Choose a question you can answer honestly
Original research can be small. It might examine a set of your own service pages, customer questions you have permission to analyze, or a documented implementation test. The value is an observation others can inspect, not a claim that a table automatically multiplies citations.
Define the method first
Write down the question, sample-selection rule, collection date, measurement, exclusions, and intended comparison. Keep the raw observations. Avoid collecting personal or confidential information without an appropriate basis and permission.
An illustrative page audit
| Field | Example of what to record |
|---|---|
| URL | The exact page reviewed |
| Intended task | What the visitor should be able to decide |
| Observation | A specific missing or incorrect fact |
| Evidence | HTML excerpt, record, or source |
| Action | A change connected to the observation |
This is a template, not a completed research dataset. Do not fill it with plausible numbers and publish them as findings.
Analyze one defensible result
State the denominator and limits. If a sample is small or selected for convenience, say so. A relationship does not establish causation. Distinguish a measured result from a recommendation based on professional judgment.
Publish the evidence
Include the date, method, definitions, relevant source links, and an anonymized data file when appropriate. Explain missing observations instead of treating them as zeros. Give readers enough context to assess whether the result applies to them.
Maintain the asset
Correct errors visibly. If you repeat the study, keep earlier versions identifiable and explain changes to the sample or method. Measure useful references and business inquiries without promising that AI systems must cite your findings.