How to Brief an AI Agent So It Doesn't Produce Garbage
Brief your AI agent like you're briefing a new hire on day one who has never seen your business, your customer, or your last twelve campaigns. Most people brief like the agent already knows. It doesn't. That gap is why you get garbage.
The short answer
Context first, task second, format third. That's the order. Not the task first. Most people open with "write me a Facebook ad" and wonder why they get mush. The agent has no idea who the customer is, what the product does, what's already been tried, or what good looks like for you. Give it that before you give it the task, and the output changes completely.
What changes the timing
I don't mean timing like a clock. I mean when in the process you load the context matters just as much as what you load.
- Task complexity. A one-line social caption needs less setup. A positioning document or a full email sequence needs a real context dump, or the agent will invent your brand voice for you.
- How much the agent already knows. If you're working in a persistent thread where you've already fed it brand docs, customer research, and past campaign results, you can brief lighter each time. Fresh session, full context, every time.
- How much you're willing to edit. If you have 20 minutes to review and fix output, you can brief loose and clean up after. If you need to ship close to as-is, the brief needs to be tight up front.
- Whether the output needs to sound like you. Anything customer-facing needs voice samples. Internal strategy docs don't.
The prompt structure I actually use
Here's the template, in order, every time:
- Who we are. Two or three sentences. Company, product, what makes us different. Not a mission statement. Facts.
- Who the customer is. Specific. Not "busy professionals." Something like "DTC buyer, 32-45, has tried two other brands in this category and been disappointed, price sensitive but will pay more for proof it works."
- What's already been done. Paste in the last campaign, the last three subject lines, whatever exists. The agent needs to know what's already been said so it doesn't repeat it or contradict it.
- The specific task. One task. Not "help with marketing." Say "write five subject lines for a cart abandonment email, targeting the customer above, that use urgency without discounting."
- Constraints. Word count, tone words, what to avoid, banned phrases. I keep a standing list of words I never want in outbound copy. I paste it in every time.
- Format. Exactly how I want it back. Numbered list, table, three options with reasoning, whatever it is. If I don't specify, I get a wall of text I then have to reformat myself.
On the DTC brand I scaled from $100K to $3M+, our agent briefs got long. A single product launch brief ran about 400 words of context before the actual ask. That felt like overkill the first time I did it. Then I compared output against a 40-word brief for the same task. The long-context version needed one round of edits. The short one needed a full rewrite. That ratio held up every time I tested it.
Signs you are overdue for a better brief process
- You're rewriting more than you're keeping. If you touch over half the output, the brief failed, not the agent.
- The copy sounds generic, like it could be for any brand in your category.
- You're pasting the same context into a new chat every single day because nothing persists and nothing's been saved as a reusable brief.
- Two people on your team get wildly different output from the same agent because they're briefing it differently with no shared template.
- You're asking the agent to "try again" more than twice on the same task. Two tries means the brief was unclear. Three means something's structurally wrong.
What happens if you keep briefing badly
The direct cost is time. If your team spends 15 minutes per piece of content cleaning up vague-briefed AI output, and you're producing 20 pieces a week, that's five hours burned weekly fixing a prompt problem. Over a quarter that's 65 hours, close to two full work weeks, spent re-doing work that a better brief would have gotten right the first time.
The less obvious cost is trust. I've watched marketing leads give up on agents entirely after a few bad runs, decide "AI doesn't work for our brand," and go back to fully manual work. Almost every time, the problem wasn't the tool. It was a brief that handed the agent a blank slate and expected brand-perfect output anyway.
The most common mistake
People lead with the task and skip the customer. I see this constantly: "Write ad copy for our new skincare serum" with zero mention of who buys it, why, or what objections they have. The agent fills the gap with generic category language, because that's all it has to work with. Garbage in, garbage out isn't an AI problem. It's a briefing problem that existed long before AI did.
Practical takeaway
Build one reusable context doc per brand: who you are, who the customer is, voice rules, banned words, and three examples of your best past work. Paste that into every new agent session before you ask for anything. It takes five minutes to assemble once and saves hours every week after. Treat the agent like a smart freelancer on their first day, not a mind reader, and the output stops being garbage.