If you read enough LinkedIn posts about AI, you’ll come away thinking the game is about clever prompting. Get the “magic words” right and the model unlocks a hidden chamber of brilliance. Get them wrong and it collapses into gibberish. This isn’t how large models actually work, and a new study on visual generative AI in advertising helps clarify the real issue – AI needs guidance, but it doesn’t need to be micromanaged.

The researchers compared three types of ads.

• Human-designed ads by ad experts.  These were separately rated by other ad industry folks to affirm quality.
• Human-designed by experts but then “improved” by GenAI.
• And GenAI-created from scratch.

The results weren’t ambiguous.

• AI-made edits to human ads consistently made them worse – lower purchase caused by lower realism and lower aesthetic quality.

When the AI was forced into procedural, prescriptive instructions that boxed it into a corner – add a face, add scenery, add an illustration, keep the layout, don’t touch anything else…it became an exercise in patching a design it never understood.

But when the AI was allowed to create from a strategic brief instead of a laundry list, the whole picture flipped.

• The AI-generated ads outperformed the human ads

The deeper point is not about advertising, it’s about how constraints work.  In our experience and in these experiments, high level direction amplifies AI, low level instruction cripples it.

This aligns with a broader pattern emerging across creativity research. Multiple studies now show LLMs outperform humans on open ended creative tasks by generating more original ideas that score higher on divergent thinking.  Specifically, LLM outputs scored higher than human work on the Torrance Tests of Creative Thinking, long considered the gold standard of human creativity assessment. Anotherline of research showed LLMs surpass humans on metaphor generation, naming, design concepts, and ideation when given high level direction.

The emerging pattern is clear, creativity is not small bore, it’s holistic. You can’t prompt a system to produce creativity by telling it to “improve this part but not that part.”

If you are working with humans or machines and you manage creative output the way you might manage an intern with endless bullet points, superficial guardrails, prescriptive language requirements, hyper specific formatting, etc. then it’ll likely be low coherence, generic work.

The alternative is not “let the AI run wild.” The winning pattern: give direction at the strategic level, not the tactical one. In our world that means,

• Define the reader’s values, goals and motivations
• Identify the moral framing that fits the reader
• Have a coherent outline for your appeal – ours is Story, Need, Solution, Ask.

A strong AI creative brief looks like strategy, a weak one looks like compliance.

AI does not beat humans at creative because it copies, it does so because it operates differently that most humans where we tend to edit ourselves too early, converge too quickly, cling to familiar structures etc.  AI does none of that unless you force it to with petty constraints.

Kevin