Marketing prompts is one of those topics that sounds simple until you actually try it. After a few weeks of running real workflows through ChatGPT and Claude, the patterns that move output quality forward become obvious — and they're not the ones most tutorials lead with.
This guide is a distillation of what we've learned shipping marketing campaigns day to day. Every example is something we've copied, edited and shipped — not a hypothetical.
Why this matters
The difference between a prompt that almost works and one that ships isn't intelligence on the model's side — it's clarity on yours. ChatGPT and Claude can reliably produce great output when the inputs are concrete; it cannot read your mind.
- A defined role makes the model write with consistent voice.
- Explicit constraints prevent the model from optimizing for the wrong goal.
- A clear output format keeps you from rewriting the response by hand.
The pattern that works
We landed on a four-part pattern that gives us a strong first draft on the first run with ChatGPT and Claude. It's not novel — it's just the discipline of doing all four every time.
1. Assign a role
Tell the model who it is. "You are a senior marketing campaigns strategist" is a different starting point than no role at all, and the output reflects that.
2. Provide context
Give it the audience, the goal, and any constraints. Two extra sentences here save twenty rounds of follow-up edits later.
3. Define the output
Bullets, headings, JSON, a table — pick the shape and ask for it explicitly. The model will use whatever you ask for.
4. Iterate, don't rewrite
Run the prompt once, then refine with follow-up messages. Editing the prompt itself between runs creates drift that's hard to debug.
A worked example
Here's a stripped-down version of the prompt we use as a starting point for marketing campaigns. Notice the role, the context block, and the explicit output format.
You are a senior marketing campaigns expert helping me ship a polished draft.
Context:
- Audience: {audience}
- Goal: {goal}
- Constraints: {constraints}
Task:
Produce a structured marketing prompts that opens with a sharp hook, communicates the core value in plain language, and ends with a single call-to-action.
Output format:
- Subject / headline
- Body (3 short paragraphs max)
- Call-to-action
- Two alternative variants I can A/B test“Good prompts read like a brief, not a wish. The model isn't guessing what you want — you've already told it.”
Common pitfalls
Most of the time, when a prompt produces mediocre output, the bug isn't in the model — it's in one of these four places:
- Vague placeholders that were never replaced with real values.
- Asking for too many variants at once (quality drops after about three).
- Skipping constraints, so the model optimizes for length instead of clarity.
- Editing the prompt mid-conversation instead of using a follow-up message.
How models compare for this task
Different frontier models have different sweet spots. The table below is what we've observed over the last few months of testing.
| Model | Best for | Watch out for |
|---|---|---|
| ChatGPT | Fast, structured drafts | Generic tone without explicit voice cues |
| Claude | Long-form writing, nuanced tone | Slower on short tasks |
| Gemini | Research and multimodal inputs | Needs grounding to stay specific |
| Grok | Punchy, opinionated copy | Less reliable for long structured output |
Putting it into practice
Take the pattern above and apply it to one task you actually do every week. Within a few iterations you'll have a reusable prompt that consistently produces output you can ship with light editing — and you'll spend more of your time on judgment, less on rewrites.
What to read next
If this resonated, the related articles below pick up where this leaves off. The related prompts are the exact templates we ship — copy any of them and adapt the placeholders to your work.
