How to Create a Campaign Plan With AI

Start with a campaign decision spine

Write the campaign as seven linked decisions before drafting prose: the outcome sought, the audience whose behaviour or understanding must change, the barrier to that change, the proposition, the intervention sequence, the evidence that will change the plan, and the accountable owner. If any cell is empty, the document is not ready for channel planning.

The UK Government communication standard says campaign planning should connect an objective, purpose, audience, issues, messages, channels, accountabilities, measurement, timescales, resources, influencers, and partners. Its OASIS lifecycle links objectives, audience insight, strategy, implementation, and scoring rather than treating them as independent headings. Use that structure as a completeness check, not as proof that a plan will work. Government Functional Standard GovS 011: Communications

Sources: [1]

Assemble a source packet before prompting

Give AI a bounded source packet and a source register. Include the approved business objective, prior campaign evidence, audience research with dates and sample notes, offer or policy facts, brand and legal constraints, available channels, delivery capacity, budget range, and the measurement owner’s definitions. Mark each input as approved fact, current observation, working assumption, or open question.

Ask AI first to produce a contradiction table, not a finished campaign. For every proposed statement, require the source item, freshness date, intended use, and any conflict. A useful output might reveal that the brief calls the audience “new managers” while the mailing list only identifies department and tenure, or that a proposed conversion event cannot be observed within the campaign window. Resolve those conflicts with the relevant owner before asking for creative routes.

Draft the strategy as hypotheses

Convert the decision spine into explicit hypotheses. A proposition hypothesis states why the audience would act. A channel hypothesis states why a particular route can reach them in the relevant context. A sequence hypothesis states why one interaction should precede another. A measurement hypothesis states which observable change would be consistent with progress and which alternative explanations remain.

The current Government Communication Service OASIS guide defines a campaign as a planned sequence of communications and interactions intended to deliver a defined and measurable outcome. It also distinguishes outcomes from outputs and calls for objectives that are measurable and focused on attitudes or behaviour where appropriate. GCS guide to campaign planning

AI can generate three genuinely different strategy routes, but require each route to name its mechanism, dependency, disconfirming evidence, resource requirement, and principal risk. Reject alternatives that merely rename the same channel mix.

Sources: [2]

Build the campaign control sheet

Use one control sheet as the evidence artifact behind the narrative plan. Give every row an identifier so later changes can be traced.

For paid activity, keep the business objective separate from platform configuration. Google Ads documentation notes that campaign type and optimization depend on the selected objective and conversion goals, and that stale or irrelevant goals should be removed. That is a configuration dependency, not evidence that the selected objective or conversion represents business impact. Google Ads campaign objective guidance

• Decision ID and accountable owner
• Approved objective and baseline date
• Audience segment and evidence source
• Barrier or opportunity, with confidence level
• Proposition and message promise
• Intervention and channel hypothesis
• Required asset, approval, and delivery owner
• Earliest observable signal and decision date
• Cost or capacity constraint
• Stop, repair, or scale condition

Sources: [3]

Worked example: a deadline-preparation campaign

Suppose an internal operations team needs regional managers to submit a readiness packet before a system migration. The source packet shows 180 managers in scope, 112 complete packets last cycle, recurring confusion about one evidence field, a four-week window, and no reliable way to attribute action to a single message.

The control sheet sets the outcome as “complete, reviewable packets received by the deadline,” not email opens. It segments managers by prior completion status and identified blocker. The proposition is faster first-pass acceptance using a worked example and office-hour route. The sequence is: manager briefing, example packet, targeted reminder after a completeness check, then escalation through the existing governance owner. The earliest signal is use of the completeness check; the decision measure is valid packets by segment. The plan records that a rise cannot be attributed to any one channel because timing, manager follow-up, and system prompts overlap.

AI helps classify recurring questions, draft message variants from approved facts, and scan the plan for owner or date gaps. The campaign owner approves the segmentation, message promise, escalation rule, and any use of personal data. A measurement owner freezes denominator rules before launch.

Run a pre-launch consistency test

Read the control sheet in both directions. Forward, every source-supported audience barrier should lead to a proposition, intervention, owner, and measure. Backward, every asset and channel should trace to a defined decision and audience need. Recalculate totals, test dates against delivery capacity, and verify that every promised action has an operational owner.

Then run three failure scenarios: the intended signal is unavailable, the primary channel is delayed, and a key assumption is false. Record the fallback, the person who can authorize it, and the point at which the campaign should pause rather than improvise. This turns contingency planning into an executable decision rather than a generic risk paragraph.

Keep the claims and AI role bounded

Label synthetic examples, forecasts, and untested hypotheses. Do not present platform-reported conversions as incremental impact, channel activity as behaviour change, or a polished AI draft as audience evidence. A campaign plan can make responsibilities and tests explicit; it cannot ensure performance.

Two Prune can provide a bounded assessment environment in which a participant creates and verifies a campaign-plan artifact from supplied evidence. It does not run advertising platforms, approve campaigns, certify marketing compliance, prove business impact, or make employment decisions. Any use of assessment evidence remains subject to human interpretation, evidence strength, and the limits of the simulated task.