Every week someone shows me a screenshot of ChatGPT writing something terrible and says "see, AI can't do copy."
Then I look at the prompt. It says "write a LinkedIn post about our new product, make it engaging."
That's not a prompt. That's a sticky note. If you handed that to a junior copywriter on their first day, you'd get the same beige paragraph back, plus a rocket emoji. The model isn't the problem. The brief is.
We solved this problem decades ago
Agencies have been fighting vague requests forever. The fix was the creative brief: one page that forces the person asking for work to decide what they actually want before anyone writes a word.
Nobody loves writing briefs. They're a bit boring and they make you commit. That is exactly why they work, and exactly why they're the best prompt template ever made. A model, like a creative team, can only be as sharp as the thinking you hand it.
The brief, line by line, as a prompt
Here's the structure I use. It's the old agency brief with almost nothing changed.
- What are we making? Format, length, channel. "A 120-word LinkedIn post" beats "some content."
- Who is it for? One real person, not a demographic. "A head of marketing at a 200-person SaaS company who has been burned by an agency before."
- What do they think now? The belief you're up against. "AI tools are toys that create more work than they save."
- What do we want them to think? The shift, in one line.
- The single-minded proposition. One thing. Not three. If you can't write it in a sentence, you're not ready to prompt.
- Why should they believe it? Proof. Numbers, a customer, a demo, a fact. The model cannot invent this and you should never let it try.
- Tone. Two or three words, plus one example of writing that nails it and one that misses.
- Mandatories. The link, the legal line, the words you never use.
Paste that in, and the same model that gave you beige will give you something you can actually edit. Not finished. Editable. That's the realistic goal.
The proposition is the whole game
If I had to keep only one line from the brief, it's the single-minded proposition.
Most weak prompts, like most weak briefs, ask for everything at once. "Show that we're innovative, affordable, trusted and fun." A human creative would push back on that. A model won't. It will happily write you a paragraph that says all four things and means none of them.
Force yourself to pick one. Innovative or affordable. Then make the model prove it, using the proof you gave it.
Things the model should never write for you
There are parts of the brief I don't let AI touch, and I'd suggest you don't either.
- The proposition itself. That's a strategic decision. Asking a model to pick your positioning is like asking a printer to pick your font. It will choose something. It will be Arial.
- The proof. If the claim isn't true, the copy is a liability, however good it sounds.
- The final read. Out loud, by a human, before anything goes out with your name on it.
Everything else is fair game. Thirty headline options, ten openings, a shorter version, a version for a different audience, a sanity check against your tone examples. That's where the hours come back.
Try it on your next request
Next time you open a chat window, don't type the request. Type the brief.
It will take you ten extra minutes. You'll find out that half the time you didn't actually know what you wanted, which is useful to learn before the campaign goes live rather than after.
Old-school marketing had a name for the people who skipped the brief and went straight to execution. We called them the reason for the second round of revisions.