Outline before you draft

Once a brief defines purpose and audience, an outline decides the path between them. Each section should perform a job: establish relevance, explain a concept, support a claim, address an objection, compare options, or ask for action. A heading that merely names a subject—“Background,” “Benefits,” “Conclusion”—does not yet reveal that job.

AI is useful at this stage because outlines are cheap to compare and revise. Asking for three structures exposes alternatives before paragraphs create attachment. The model can suggest familiar patterns, but you must choose based on the reader and evidence.

Common structures include:

• Problem → causes → options → recommendation for decisions;
• Question → short answer → explanation → example for guides;
• Claim → evidence → limitation → implication for analysis;
• Before → intervention → after → lesson for cases;
• Chronology with turning points for narratives;
• Objection → response for skeptical audiences;
• Task steps → checks → recovery for procedures.

These are tools, not rules. A document can combine them, but an unexplained mixture often produces repetition or abrupt transitions.

For every proposed section, write a section promise: one sentence stating what the reader will gain and why it belongs there. Then list the evidence or example needed to keep that promise.

Section: Why decision records matter
Promise: Show that the real cost is repeated reasoning, not missing paperwork.
Evidence: Two approved examples of teams revisiting settled choices.
Reader question answered: “Why add another process?”
Transition: If repeated reasoning is the problem, what is the smallest useful record?

This level of detail prevents an outline from becoming eight synonyms for the topic. It also reveals evidence gaps. If a section promises that a practice “cuts onboarding time in half” but no source supports the number, change the promise or acquire valid evidence before drafting.

Give the model the brief, then constrain the outline task:

Propose three genuinely different outlines for this brief.
For each, name the organizing logic, reader journey, strengths, and risks.
Under every section include its purpose, key claim, required evidence,
and transition. Do not draft prose or invent facts.

“Genuinely different” matters. Models often return the same generic sequence with renamed headings. Compare structures at the level of logic, not wording. Ask what each version delays, foregrounds, repeats, or assumes.

Then stress-test the best candidate. Ask questions such as:

• Where will the target reader lose interest or lack context?
• Which section repeats a prior function?
• Does any conclusion appear before its supporting evidence?
• Which objection remains unanswered?
• Does the requested action follow from the argument?
• What can be removed without harming the purpose?

The model’s critique is a hypothesis. Verify it against the brief and your own knowledge of readers.

A strong outline has visible levels. Top-level sections advance the main argument or task. Subsections support one parent idea. If a subsection could stand anywhere, its relationship may be unclear.

Use parallel headings when items have equal status. For a comparison, headings might be “Cost,” “Reliability,” and “Implementation effort.” Mixing “Option A,” “Risks,” and “What leadership should know” confuses categories.

Limit scope early. When AI suggests adjacent ideas, sort them into four buckets:

1. required for the purpose;
2. useful if space allows;
3. better as a link or appendix;
4. outside scope.

This protects the reader from a document that tries to prove expertise by covering everything. Depth is not the same as breadth.

Outlining includes movement between sections. Write the logical bridge: “Now that the constraint is clear, compare the available options.” If you cannot explain why section four follows section three, the order may be arbitrary.

Assign rough word budgets. In an 800-word recommendation, 350 words of background may leave too little room for evidence and action. Budgets force priority and prevent the model from giving every heading equal weight. They are planning estimates, not rigid quotas.

Also choose where uncertainty belongs. Limitations should appear near the claims they qualify, not hidden at the end. Alternatives deserve fair treatment before the recommendation. Definitions should arrive before dependent arguments.

An outline is ready when it covers the brief, has a coherent sequence, assigns evidence, and can survive a skeptical review. It does not need to predict every sentence. Excessive outlining can become avoidance.

Freeze a working version, then draft section by section. If drafting reveals a structural problem, update the outline explicitly. Do not quietly force new material into old headings. Keeping outline changes visible helps distinguish a better argument from uncontrolled expansion.

For collaborative writing, obtain agreement on the outline before polishing prose. Reviewers can debate claims, order, evidence, and omissions without becoming distracted by word choice. This is one of the highest-leverage uses of AI-assisted planning.