How to brief an AI like a senior planner

A senior planner given a vague brief does what the model does: makes assumptions, fills in what's missing from general knowledge, produces something plausible that may or may not be what you needed. The difference is that a senior planner can push back. The model can't.

Which means the brief has to do more work.

What makes a brief work for humans and AI

A brief that works for a human planner works for the model too — with one difference. The human can ask questions when the brief is unclear. The model will proceed regardless.

A brief that works for both tells the model:

• Who the audience is in specific terms — not "marketing decision-makers" but "a VP of marketing at a mid-market B2B company who's been burned by AI tools that produced generic copy and is skeptical that another tool will be different"
• What the piece should do — not "inform" but "shift the reader from 'AI doesn't work for our use case' to 'this approach might be different'"
• What the brand is prepared to say — the specific positioning claim this piece advances, not just the general topic
• What "done" looks like — one example of an approved piece in this format that represents the right quality and voice

Every element of a brief that a human planner would need, the model needs. Every assumption you'd correct in a kick-off call, the brief has to catch in advance.

The five elements every AI brief needs

1. Audience in decision context. Not demographics — the decision-maker's current belief or objection. What does this person think right now, and what should they think after reading this? A brief that specifies this produces content aimed at a real shift. A brief that says "marketing managers" produces content aimed at a category.

2. A specific objective. What this piece should produce: a shift in belief, a decision to act, an understanding of something the reader didn't have before. "Inform" and "engage" are not objectives. "Convince a skeptical CMO that brand consistency at scale requires persistent context, not more review cycles" is an objective.

3. The brand's positioning claim. What this brand is saying that distinguishes it from what competitors say. The model defaults to category language without this. With it, the output reflects a specific point of view rather than a reasonable-sounding average.

4. Constraints and what to avoid. Claims the brand hasn't approved, angles that were considered and rejected for this client, comparisons that aren't defensible, terms that are off-brand. These are harder to specify than the positive direction, but failing to specify them produces the most recognizable brief failures — the model does exactly what you didn't think to rule out.

5. An output example. One approved piece in this format that represents the right quality, length, and voice. Not a description of what good looks like — an actual example the model can compare against. This is the highest-impact single element in any AI brief.

The brief format that works across campaign types

A brief that travels:

Audience: [Specific decision-maker + current belief or objection + what should shift]
Objective: [What this piece should produce — a belief change, a decision, an understanding]
Key message: [The one thing this piece is arguing]
Positioning: [What the brand is claiming that competitors aren't]
Avoid: [Terms, angles, claims, or comparisons that are off-limits]
Format: [Length, structure, and what "done" looks like]
Example: [Link to or paste one approved piece in this format]

This format takes five to ten minutes to complete on a piece you know well. It takes longer the first time for a new client or campaign — which is the right time to spend it. The brief is the most expensive part of a campaign to do wrong.

How to show the model what "good" looks like

The most underused element in most AI briefs: the output example. Most briefs describe good output ("professional but warm, direct without being aggressive, confident without being arrogant"). The model interprets those descriptions. There's room for drift on every word.

An example closes the interpretation gap. The model can compare its output against the example in ways it can't compare against a description.

What makes a good example:

It's in the same format as the piece you're requesting — an email example for an email brief, a blog example for a blog brief. Format similarity matters more than topic similarity.

It's approved and representative — not a piece the team considers acceptable, but one they consider good. The model calibrates to the example; calibrating to acceptable produces acceptable output.

It's recent enough to reflect the current voice. A brand's tone shifts over time. An example from four years ago may not reflect how the brand sounds now.

What to do when the brief is solid but the output still misses

If the brief is genuinely solid and the output still misses, one of three things is happening:

The example wasn't close enough in format or quality. The model calibrated to a different bar than you intended. Fix: find a better example or supplement with a second one that specifically shows the dimension where the output missed.

The positioning claim was too abstract. "We help teams produce better content" is a positioning claim in the grammatical sense but not in the strategic sense — it's what every tool in the category says. Fix: make the specific claim the brand is actually prepared to defend.

The audience framing wasn't specific enough. "Marketing professionals" is a placeholder, not an audience. The output will reflect the placeholder. Fix: specify the belief or objection the audience carries into the piece.

In each case: fix the input, not the output. The brief was the problem; editing the output patches the symptom.

For the campaign workflow that makes briefs cumulative: The AI-assisted campaign workflow, start to finish. Why context beats prompt sophistication: Why context beats prompts in AI marketing work. The concepting stage where the brief becomes creative direction: From brief to concepts without a blank prompt.

Frequently Asked Questions

What should I include in a prompt for campaign work?

Everything a senior planner would need: who the audience is in decision-context terms (not just a job title), what this piece should produce (a specific shift, not just "awareness"), what the brand is claiming that competitors aren't, what to avoid, and one approved example in the same format. That's the brief. The prompt is how you deliver it; what the model needs is the substance those elements contain.

How specific does my brief need to be for AI?

More specific than you'd give a senior writer, because the writer can ask questions and the model can't. The elements where specificity matters most: the audience's current belief or objection (not their job title), the objective in terms of a shift or decision (not a vague goal like "inform"), and the positioning claim (not the category language everyone uses). Over-specify; the model ignores what it doesn't need and uses what it does.

How do I show AI what style I want?

Provide an example — not a style description. "Professional but warm, direct without being cold" describes a continuum; the model places itself somewhere on it and there's room for drift. An approved piece in the right format shows the model where on that continuum the brand actually sits. One good example is worth more than three paragraphs of style guidance.

What's the difference between a prompt and a brief?

A prompt is the instruction you give the model at the start of a session. A brief is the complete set of context that makes the instruction meaningful: who the audience is, what the piece should do, what the brand is claiming, what good looks like. Most "prompt engineering" problems are brief problems — the prompt was fine; the context was missing. The brief is the more important document.