Before generating
Set the context
Make sure the workspace, brand, product, and audience context are in the right place. Durable facts belong in context, not only in the prompt.
Pick the variable
Decide what you are testing: hook, audience, value prop, format, offer, visual style, or CTA.
Attach visual references
If the output must look like something, attach it. Product shots, screenshots, winning ads, and layouts matter more than adjectives.
Prompt like a creative brief
A strong prompt includes:- Audience: who the ad is for.
- Angle: the pain, desire, or value proposition.
- Format: static, UGC video, app demo, comparison, testimonial, offer-led.
- References: which assets or winners to use.
- Guardrails: what must not change.
- Test variable: what should vary across the batch.
“Create five Meta feed statics for tired parents buying our sleep supplement. Keep the product packaging exact. Use the attached testimonial as the hook source. Test five different first-line pains, but keep the layout clean and premium.”
How to fix bad outputs
It feels generic
It feels generic
Add real customer language, reviews, objections, and competitor examples. Generic inputs produce generic ads.
It is off-brand
It is off-brand
Add approved ads, visual references, logo rules, brand colors, and examples of what not to do.
The product is wrong
The product is wrong
Use a narrower reference set. Attach the exact product or UI asset and say what must stay fixed.
It changed too much during editing
It changed too much during editing
Ask for one surgical change at a time. If only the hook should change, say that everything else should stay fixed.
Save what works
When a creative works, save it as a reference. The best brief for the next ad is a previous winner.Reference images & winning ads
Build the visual layer that keeps future outputs sharp.