Iterative AI Creative Briefing
Brief, critique, and refine AI creative until the concept becomes usable
- Difficulty
- Easy
- Time to result
- ~days to results
- Steps
- 6
- Confidence
- 98%
This method treats an image-generating AI like a creative partner working through normal design revisions. Start with the campaign idea, product context, format, and desired message, then inspect the first output for misunderstandings. Convert every failure into clearer direction: explain wordplay, identify the intended meaning of ambiguous terms, remove unnecessary elements, and describe the desired scene precisely. Continue briefing and critiquing rather than abandoning the concept after a weak first result. The mechanism is a feedback loop in which each generated image reveals missing information in the brief. That loop can compress early concept development from weeks to hours, although human designers may still be needed for typography, brand fidelity, legal review, and final production.
Origin
Extracted from Marketing Against the Grain during Kieran Flanigan's demonstration of creating Zapier campaign concepts with ChatGPT, Bing, and image-generation tools.
Core principles
- 01Treat AI as a creative partner, not a one-shot generator
- 02Specific visual direction improves creative relevance
- 03Explain puns, metaphors, and brand context explicitly
- 04Use each output as feedback for the next brief
How to run it
- 1
Frame the campaign
State the product, audience, campaign objective, and creative formats you want the AI to produce.
Pro tip Request several placements, such as a billboard, bus advertisement, and digital ad.
Watch out A catchy line without product context can send the model toward the wrong visual meaning.
- 2
Explain the concept
Clarify the intended meaning of puns, metaphors, product names, and brand references.
Pro tip Describe what an unfamiliar term refers to instead of assuming the model understands it.
Watch out Ambiguous words may be rendered literally.
- 3
Generate a first revision
Ask the AI to produce an initial concept and treat it as a diagnostic draft rather than final work.
Pro tip Generate multiple variants or compare more than one model.
Watch out Do not evaluate the entire method from one weak output.
- 4
Critique concrete elements
Identify exactly what failed, including composition, text, product representation, mood, or visual clutter.
Pro tip Turn each criticism into a positive instruction describing what should appear instead.
Watch out Generic feedback such as 'make it better' gives the model little usable guidance.
- 5
Increase visual specificity
Describe the scene, subject, action, hierarchy, and omitted elements in explicit terms, then regenerate.
Pro tip Specify where the product benefit should appear visually.
Watch out Generated text may still require manual replacement even when the image concept works.
- 6
Refine and hand off
Repeat the loop until the concept is useful, then pass the strongest direction to production or testing.
Pro tip Use the generated concept as a fast creative brief for a human designer.
Watch out Review final assets for brand, copyright, accessibility, and factual issues.
In the wild
Kieran began with the line “Zapier automates the sheet out of work,” but early generations confused spreadsheets with bed sheets and even treated Zapier like washing-up liquid. He then specified that sheets should float in the sky while a happy person pressed Automate on a laptop. The resulting billboard concept moved much closer to something usable.
→ A vague pun became a credible visual concept after several increasingly specific briefs.
Common mistakes
Stopping after the first revision
The first output is often a test of the brief rather than evidence of the model's full capability. Use its errors to identify missing instructions.
Leaving wordplay unexplained
The model may interpret a pun literally or attach it to the wrong product category unless its intended meaning is stated.
Treating mockups as finished assets
AI concepts can contain malformed text, incorrect logos, or unsuitable details that require human production work.
Is it for you?
Best for
Marketers who need rapid billboard, display, social, or campaign mockups before involving a design team.
Not ideal for
Teams expecting polished, brand-compliant production assets from a single vague prompt.
From the transcript
“but just like you would any designer or person the first rev is not going to be good”
“turns out when you actually brief the AI like a marketer it actually does something pretty good”
“AI is a good creative partner if you brief brief it keep itting just like you would with hum by the way you have lots…”
From the episode
I Used ChatGPT To Create A Marketing Plan In 30 Minutes