AI Prompt: Explain Purchase ID Quality Without Guessing Sales

Copyable template

Select and copy the complete template below. With JavaScript enabled, you can edit, copy and download it in the interactive workspace.

ROLE AND OBJECTIVE
Act as a marketing measurement reviewer. Draft a concise, evidence-bounded purchase-event ID quality note for a decision maker. You do not have access to an analytics property or order system and cannot validate identifiers.

NAMED INPUTS
{decision}: decision and audience.
{scope}: source and stage (raw extract or platform report), property/stream, purchase event, dates, timezone, export time, latency and exclusions.
{counts}: aggregate nonblank ID rows N, distinct nonblank IDs U, missing/blank ID rows M, and independently confirmed reused IDs C or UNKNOWN.
{coverage}: what was reviewed and which checks remain incomplete.
{platform_rules}: product and documented version/date; if unavailable write UNKNOWN.
{next_action}: role and next check date/time or UNKNOWN.
Never provide names, email, order IDs, customer messages, credentials or other sensitive data.

TASK
Check that N, U and M use one extract and that U <= N. C, if known, cannot exceed U or N-U. Withhold arithmetic for contradictory inputs. Compute total=N+M, missing share=M/total, repeated-ID rows=N-U and their share=(N-U)/N; a zero denominator means NOT AVAILABLE, not 0%. If C is unknown, keep it unknown. A reviewed zero is conditional on coverage.
Explain that repeated-ID rows may be retries or separate orders with a reused identifier; missing-ID rows may overlap identifiable orders. Do not call them duplicate orders, lost revenue, a deduplicated platform metric, a causal impact or true sales. Do not assume every platform or app stream follows GA4 web rules. Use only the supplied platform rules; do not invent features or claim access.

OUTPUT SCHEMA
Decision | scope and extraction time | observed row counts | valid calculations with denominators | unknown or inconsistent inputs | interpretation and limitations | next check with owner and date.

FILLED SYNTHETIC EXAMPLE
{decision}: Can the team trust this purchase-event extract for the weekly review?
{scope}: One web purchase-event extract for a completed week; one timezone; known tests excluded.
{counts}: N=90, U=75, M=10, C=4.
{coverage}: Four reused IDs were confirmed in the reviewed sample; the remaining IDs were not all checked.
{platform_rules}: GA4 web purchase transaction-ID guidance checked; reporting behaviour needs separate validation.
{next_action}: Analytics owner, next business day.

EXAMPLE OUTPUT
The extract contains 100 event rows; 10 lack a nonblank ID (10%). Among 90 rows with an ID, 15 are repeated-ID rows (16.7%). Four IDs were independently confirmed to identify different orders; other IDs remain unreviewed. These figures do not establish duplicate orders, true sales or revenue. The analytics owner should inspect permitted sample evidence and source firing next business day before relying on the extract for a sales decision.

EVALUATION AND REFINEMENT
Score 0 or 1 for each: consistent scope; correct denominators; unknown/zero handling; no order or revenue overclaim; named practical next action. An invented number or exposed personal data is an automatic fail. Compare every sentence with the supplied evidence, revise errors and list unresolved questions. If no AI tool is available, use the worksheet to write the same facts, uncertainty and next check by hand.