MMarketing Against The Grain
← All frameworks
Innovation

Three-Window AI Creative Workflow

Separate research, prompt design, and generation to multiply output quality.

Difficulty
Easy
Time to result
~days to results
Steps
5
Confidence
97%

The workflow replaces a single improvised prompt with three specialized AI workspaces. The first performs deep research and produces detailed context such as a style guide, audience analysis, or creative principles. The second uses a conversation with AI to turn the objective and research into a precise generation prompt. The third combines the research and refined prompt to produce the final asset. This separation prevents complex instructions from becoming muddled and lets each stage optimize for a distinct output: knowledge, direction, and execution. The resulting asset is then evaluated against the brief and iterated selectively. The hosts argue that this modest increase in process can produce dramatically higher-quality results than typing a basic request into one window.

Origin

Extracted from Marketing Against The Grain during a demonstration of advanced workflows for ChatGPT image generation.

Core principles

  • 01Rich context improves creative output.
  • 02Prompt design deserves its own collaborative step.
  • 03Separate research from generation.
  • 04Use AI to help construct advanced prompts.
  • 05Treat first outputs as material for iteration.

How to run it

  1. 1

    Research the target

    Run a deep-research project on the brand, style, audience, format, or expert principles you want the output to follow. Ask for detailed, machine-usable findings rather than a short human summary.

    Pro tip Request concrete rules, examples, dos and don'ts, colors, typography, layouts, and messaging principles where relevant.

    Watch out A broad topic summary may not contain enough operational detail to guide generation.

  2. 2

    Package reusable context

    Turn the research into a structured reference that another AI conversation can apply consistently. Preserve the few principles that most strongly determine quality.

    Pro tip If the model loses direction, restate the highest-priority principles separately from the full guide.

    Watch out An excessively long guide can cause the model to lose sight of the most important constraints.

  3. 3

    Co-create the prompt

    Open a separate AI conversation and work back and forth on the generation prompt. Clarify the objective, audience, message, format, and evaluation criteria.

    Pro tip Use a dedicated prompt assistant or a saved prompt template for recurring asset types.

    Watch out Do not assume your first prompt draft expresses the creative objective precisely.

  4. 4

    Generate in a fresh window

    Provide the refined prompt and research context to a new generation conversation. Ask for the specific creative outcome rather than another round of planning.

    Pro tip Keep the research and prompt distinct so the model can recognize reference material versus instructions.

    Watch out Do not omit the research after investing time in creating it.

  5. 5

    Evaluate and iterate

    Compare the output with the brief, identify the weakest element, and request a targeted revision. Continue until the message, visual, and format work together.

    Pro tip Correct one concrete issue at a time, such as the value proposition, headline, or visual metaphor.

    Watch out Repeatedly regenerating the whole asset can discard elements that already work.

In the wild

From Ogilvy research to a HubSpot ad

The hosts researched David Ogilvy's advertising principles, converted them into a detailed style guide, and used that context to generate a B2B ad. They then corrected a confusing headline through conversation until the model reached the clearer message “scale revenue not headcount,” before applying HubSpot's brand style.

The process produced an ad the hosts considered close to production quality in copy, visual concept, branding, and call to action.

Creating a campaign without a research stage

A marketer asks one AI window to make a launch ad using only a product name and receives a generic result. They restart with separate audience and category research, refine a prompt around one value proposition, and generate the ad in a fresh window using both artifacts.

The revised ad has a clearer message, more distinctive visual direction, and fewer rounds of random regeneration.

Common mistakes

Doing everything in one prompt

A single window must simultaneously discover context, interpret the objective, and execute the creative, which commonly produces merely acceptable work.

Letting detail bury the core idea

A large style guide can overwhelm the central value proposition. Re-emphasize the small number of principles that should dominate the output.

Accepting polished nonsense

An asset can look professional while its headline or strategic claim makes little sense. Evaluate meaning separately from visual polish.

Is it for you?

Best for

It is best for marketers producing ads, campaign concepts, landing pages, or other high-stakes creative assets with AI.

Not ideal for

It is unnecessary for trivial requests where speed matters more than consistency or production quality.

From the transcript

if you were just opening up chat GPT and typing stuff in you're going to get an okay result

15:00

you're doing three chat GPT windows instead of one you're going to be 10 to 20 times more successful in terms of quality of output

15:00

you are using so it's like that's a form of AI helping me craft prompts and you're doing the same thing you're going back and…

14:30

From the episode

GPT-4o Just Replaced Your Creative Team (For Free)