Begin with a brief and audience

AI makes producing sentences almost frictionless. That convenience can encourage premature drafting: generating several paragraphs before purpose, audience, evidence, or scope is settled. The output may look complete, so unresolved decisions become harder to notice. You then spend time polishing a document that was pointed in the wrong direction.

A writing brief separates thinking from phrasing. It is a compact agreement about the job the document must perform. It gives the model boundaries, gives the writer a review standard, and makes disagreements visible before they are buried in prose.

Start with purpose. “Write a project update” names a format. “Help sponsors understand why the launch date is at risk and decide whether to reduce scope” names an outcome. The second statement guides what belongs in the document, what can be omitted, and where the requested decision should appear.

Audience is not just “general readers” or a job title. Useful audience notes describe:

• what readers already know and what they do not;
• why they are reading and how much time they have;
• the decision, task, or concern they bring;
• technical terms they understand or need explained;
• their relationship to the author;
• likely objections, accessibility needs, and context.

Consider a report about a delayed software release. Engineers may need failure evidence and dependencies. Executives may need impact, options, costs, and a recommendation. Customers may need an accurate effect on their work and a realistic next update. The underlying facts should remain consistent, but selection, sequence, definitions, and detail change.

Do not use audience modeling to manipulate. “Explain the trade-off in plain language for a time-limited reader” is legitimate. “Exploit the reader’s fear so they approve” is not. Audience awareness should reduce the reader’s effort and support informed judgment.

A practical brief contains seven fields:

1. Purpose: what should change after reading.
2. Audience: knowledge, needs, context, and relationship.
3. Core message: the one idea the reader should retain.
4. Evidence: approved sources, examples, data, and direct observations.
5. Scope: what is included and deliberately excluded.
6. Constraints: length, format, deadline, terminology, policy, and accessibility.
7. Success criteria: observable qualities the final piece must satisfy.

Write unknowns as [NEEDS INPUT]. Do not invite the model to fill them with plausible detail. If the release date is not approved, the brief should say so. If a quote has no source, exclude it until verified.

Success criteria should be testable. “Make it engaging” is subjective. “Open with the customer problem, define two unfamiliar terms, support each numerical claim with a supplied source, and end with three options” can be checked. Include negative criteria too: no invented examples, no unsupported certainty, no imitation of a living writer, and no confidential information.

Models can blur source material and commands when both appear in an undifferentiated block. Label them:

PURPOSE
Help new managers recognize when a meeting should become an async update.

AUDIENCE
First-time managers; familiar with team meetings, unfamiliar with facilitation theory.

APPROVED FACTS
- Weekly status meetings average 45 minutes in our observed sample.
- The team already uses a shared project board.

INSTRUCTIONS
Propose a 700-word internal guide. Explain terms in plain language.

BOUNDARIES
Do not invent productivity statistics or claim that async work suits every decision.

The labels do not guarantee accuracy, but they make review easier. Ask the model to identify missing information and contradictions before it drafts. You might discover that “new managers” spans very different teams, or that the evidence supports only a local recommendation rather than a universal claim.

Keep the brief beside every draft. When the prose expands, ask whether each section serves the purpose and audience. When reviewers request changes, update the brief if the assignment itself changed. Otherwise, evaluate the request against the existing agreement.

The author remains responsible for the choices. AI can suggest audience questions, challenge scope, or reveal ambiguity, but it does not know organizational intent, permissions, or reader consequences unless you provide them. For academic, legal, medical, employment, or regulated writing, follow the relevant rules and involve qualified reviewers.

A good brief also supports stopping. If you lack permission to use the source material, cannot verify the central claim, or do not know who the audience is, producing more prose is not progress.