A reference standard for recording answer inclusion, citation presence, entity accuracy, prompt conditions and changes over time.
Minimum observation standard
| Field | Status | Reason |
|---|---|---|
| Exact prompt | Required | Allows replication. |
| Surface/model label | Required | Products change; “AI” alone is insufficient. |
| Date/time | Required | Answers and citations are volatile. |
| Answer text/capture | Required | Prevents summary bias. |
| Citations/links | Required when available | Shows source selection. |
| Entity outcome | Required | Correct / competing / ambiguous / absent. |
| Source independence | Required | Separates owned, primary, editorial and syndicated evidence. |
| Personalisation context | Recommended | Location/account settings can affect results. |
| Repeat count | Recommended | One response can be an outlier. |
Reporting rule
Do not publish “ChatGPT says X” without the prompt, date and product context. Do not convert ordinary web-search positions into “AI citations.” The purpose of a standard is to make bad news publishable too: a competing entity or no citation is still valid data.