Overview#

Most people use AI for writing by dumping a vague prompt in and hoping something usable comes back. Sometimes it does. More often you get a wall of competent-sounding text that doesn’t say what you meant, structured in a way you wouldn’t have chosen, making points you didn’t ask for.

This playbook flips that. Instead of asking the AI to write and then fixing what it produces, you build the document in stages — thesis, then outline, then content — where you control the thinking and the AI does the prose. Each stage is a checkpoint where you decide whether to proceed, revise, or change direction.

The writing sounds like yours because the ideas and structure are yours. The AI just got them on the page faster.

When to Use#

• You need to produce a document (report, blog post, proposal, memo, white paper) and you know roughly what you want to say but haven’t organized it yet.
• You’re staring at a blank page and the friction of starting is the bottleneck, not a lack of ideas.
• You want AI help with writing but you’ve been burned by output that drifts from what you actually meant.
• The document matters enough that you can’t just publish whatever the AI gives you, but you don’t have time to write every word from scratch.

The Play#

Start with the thesis#

Write your thesis yourself. One to three sentences that capture the core argument or purpose of the document. What is this thing trying to say? What should someone walk away believing or understanding?

Most people skip this, and it’s why most AI-assisted writing feels hollow. If you can’t state the point in a few sentences, you’re not ready to write yet. If you’re struggling to pin it down, try working through it in a rubber duck session — talk through what you’re trying to say, let the AI push back with questions, and distill what comes out into your thesis. But the thesis has to be yours.

Be specific. “AI is changing how we work” is a topic, not a thesis. “Teams that use AI for first drafts and human judgment for editing produce better documents in less time” is a thesis. The more specific you are here, the less you’ll fight the AI later.

Build the outline collaboratively#

Give the AI your thesis and ask it to propose an outline. Be clear about format, length, and audience — a five-section blog post for practitioners needs a different structure than a twenty-page white paper for executives.

Review the outline carefully. Structure is thinking. Look for sections that don’t earn their place, gaps in the logic, ordering that doesn’t flow. Reorganize, cut, and add until the outline represents the argument you want to make.

A useful practice: annotate each section with a sentence or two about what it needs to accomplish. Not the content itself, just the job. “Establishes the cost of the status quo” or “Addresses the most likely objection.” These annotations become instructions the AI can use when filling in content, and they keep you honest about whether every section has a purpose.

Fill in content section by section#

Work through the outline one section at a time rather than asking the AI to generate the whole document at once. This gives you control over pacing and lets you course-correct early.

For each section, give the AI your outline, your annotations, and any specific points, data, or examples you want included. The more material you provide, the less it has to invent. AI is much better at organizing and articulating your thinking than it is at coming up with thinking worth articulating.

Read each section before moving on. Does it make the point from your annotation? Does it fit with what came before? If something feels off, fix it now rather than hoping it resolves itself in the full document.

Revise in chunks before looking at the whole#

Once all the sections are filled in, resist jumping straight to a full document review. Work through revisions at the chunk level first. If your outline has sections 2.1, 2.2, and 2.3, pull all three together and revise them as a unit. Look at how they flow into each other, whether they repeat points, whether the progression actually builds.

Section-by-section writing creates a specific category of problem: each section might be fine on its own, but adjacent sections often overlap, contradict each other in tone, or fail to build on what came before. Revising at the chunk level lets you smooth these out while the scope is still manageable. Fixing flow across an entire document at once is harder, and you’ll miss things.

Review the whole document as a unit#

Once your chunks are solid, read the full document start to finish. Even after chunk-level revisions, some issues only surface at the document level — the argument might lose its thread between major sections, the introduction might promise something the conclusion doesn’t deliver, or the pacing might feel uneven.

This is a good place to bring in a rubric. If you built one for this type of document, score the draft against it. If you didn’t, this might be the moment that convinces you to build one for next time.

Iterate on specific problems, not the whole thing#

When you have feedback — from yourself, from the rubric, from a colleague — resist pasting the whole document back in with “make it better.” Identify the specific problem. “Section 3 doesn’t connect the cost data to the recommendation in section 5” gives the AI something to work with. “Polish this up” does not.

If the document has been through a few rounds of revision, consider running it through a secondary AI review to catch the AI-isms that accumulate: filler phrases, em-dashes everywhere, that characteristic symmetry where every paragraph has exactly three points.

Related Playbooks#

• Rubric Driven Assessment — Score your document against a rubric at each stage. Catching structural issues at the outline stage is much cheaper than catching them in a finished draft.
• Secondary AI Review — After iteration, use a fresh session to clean up AI writing patterns that crept in.
• Audience Translation — Once you have a strong base document, adapt it for different audiences rather than writing from scratch each time.
• Generating Variations — If you’re stuck on framing early in the process, generate variations of your thesis or key messages before committing to a direction.
• Rubber Duck with Memory — If you’re struggling to articulate your thesis, talk it through first. The thesis often emerges from conversation before it emerges from writing.

Examples#

Examples coming soon.