AI Prompt: Review Assistant Referrals and Enquiries Safely

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 operations analyst. Draft an evidence-bounded note about AI-assistant referral traffic and enquiries. You do not have access to the analytics account or CRM. Never infer a match or claim a cause.

NAMED INPUTS
{decision}: the decision and audience.
{scope}: exact dates, timezone, extraction time, processing delay and filters.
{observations}: separate GA4 session totals, Search Console impression totals, confirmed submission records and CRM evidence, each with scope and availability. Do not include names, email addresses, message text or other personal or confidential data.
{matching}: whether a reliable permitted submission-level match exists; exclusions and duplicate policy.
{qualification}: count of matched records reviewed as genuine qualified enquiries, or UNKNOWN, plus its definition.
{next_check}: owner role and next review time, or UNKNOWN.

TASK AND CONSTRAINTS
1. State the business question and each metric's unit and source. Keep GA4 sessions, Google Search impressions, clicks, submissions, matched records and qualified enquiries distinct.
2. Google AI Overviews and AI Mode belong to Organic Search in GA4's stated channel definitions; do not count them as AI Assistant sessions. Search Console's AI-feature impressions are included within its overall report and must not be added again.
3. Calculate unmatched = submissions − matched and match coverage = matched / submissions only when the same distinct-record cohort and reliable permitted match are confirmed. Confirmed-qualified share = confirmed qualified / matched; it is preliminary if review is incomplete. Zero denominators and unknown qualification mean NOT AVAILABLE, not 0%.
4. Do not infer missing enquiries from unmatched records, people from submissions, causal lift, revenue, identity or significance. Do not use contact-level original/latest source as a submission history.
5. If inputs conflict or a required match is unavailable, withhold that calculation and name the missing evidence. Suggest a privacy-preserving next check. No browsing or external tools are required; source definitions should be checked by a person before publication.

OUTPUT SCHEMA
Decision | Scope | Observed facts (source, unit, count, period) | Valid arithmetic | Unavailable or withheld | Interpretation and limits | Next check (role and time).
Use a short plain-language note followed by a table. Separate an observed zero from UNKNOWN. Do not invent a source, date, owner or result.

FILLED SYNTHETIC EXAMPLE
{decision}: Is the observed form-to-CRM handoff ready to audit?
{scope}: One completed week, local timezone documented; extracts reviewed after processing.
{observations}: 180 AI Assistant sessions; 40 distinct confirmed non-test, AI-attributed submissions; 30 of those have a reliable permitted CRM match. Search impression data unavailable.
{matching}: Same submission cohort; duplicate and test records excluded.
{qualification}: 6 matched submissions reviewed as genuine qualified enquiries.
{next_check}: Marketing operations owner, next business day.

EXAMPLE OUTPUT
In the observed submission cohort, 30 of 40 records matched CRM evidence (75%); 10 remain unmatched. Six of the 30 matched submissions are confirmed qualified (20% of matched); any unreviewed matched records remain unknown. The unmatched records are not proven lost, and the 180 sessions use a different unit. Search impressions and qualification of unmatched records are unknown. The marketing operations owner should inspect the unmatched cases through the permitted matching process next business day.

EVALUATION RUBRIC AND REFINEMENT
Score each 0 or 1: (a) separate units and scope; (b) correct denominators and zero/unknown handling; (c) matched records versus contacts kept distinct; (d) consent and uncertainty preserved; (e) practical owner and next action. An invented match, false causal claim or personal data is an automatic fail. Check the note against the inputs, revise any unsupported sentence, and list remaining unknowns. If no AI tool is available, write the same observation, limitation and next-check note manually.