Generative AI can produce a campaign asset before the approval meeting has finished discussing the campaign asset.
That speed is useful. It also creates a new production risk: work can look finished before it has been properly judged.
Visual polish is not evidence of strategic clarity, factual accuracy or cultural sense. A fluent model can produce a confident mistake wearing excellent lighting.
This checklist is designed for the point between “the output is impressive” and “the public should see it.”
Use it twice: once on the first near-final route, while large changes are still affordable, and once on the exact files, captions, links and destinations scheduled for release. A review of the master visual is not a review of the campaign.
1. Strategic fit
Before inspecting fingers, reflections and typography, ask whether the idea deserves to exist.
- Is the audience specific?
- Is the business problem clear?
- Does the work express one strategic idea?
- Is the intended memory identifiable?
- Does the treatment help the audience, or mainly advertise the tool?
- Is the requested action proportionate?
If the strategic answer is vague, visual correction will produce a cleaner version of the wrong work.
2. Truth and evidence
Review every factual element:
- Product features and appearance
- Performance claims
- Prices, locations and availability
- Customer or expert representations
- Statistics and quoted material
- Before-and-after implications
- Disclosures and qualifications
Distinguish concept visuals from documentary evidence. Do not generate a customer, result, facility or product capability and allow the audience to assume it existed.
3. Visual continuity
AI errors often hide inside change.
Check across every frame and format:
- Character identity, age and clothing
- Product dimensions, controls and branding
- Object count and position
- Lighting direction and time of day
- Architecture and geography
- Reflections and shadows
- Text, logos and packaging
- Left/right orientation
Create a continuity sheet before generating final variants. The more assets you produce, the more important a fixed reference becomes.
4. Physical plausibility
Inspect anatomy and object behaviour without assuming “AI weirdness” is charming.
Look for:
- Hands, teeth, eyes and joint movement
- Contact between people and objects
- Liquid, fabric, smoke and hair behaviour
- Weight, balance and gravity
- Product assembly and use
- Impossible camera movement
- Background people or objects changing mid-shot
Stylised work can break physical rules deliberately. The key word is deliberately.
5. Audio and language
Review sound separately from the visual excitement.
- Is pronunciation correct?
- Does speech sound natural for the person and context?
- Are names, places and category terms accurate?
- Does lip movement support the language where necessary?
- Are music and voice rights clear?
- Are captions accurate and timed?
- Is the message understandable without audio?
Synthetic speech can sound technically smooth and socially misplaced. Local review matters.
6. Cultural and representational sense
Ask people with relevant lived context to review:
- Casting and identity
- Clothing and setting
- Gesture and social relationships
- Humour
- Religious or cultural symbols
- Class and regional representation
- Stereotypes and exclusions
- Whether the scenario feels observed or assembled
Do not make one reviewer responsible for representing an entire group. Use specific questions and allow disagreement to inform risk.
7. Rights, permission and disclosure
Confirm:
- Source material and reference rights
- Model and voice permissions
- Music, typeface and asset licensing
- Trademark and product depiction
- Contractual restrictions
- Platform disclosure requirements
- Whether synthetic treatment should be identified
Record which tools, sources and approvals were used. Documentation becomes more valuable when a client, platform or audience asks later.
8. Brand coherence
Evaluate more than colour and logo.
- Does the work carry the brand’s point of view?
- Does it strengthen an existing memory structure?
- Is the tone recognisable?
- Are category clichés doing the creative work?
- Could a competitor publish the same asset after replacing the mark?
- Does every variation still belong to one campaign world?
Use brand anti-references to catch high-probability sameness early.
9. Accessibility and format
Check the actual placements:
- Caption accuracy and readability
- Text size, duration and contrast
- Alt text or meaningful description
- Flashing and motion intensity
- Safe areas and crops
- Small-screen legibility
- Comprehension without sound
- Reduced-motion alternatives where applicable
A beautiful master frame can become unreadable after platform cropping. Review the delivered asset, not only the artboard.
10. Operational readiness
The campaign journey must be ready for the attention:
- Landing page continues the promise
- Offer and inventory are correct
- Customer-facing teams know the campaign
- Questions and complaints have owners
- Corrections can be issued quickly
- Files and versions are clearly controlled
- Performance signals are defined
An AI-native campaign can still have a very traditional handoff problem.
Use a red–amber–green review
Run the checklist in a shared record, not in scattered comments. Every amber or red finding needs the asset, the issue, the owner, the required decision and the retest date. That turns review from collective concern into accountable work.
| Status | Meaning | Action |
|---|---|---|
| Green | Verified and ready | Approve |
| Amber | Judgement or clarification required | Named owner resolves |
| Red | Factual, rights, safety or strategic failure | Do not publish |
Do not average red issues into an acceptable score. One fabricated claim or unlicensed voice cannot be cancelled out by excellent composition.
Name the approvers
At minimum, identify owners for:
- Strategy and brand
- Factual/product accuracy
- Creative craft and continuity
- Cultural or market review
- Rights/legal review where required
- Final publishing approval
For a fuller workflow, use human approval gates for AI production .
The final five questions
Before release, ask:
- Is the idea worth the audience’s time?
- Is every consequential claim true and supportable?
- Would we be comfortable explaining how this was made?
- Can we identify who approved the final work?
- Does AI improve the campaign rather than become the campaign?
If any answer requires a very long pause, the work is not ready.
Read A strong prompt is a compressed strategy for the briefing system behind this checklist.
Building an AI-assisted campaign? We can help design the strategy, production and review system .