Most people brief AI the way they brief a search engine. Type in what you want, hit go, evaluate the result. The output is what you'd expect from that input: generic, polished, forgettable. Then they blame the AI.
The fix isn't a better tool. It's a better brief. Brief the AI the way a creative director briefs a designer, and the output changes immediately. Most of the value of any creative work happens at the brief stage. The same is true for AI work.
WHAT A REAL CREATIVE BRIEF DOES
A real creative brief sets the conditions under which good work can happen. It defines the problem, names the audience, specifies the constraints, identifies the failure modes, and points at the success target. The designer reads it and knows what shape of work to produce. They also know what would be wrong, which is usually the more important information.
Most AI prompts skip every part of this except the ask. "Write me a headline for X." That's not a brief. That's an order. The output is what you get when you place an order with no further instructions: the most likely interpretation of your minimum specification.
THE BRIEF FORMAT
For any creative work you want from AI, structure the brief like this:
Context. What is this for? Who's the audience? What's the moment in their day or year you're trying to land in? Be specific. "Customers" is not specific. "Marketers in their second year of running a brand who are tired of generic strategy advice" is specific.
Constraints. What can't this say? What can't it look like? What competitors have already taken? Where are the third rails? What is the brand's voice? Constraints make the work better. The model will produce safer work without them; constraints push it into a smaller, sharper space.
Failure modes. What would be wrong? Not "wrong" in the sense of incorrect — wrong in the sense of off-brand, off-tone, generic. Tell the model what to avoid. Specifically. "Don't sound like a marketing email" is more useful than "be conversational."
Reference. What's the kind of work that would be right? Paste a paragraph or two of writing that has the texture you want. The model can't read your mind about voice. It can mimic specific examples.
The ask. Now the actual request. With all the above context, the ask is doing a fraction of the work. Most of the shape has already been set.
EXAMPLE
Bad: "Write a headline for our productivity app launch."
Better: "We're launching a productivity app for freelance designers who already use Notion and Things and have hit the limit of what a calendar can do for them. The category is full of bright, friendly, optimistic productivity messaging. We're going to do the opposite: dry, almost-cynical, the kind of voice that respects how tired they are of being marketed at. Avoid: 'unleash,' 'productivity hacks,' 'level up,' anything in the get-stuff-done genre. Reference voice: paste 200 words of Craig Mod or Robin Sloan writing about tools. The ask: 30 headline options that fit this voice and would make a freelance designer click."
The first prompt produces 30 headlines that could be from any productivity app. The second produces something specific.
WHY MOST PEOPLE WON'T DO THIS
It's slower. The brief takes 5-10 minutes to write. You could have written 10 mediocre prompts in that time and looked at 300 mediocre outputs. The brief approach produces fewer outputs but they're closer to usable. For most creative work, that's the better trade.
The other reason: the brief forces you to think before you ask. Most people use AI specifically because they don't want to think. They want the model to do the thinking. The model can't. The model can only produce the most likely answer to whatever you put in front of it. If your input is half-baked, your output will be too.