Why Magic Prompts is the Past. Long Live the Brief
For the last two years the advertising industry chased incantations. That era is over. What separates good output from noise today is a discipline agencies have practised for decades.
For the last two years the advertising industry chased incantations. That era is over. What separates good output from noise today is a discipline agencies have practised for decades.
Remember when prompt engineering felt like learning spells? You told the model to take a deep breath. You asked it to imagine itself as an award-winning creative director at a New York agency. You buried small formatting tricks in the text to force it to think in the right order.
It worked. For a while.
If you are still working that way, you are working harder for a worse result. Context windows have grown. Reasoning has matured. The rules changed while most of us were still rehearsing the old ones.
It is not about phrasing. It is about framing.
Modern LLM models do not need to be walked through basic logic. They need to be briefed. What the field now calls context engineering comes down to three familiar questions. How sharply have you defined the task? What data have you put on the table? Where are the boundaries?
This is not a technical problem. It is a communication problem. And it is precisely the exercise a good account director runs every Monday morning.
The anatomy of a brief that works
Drop the wall of text. Treat the prompt as a tight brief with four parts.
Task. Open with a decisive verb. Draft. Rewrite. Synthesise. Critique. No polite warm-up.
Context. Hand over the raw data, the previous campaign material, the audience insight, the client's tone of voice. Context is the sandbox the model works inside. Without it, the model guesses. It guesses convincingly, which is exactly the problem.
Constraints. Say what it must not do. Ban the clichés. Set a hard word count. Outlaw generic openings. Negative constraints are the cheapest edit you will ever commission.
Format. Specify the deliverable precisely. Three columns. Five bullets. Raw JSON. A script with dialogue and voiceover.
The difference in practice
The weak brief: "Write some social copy for our new sustainability campaign."
The strong brief: "Write three LinkedIn options of no more than 60 words, aimed at procurement directors in construction. The campaign is about documented CO2 reduction in concrete. Use only the figures in the attached fact sheet. Avoid the words 'game changer' and 'green transition'. No throat-clearing openings. Deliver three numbered options with two lines of rationale under each."
Same model. Same minute. Two entirely different results. The difference sits with the sender's clarity, not the technology.
From prompt writer to workflow architect
The direction of travel is clear, and it leads away from the chat window. We are moving toward systems that work in the background, read the live project context and put the analysis on the table before anyone has opened a dialogue box.
When you do prompt, you will not sit correcting line by line. You define the intent. You set the criteria for what good looks like. Then you let the system execute and self-correct until it clears the bar.
You stop being the copywriter inside the machine. You become the editor-in-chief.
Playbook: three moves for tomorrow
Build a context sheet for every client. One page: audience, tone of voice, forbidden words, most recent results. Attach it every time. That is your factual anchor.
Write the ban list down. Start with your own agency's favourite clichés. They cost the most, because nobody can see them any more.
Define "good" before you ask for anything. If you cannot describe the criterion, the machine cannot hit it. And you are back to editing by hand.
The good news
The future of working with AI does not lie in remembering tricks that outwit a machine. It lies in thinking clearly and delegating precisely.
That is an old-fashioned discipline. It is called briefing. And the advertising industry has every qualification to own it.