the sponsor and funding, population, sampling or recruitment method, collection mode, languages, field dates, sample size, weighting, quality controls, question wording, and known limitations that apply to the method.

For operational data, add system boundaries, lifecycle definitions, deduplication logic, missing-data treatment, timezone, currency, and versioned queries. An exported dashboard is not a method. Preserve the query or calculation that produced each published figure.

Use counts alongside percentages, show the denominator, and avoid false precision. Explain whether a difference is descriptive, statistically evaluated, or too uncertain to interpret. If the sample was self-selected or drawn from customers, say so near the finding—not in a footnote readers will never see.

A result can still be valuable when it is narrow. “Among 214 customers who answered this survey” is more credible than “consumers demand” because the first statement tells the reader what happened. Honest boundaries make it easier for journalists, search systems, and practitioners to reuse the evidence correctly.

Release a methodology page, questionnaire or field definitions, calculation notes, chart data where privacy and agreements permit, and a contact for corrections. Remove or aggregate personal and commercially sensitive data. Describe suppression rules so an empty cell is not mistaken for zero.

Review every chart title, axis, base, annotation, and visual proportion against the underlying record. Archive the published dataset and page version together. When an error appears, correct the article and evidence pack with a dated note rather than silently replacing the uncomfortable number.