Concrete criteria for assessing strategy, originality, craft, brand, and risk in AI-generated work before approving it or scaling it with sound judgment.
First evaluate whether it addresses the right problem
A piece can be beautiful and fail because it does not answer the brief. The first criterion examines the audience, tension, objective, promise, and expected action. The question is whether it expresses an idea capable of moving the defined decision, rather than whether it looks like advertising. Also check that claims are supported and the execution does not shift the focus toward a technology demonstration. If strategic relevance fails, polish should not save it.
Abundant outputs make judgment more important. A useful rubric separates strategy, idea, execution, identity and risk; defines critical failures and requires decisions to be explained. It does not seek to turn creativity into a mechanical score, but to reveal why one direction deserves more work and another must stop.
Distinguish visual novelty from conceptual originality
AI can produce surprising surfaces that conceal a familiar idea. The rubric asks whether there is an unexpected yet understandable connection, whether the brand could own the territory and whether the asset adds more than style. Comparing directions without revealing who or which tool generated them helps reduce novelty bias. Recording similarities to references also helps prevent a direction from depending too heavily on someone else's aesthetic.
Review craft with a medium-specific checklist
Images need composition, anatomy, light, continuity, legibility and detail; video adds motion, timing, audio and consistency between frames; text demands structure, precision, rhythm and voice. A shared rubric should not erase those differences. Each medium retains technical criteria while being evaluated within the complete family. An asset approved for a feed can fail outdoors or on a small screen if it is simply reduced in size.
Make identity a verifiable criterion
Brand review checks positioning, distinctive codes, tone, hierarchy and relationships with the visual system. Adding colors and a logo is not enough. The question is whether the idea would feel proprietary even without a signature and whether an adaptation preserves what makes it recognizable. Specific deviations are recorded to update instructions, references or components; saying it does not feel like Wiwo teaches neither the system nor the team.
Separate critical failures from improvement opportunities
False information, unauthorized use of identity, discriminatory content, sensitive data, or regulatory noncompliance can stop a piece regardless of its average score. Other aspects, such as pacing or brand intensity, allow iteration. The rubric must state these gates and preserve the reason for the result. By analyzing rejections and corrections by category, the team discovers where the workflow needs better sources, controls, or training.
A rubric you can use in your next review
Score from 1 to 5. Multiply each score by its weight and divide by 5. An unresolved rights or accuracy criterion prevents approval, even if the average is high.
| Criterion | Weight | Fictional example: a piece for a café |
|---|---|---|
| Clarity | 25% | 4/5: the invitation is understood |
| Brand voice | 20% | 4/5: approachable tone, adjust the ending |
| Relevance | 20% | 3/5: local context missing |
| Execution quality | 20% | 4/5: review contrast |
| Rights and accuracy | 15% | 5/5: authorized materials in this example |
Illustrative example: 79/100. With a proposed threshold of 80, it returns to editing. The threshold must be agreed by brand; this assessment does not correspond to a real client.
Download the rubric and evaluated example (CSV) ↓