How to Write a Brief That AI Can Use

A brief written for a human writer contains instructions, context, and expectations. Some of it is implicit: the writer infers the register from the brand, asks clarifying questions when something is unclear, brings their own knowledge of the topic to fill gaps.

A brief written for an AI model cannot rely on any of that. The model does not ask clarifying questions unless you invite it to. It does not infer implicit expectations. It fills gaps with the statistical average of all the content it has processed, which produces generic output. The brief has to be explicit where a human brief could be implicit, and complete where a human brief could have gaps.

What to include

The argument, stated as a single sentence. Not the topic. The claim. "Content calendars fail solo operators because they were designed for teams" is an argument. "About content calendars" is a topic. The model produces a piece around what you give it. Give it a topic and it covers the topic. Give it an argument and it argues the claim.

The target reader, described specifically. The more specific the description, the more specific the output. "Freelance content writers who are managing four or more clients, producing eight or more pieces per month, and looking to systematize their production" narrows the audience to people in a particular situation. "Content writers" does not.

The scope constraint. What the piece does not cover. This is the most underused element of AI briefs and the most effective at preventing generic output. "This piece does not compare content calendar tools and does not recommend software" removes a category the model will naturally generate when writing about content calendars.

The structural preference. If you have a preferred structure, state it. "Open with a specific failure scenario rather than a definition" tells the model something about the entry point. "Use no more than three main sections" prevents the model from producing eight equally-weighted sections of mediocre depth.

The register notes. Short notes about what the voice is not. "Does not use hedging phrases like 'it depends on your situation.' Does not present balanced views on questions where there is a correct approach."

The output specification. Length, format, any specific requirements. This is usually the shortest part of the brief and goes last.

What to cut

Background information about the topic. The model does not need you to explain what a content calendar is. It has processed thousands of articles about content calendars. Background information inflates the brief without improving the output.

Style guidance that is not specific. "Write clearly and engagingly" does not change model behavior. Every brief contains this instruction implicitly. Specific guidance changes behavior: "sentences average under 15 words in the main body" is specific.

Examples of similar articles, unless they are your own work in the style reference. Showing the model articles by other writers nudges it toward those writers' styles. If you want your style, the examples should be yours.

Hedging about the scope. "Feel free to add anything you think would help" is an invitation for the model to fill space with coverage. If you want something included, include it in the brief.

Format

Short paragraphs with labeled sections work better than long prose briefs. The model processes structured information more reliably than it processes prose where context is buried in sentences. Label each element: Reader, Argument, Scope constraint, Structure, Register, Output.

A brief template is more efficient than writing a fresh brief every time. The template forces completeness and eliminates the question of what to include. You fill in the slots and the template handles the structure.

The brief should fit on one page. A brief that requires two pages to state what you want usually means the argument is not yet clear enough to write. Clarify the argument before you hand it to the model.