Prompt Before Production
Refine cheap text prompts before generating expensive creative assets
- Difficulty
- Easy
- Time to result
- ~days to results
- Steps
- 6
- Confidence
- 98%
Separate inexpensive thinking from expensive production. Begin in a capable text model, where the product, audience, advertising practices, creative angles, and constraints can be researched and converted into a detailed production prompt. Inspect that prompt section by section rather than accepting the first draft, replacing weak concepts and adding explicit creative direction before opening the asset-generation tool. Only after the written specification is coherent should the marketer generate ads. This ordering makes errors visible while they are still cheap to correct, limits waste from repeated image generation, and creates a stronger baseline for subsequent visual iteration. It is especially valuable when every generation consumes paid credits or takes meaningful time.
Origin
Extracted from Marketing Against The Grain during a demonstration of preparing Replit ad-generation instructions in Claude before producing the ads.
Core principles
- 01Fix instructions before paying for production
- 02Use text iteration to expose weak assumptions cheaply
- 03Treat the prompt as a creative specification
- 04Move to asset generation only after reviewing every concept
How to run it
- 1
Define the deliverable
Specify the product, advertising platforms, number of variants, audience, and intended outcome.
Pro tip Make every requested output count explicit.
Watch out A vague deliverable will produce an unfocused research and production prompt.
- 2
Research in text
Use a capable text model to investigate the product, positioning, customer benefits, and platform-specific advertising practices.
Pro tip Ask the model to show the resulting product context and creative rationale inside the prompt.
Watch out Do not assume the generation tool already understands the product accurately.
- 3
Draft the production specification
Turn the research into one detailed prompt containing brand guidance, formats, angles, proof points, and quality standards.
Pro tip Structure it by platform and individual ad concept.
Watch out Do not move directly from a one-line idea to paid asset generation.
- 4
Review section by section
Read every proposed concept and highlight instructions that conflict with your taste, brand, or campaign goal.
Pro tip Spend deliberate review time even if the draft initially looks comprehensive.
Watch out Length is not evidence that the prompt is good.
- 5
Refine cheaply
Iterate in text until the concepts and constraints are production-ready.
Pro tip Use precise corrections such as changing an angle, visual style, or level of polish.
Watch out Do not burn generation credits fixing problems that could have been corrected in prose.
- 6
Generate the assets
Transfer the approved prompt into the specialized generation tool and produce the first creative batch.
Pro tip Preserve the reviewed prompt as a reusable campaign specification.
Watch out The improved prompt reduces iteration but does not eliminate the need to inspect creative output.
In the wild
A marketer needs three LinkedIn, three Instagram, and three Google ads for an AI customer-support product. They first ask a text model to research the product and platform conventions, then review and rewrite the proposed angles before moving the approved prompt into the ad-generation tool.
→ The expensive generation phase starts from a researched, reviewed specification rather than a rough idea.
A founder drafting a product video resolves the positioning, scene sequence, brand constraints, and prohibited imagery in text before requesting any video renders.
→ Fewer paid renders are discarded because conceptual problems were corrected before production.
Common mistakes
Generating from the first idea
Starting production with an unreviewed prompt transfers every ambiguity into an expensive asset.
Confusing detail with quality
A very long prompt can still contain weak angles, inaccurate claims, or unsuitable creative direction.
Skipping human taste
Research and best practices provide a baseline, but the marketer must still replace concepts that do not fit the desired campaign.
Is it for you?
Best for
Marketers producing costly AI-generated images, videos, ads, or other multimedia assets.
Not ideal for
Tiny text-only tasks where generation is effectively instantaneous and free.
From the transcript
“I want to underline that it's really important to take this step of working to refine and iterate your prompt before you go and actually…”
“So I'm gonna save a lot of time and I'm gonna save a lot of money if I iterate on the prompt in text before…”
“Prompts are much cheaper. Iterating on the ads themselves through AI is expensive.”
From the episode
Can AI Actually Make Good Ads? Replit Ad Maker Review