AI Campaign Plan Verification Checklist

Set the verification packet

Verify a controlled packet rather than the prose in isolation. Include the approved brief, audience evidence, baseline, plan, channel assumptions, delivery schedule, budget or capacity limit, measurement specification, risk register, and approval record. Give each item an identifier, owner, date, and status: approved fact, observation, assumption, or unresolved question.

The current Government Communication Service standards require use of OASIS for campaign planning and delivery and call for evaluation through the GCS evaluation cycle. That establishes a useful planning and evaluation frame, but it does not show that a particular plan is complete or effective. Government Communications professional standards

Ask AI to build a source-to-claim table from this packet. For every material statement in the plan, record the supporting item, the exact passage or field, its date, and any contradiction. A claim with no traceable source becomes an assumption or is removed.

Sources: [1]

Check the objective and audience decision

Confirm that the objective describes an observable audience or operating change, not an activity count. Record the baseline, denominator, time window, decision owner, and the point at which the result will be judged. If the campaign has several objectives, name their priority and identify conflicts between them.

Then test the audience definition. Each segment should have a documented reason for being distinct, an evidence source, a reachable route, and a relevant barrier or opportunity. Reject labels that cannot be reproduced from available data. Check that personal data use, exclusions, and contact permissions have an accountable review route.

The GCS Evaluation Cycle links objectives, audience insight, strategy, implementation, and evaluation, and recommends planning evaluation at the beginning. Use that linkage to test whether the stated objective can actually be observed with the proposed evidence. GCS Evaluation Cycle

Sources: [2]

Trace strategy to evidence and choices

For every proposition, message, channel, and sequence, ask four questions: What evidence supports this choice? What mechanism is expected? What competing explanation exists? What observation would trigger a change? A channel preference or polished phrase is not audience evidence.

Check that the plan explains why one route was chosen over plausible alternatives. Require resource needs, dependencies, approval constraints, delivery capacity, and the principal failure for each route. If AI generated options, retain the evaluation criteria and human decision rather than only the selected wording.

Use the OASIS planning model as a structural cross-check across objectives, audience insight, strategy, implementation, and scoring. The framework can reveal a missing connection; it cannot determine whether the underlying audience evidence is sound. Campaign planning with the OASIS framework

Sources: [3]

Verify implementation and ownership

Walk every planned intervention through delivery. Confirm the asset, channel, audience rule, dependency, accountable owner, approver, earliest start, deadline, capacity, and fallback. Dates must be consistent with approval and production lead times. Costs and totals should be recalculated from their source values.

Use a two-column responsibility test. The left column lists every decision, asset, data flow, approval, measurement task, and escalation. The right column names one accountable owner and any contributors. Empty rows are release blockers. Rows with several owners need a single decision authority or an explicit handoff rule.

Test the measurement design

Confirm that outputs, audience responses, and intended outcomes remain separate. Each measure needs a definition, numerator, denominator, collection method, time window, owner, quality check, and interpretation limit. Platform events should not silently become business outcomes.

Recalculate the example metrics using raw values. Test missing data, duplicated events, delayed reporting, audience overlap, and changes to the eligible population. Record confounds such as simultaneous manager follow-up, product changes, seasonality, or existing communications. The plan should state which decisions the evidence can inform and which causal conclusions remain unavailable.

Worked check: deadline-preparation campaign

Consider a plan intended to improve the completeness of 180 readiness packets before a migration. The verifier finds that the objective and baseline are defined, but the audience table calls 68 people “previously incomplete” while the source register only identifies 64. The email schedule also assumes a legal approval two days before the approval owner is available. Finally, the dashboard counts link clicks while the plan calls them evidence of completed packets.

The checklist disposition is repair, not approval. The campaign owner reconciles the four-person segment difference, moves the first send after the approval date, and changes the dashboard decision measure to valid packets received by segment. Link clicks remain a delivery diagnostic. The plan adds a fallback office-hour route if the completeness checker fails and records that any improvement cannot be attributed to a single intervention.

AI helped compare tables, dates, and definitions. It did not decide which population record was authoritative or approve the revised schedule.

Record the disposition and residual uncertainty

Close the check with one of four dispositions: pass, repair, hold, or retire. Record failed checklist items, the evidence reviewed, repairs made, unresolved uncertainty, accountable approver, approval time, and the version approved. A pass means the specified checks were completed; it is not proof that the campaign will achieve its intended outcome.

Copy this compact release record into the plan:

Two Prune can place a participant in a bounded assessment where they verify a campaign-plan artifact against supplied evidence. It does not approve campaigns, operate channels, certify compliance, prove business impact, or make employment decisions. Assessment evidence requires human interpretation within the limits of the simulated task.

• Packet version and verification date
• Objective, audience, strategy, implementation, and measurement checks
• Calculation and source-trace results
• Privacy, brand, legal, and operational approvals where applicable
• Failure scenarios and authorized fallbacks
• Residual uncertainty and prohibited interpretations
• Final disposition, owner, and next review date