Most people prompt AI the way they’d search Google: short phrase, vague intent, hope for the best. It works often enough that nobody questions it. But there’s a different framing that gets noticeably better output, consistently — borrowed from somewhere most people wouldn’t expect: the agency client brief.
A client brief isn’t a question. It’s a structured handoff. It tells the person doing the work who the audience is, what success looks like, what constraints exist, and what tone is expected. Give AI that same structure and it stops guessing at the most statistically average interpretation of your request and starts producing something shaped around your actual situation.
Why Vague Prompts Produce Average Outputs
The mechanism matters here. When you ask a model something underspecified — “write me a cover letter for a marketing job” — it fills in every missing variable with the most generic plausible option. Generic seniority, generic tone, generic emphasis. Not because it can’t do better, but because it has no reason to choose differently.
Every undefined parameter defaults to the mean. The output is technically correct and almost immediately forgettable.
Adding positive adjectives doesn’t help. “Write a compelling, natural-sounding cover letter” doesn’t reduce the ambiguity — it just decorates the same vague request. What actually moves the needle is reducing the number of things the model has to guess.
What a Brief-Style Prompt Actually Contains
You don’t need a formal template. You need to answer, briefly, the questions a thoughtful collaborator would ask before starting. In practice that usually means four things:
- Who’s reading this, specifically? Not “a general audience” — a hiring manager at a mid-size SaaS company who has already read forty applications today. Or a non-technical client who will forward your email to their finance team. The more specific the reader, the more targeted the output.
- What should they do or feel at the end? Not “be persuaded” — click reply, approve the proposal, feel confident about the timeline. A concrete desired outcome shapes word choice and emphasis in ways vague goals don’t.
- What’s the one thing this must not do? The negative constraint is often more useful than the positive one. “Don’t make it sound like we’re desperate for the business” cuts off a whole class of mediocre outputs immediately, more reliably than any amount of “make it sound confident” would.
- What exists already that this should match or reference? A sample of your own writing, the job description you’re responding to, a previous email in the thread. Context grounding produces coherence; asking the model to generate from scratch produces generic.
None of this is complicated. It’s just slower than firing off a three-word prompt, which is why most people skip it.
A Concrete Example of the Difference
Here’s the kind of gap this actually creates.
Vague prompt: “Write a LinkedIn post about my new freelance consulting business.”
Brief-style prompt: “I’m writing a LinkedIn post to let my existing network know I’ve gone independent as a fractional CFO. My audience is former colleagues and past clients — people who already know my work, so I don’t need to explain my credentials. I want them to reach out if they know someone who needs interim finance leadership. Tone should feel like a personal update, not an announcement. One thing it must not do: sound like a victory lap or like I’m positioning this as leaving-my-last-job content. Under 150 words.”
The output gap between those two prompts isn’t subtle. The first gets you something that reads like every other “excited to announce” post on LinkedIn. The second gives the model enough to actually write toward a specific effect for a specific reader.
Where This Pays Off Most
Not every task is worth the setup time. Summarizing an article, fixing a sentence — a short prompt is fine. The brief approach earns its overhead on anything where the output will be read by someone who matters, or anything you’d otherwise spend real time editing.
Emails to important clients. Proposals. Content you’re publishing. Anything where the wrong tone or a missed constraint means a second draft.
It also compounds on recurring tasks. If you write a brief-style prompt that works well for a given format — client-facing project updates, for example — save it. The second run costs none of the setup time. That’s the actual productivity gain, not the minutes saved on a single session.
One Honest Caveat
Brief-style prompting doesn’t fix all failure modes. If the model has a gap in domain knowledge, more context about your audience won’t bridge it. And if you don’t actually know what you want — if the vagueness in the prompt is really vagueness in your own thinking — the structure of a brief will expose that rather than solve it. You’ll write “what should they feel at the end” and realize you don’t have a clear answer. That’s useful information, but it’s not the model’s problem to solve.
There’s also a real risk of over-constraining. Lock down every parameter and you occasionally rule out an approach that would have been better than anything you’d anticipated. Leaving one dimension deliberately open — “I don’t have a strong opinion on format, use your judgment” — sometimes produces something worth keeping.
The Practical Starting Point
Pick the next prompt you’d normally fire off in one sentence. Before you send it, spend sixty seconds writing down: who’s reading the output, what you want them to do, and one thing the output must not do. Add that to your prompt.
The more interesting case is when you sit down to write that brief and realize you can’t answer one of the questions. That gap is usually where the second draft was going to come from anyway.
