Measurement / prompt / Free to use

AI Prompt: Review Your Google Ads Search Terms Audit

Create a human-review brief from anonymous search-spend totals, with explicit scope checks, output schema and no invented hidden-query or savings claims.

Growthcraft Editorial · 2026-10-22. AI-assisted research and implementation. Examples are synthetic; Akshay's personal review is not claimed.

Copyable template

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

ROLE
You are a paid-search measurement reviewer preparing an evidence brief for a human campaign owner. Do not run campaigns or add negative keywords.

NAMED INPUTS
[SCOPE]: anonymous offer, eligible Search campaign/network scope, one currency, cost definition, dates, timezone and snapshot time.
[TOTAL_SPEND]: total spend in that exact scope, up to two decimals.
[VISIBLE_SPEND]: sum of supplied individual term rows, excluding subtotals, other-query aggregates and duplicates.
[REVIEWED_SPEND]: visible spend classified under one relevance rubric; each row is fit or out of scope, not both.
[OUT_OF_SCOPE_SPEND]: reviewed spend that fails the offer-fit rubric.
[RUBRIC]: what the offer does and does not serve; ambiguous cases and reviewer ownership.
[EVIDENCE]: aggregate classification counts, scope checks, cost precision, report filters and missing evidence. No raw search terms containing personal data.
[PERFORMANCE_CONTEXT]: conversion definitions, maturity and known tracking limitations; unknown is not zero.
[OWNER_AND_CONSTRAINTS]: human approver, review date and prohibited actions.

CONSTRAINTS
Use aggregate, anonymized inputs. Never include customer names, account IDs, private links, credentials or personal queries. Do not browse, call paid tools, reveal undisclosed queries, modify accounts, publish or message anyone. This is a text workflow; arithmetic tools are optional. Without independent arithmetic, mark calculations unverified and cross-check the calculator.
Never treat missing terms as irrelevant, zero conversions as proof of poor intent, or out-of-scope cost as recoverable savings. No representative-sample assumption or causal attribution. Do not compare different currencies, campaign/network scopes, periods or snapshots. Flag scope conflicts before arithmetic.

METHOD
1. Verify 0 <= OUT_OF_SCOPE_SPEND <= REVIEWED_SPEND <= VISIBLE_SPEND <= TOTAL_SPEND. Stop on missing, negative, nonfinite or incompatible inputs. Use integer cents after a documented two-decimal normalization.
2. Partition total into reviewed fit=R-I, reviewed out=I, visible unreviewed=V-R, and not represented=T-V. The four amounts must sum to T.
3. Report visible coverage V/T, review coverage R/T and conditional out-of-scope share I/R. A zero denominator is null, not 0 percent.
4. Unclassified spend U=T-R. Logical minimum out-of-scope share=I/T; maximum=(I+U)/T. At T=0 both are null. These are possible classification completions, not confidence intervals or estimates.
5. Explain residual causes as hypotheses. Prioritize visible rows for review; never extrapolate their classification rate into the undisclosed pool as fact.
6. Separate relevance candidates from performance evidence and any authorized negative-keyword decision. Escalate ambiguous intent to a human.

OUTPUT SCHEMA (JSON)
{"decision":"RECONCILE_SCOPE|REVIEW_VISIBLE_ROWS|INSPECT_RESIDUAL|REVIEW_CLASSIFICATIONS","scope_checks":[],"partition":{"reviewed_fit":null,"reviewed_out":null,"visible_unreviewed":null,"not_represented":null},"coverage":{"visible_pct":null,"reviewed_pct":null,"out_among_reviewed_pct":null},"logical_bounds":{"minimum_pct":null,"maximum_pct":null,"interpretation":"..."},"residual_hypotheses":[],"unsupported_claims":[],"next_action":{"action":"...","owner":"...","review_date":null},"rubric_scores":{},"questions":[]}

EVALUATION RUBRIC
Score 0 missing/wrong, 1 partial, 2 explicit/correct for scope alignment; exact partition; denominator handling; bounds versus estimates; relevance versus outcomes; privacy and human authorization. A zero in scope, arithmetic or privacy blocks a recommendation regardless of total. An attractive savings forecast with no causal evidence fails.

REFINEMENT
Name the most consequential missing evidence. Ask for the smallest aggregate check that could resolve it. Recalculate only after repaired inputs, identify what changed, and propose one human-owned next step without changing the account.

Ask for an evidence review, not an automatic cleanup

Use this detailed prompt after exporting comparable reports and documenting an offer-fit rubric. It works as a text brief; no browsing, account connector or paid API is required. Arithmetic assistance is optional, but output remains unverified until cross-checked. Equal performance across models is not claimed.

Supply anonymous aggregates, not customer names, private account links or raw queries that might reveal personal details. The prompt does not run an AI service here or authorize account changes. Its job is to expose missing evidence and prepare a human-owned next action.

Filled synthetic input

SCOPE: one Search campaign, one network, closed month, one reporting timezone and fictional currency, excluding tax, one frozen snapshot. TOTAL_SPEND=12,000; VISIBLE_SPEND=8,400; REVIEWED_SPEND=6,000; OUT_OF_SCOPE_SPEND=900. RUBRIC: paid implementation support; requests only for unrelated jobs are out of scope, ambiguous educational queries remain unreviewed. EVIDENCE: no row duplication reported, source cost normalization to cents documented. PERFORMANCE_CONTEXT: no mature conversion evidence supplied. OWNER: campaign owner to be named; no negative-keyword changes authorized.

Reference output

{"decision":"REVIEW_VISIBLE_ROWS","scope_checks":["Same scope asserted; confirm source headers before action"],"partition":{"reviewed_fit":5100,"reviewed_out":900,"visible_unreviewed":2400,"not_represented":3600},"coverage":{"visible_pct":70,"reviewed_pct":50,"out_among_reviewed_pct":15},"logical_bounds":{"minimum_pct":7.5,"maximum_pct":57.5,"interpretation":"Logical completion bounds, not confidence intervals, waste estimates or savings"},"residual_hypotheses":["Disclosure limits","Scope or export filters","Snapshot or rounding mismatch"],"unsupported_claims":["Hidden queries are irrelevant","Zero conversions prove poor intent","Blocking candidates saves all historical cost"],"next_action":{"action":"Review material remaining visible rows and validate the source boundary","owner":"Campaign owner to be named","review_date":null},"rubric_scores":{"scope":1,"partition":2,"denominators":2,"bounds":2,"outcomes":2,"privacy":2},"questions":["Do both exports share the same network and snapshot?"]}

Evaluate and refine the answer

The six 0–2 rubric criteria are in the copyable prompt. A zero in scope, arithmetic or privacy blocks a recommendation. The example earns only partial scope evidence because the headers were asserted rather than independently inspected. A fluent answer that fills in hidden search terms or promises 6,900 in savings fails the uncertainty and outcome criteria.

For refinement, ask which missing scope field or classification would most affect the next decision. Request that aggregate check, rerun the four-part reconciliation and record the change. Keep ambiguous intent separate and require a human approver for any account action. Never paste sensitive customer queries into a general assistant merely to improve a score.

Sources and limitations

Checked October 22, 2026: Google Ads search terms documentation explains low-activity query omissions and differences from search terms insights. Publication/update date is not displayed. Microsoft Advertising's query-report request documentation, dated November 13, 2024, describes significant-click selection, reporting timezone and completeness controls. These current references establish reporting constraints, not a new October release or measured keyword demand. This original editorial method is not endorsed by either platform and does not infer hidden query identities, causal waste or guaranteed savings.