Brief an AI like you would brief a freelancer
The fastest way to better AI output is a better briefing: context pack, constraints, examples, and a clear definition of done, just like hiring out work.
Imagine hiring a sharp freelancer who knows nothing about your business and starts work in thirty seconds. You would not send them “write something about our spring promotion” and hope. You would brief them. The single most useful mental shift in working with AI is realizing you are always briefing that freelancer.
This guide is the working companion to prompting fundamentals: same levers, packaged as the document you actually write.
The four-part briefing
Every solid briefing answers four questions. For a competent stranger or a language model, the same four.
Who is this for and what is the situation? The context pack. Your business in two sentences, the audience, what prompted this task, what the reader already knows. Models fill missing context with the most average assumption on the internet, and average is rarely your situation.
What exactly is the deliverable? Format, length, structure, tone. “A 200-word email to existing customers, plain tone, one clear call to action” beats three rounds of “shorter, warmer, no, shorter.”
What are the constraints? The things it must include, the things it must never say, the claims it may not make. If your industry restricts promises, say so in the briefing. The model does not know your compliance rules until you state them.
What does done look like? One or two examples of past work you considered good. An example transmits a hundred preferences you would never think to list. This is the highest-value section and the one people skip.
Build the pack once, reuse it forever
The context pack barely changes between tasks, so stop retyping it. Keep a short document: what the business does, who the customers are, voice notes, standing constraints, one or two gold-standard examples. Paste it at the top of any serious session, or store it in your assistant’s standing instructions if it has them.
This is the move that makes the difference between “AI gives me generic junk” and “AI drafts like someone who has worked here a month.” The same pack powers writing tasks in writing with AI and the recipes in the automation playbook. One asset, many uses.
Grant the permission to ask
A real freelancer asks clarifying questions. A model, by default, does not. It guesses, confidently. Add one line to any briefing for nontrivial work: “Before drafting, ask me up to three questions that would most improve the result.”
Modern models use this permission well, and the questions themselves are diagnostic. If the model asks something you cannot answer, you have found the hole in your own thinking before it became a hole in the deliverable.
Iterate like a client, not a gambler
When a draft misses, the gambler’s move is regenerating and hoping. The client’s move is diagnosing which part of the briefing failed. Wrong facts: the context pack was thin. Wrong shape: the deliverable spec was vague. Wrong vibe: the example was missing or mismatched. Fix that one section and rerun.
Two briefing fixes beat ten rounds of “no, more like…” Conversation drift is real: a long thread accumulates wrong turns, and the model keeps building on them. For a fresh start, take the improved briefing to a fresh session.
The sixty-second version for small tasks
Not everything deserves a full pack. For quick tasks, compress the four parts to four lines: situation, deliverable, one constraint, one example phrase. Sixty seconds of briefing still beats zero, and it builds the habit that makes the big briefings automatic.
The discipline here is the same one that makes delegation work with people. Vague requests produce confident guesses; specific briefings produce usable work. The model just removes the polite pause where a human would have asked you what you actually meant.