Start With the Argument, Not the Sections
Most outlines fail because they start with structure — Introduction, Methods, Results, Discussion — and then hunt for things to put in each box. That gives you a shaped void, not a paper. The boxes get filled with whatever you have, the logic limps, and your reader senses it.
Flip it. Before you name a single section, write your argument as one sentence: the claim your paper exists to defend. Then list the three to five things a skeptical reader must accept for that claim to hold. Those are your load-bearing points. Everything else is support, context, or cutting-room floor.
You already did the hard part if you took notes well in Chapter 6. Pull your tagged claims and evidence into one document and feed it to your AI tool with a blunt instruction:
Here are my research notes (claims, evidence, sources).
My thesis is: [one sentence].
List the 3-5 distinct sub-claims I'd need to prove this thesis.
For each, tell me which of my notes support it and which note
doesn't clearly belong to any sub-claim.
That last line is the gold. The notes that don't fit are either your scope creep or your missing argument. Decide which, deliberately.
Order the Points So Each Earns the Next
A paper flows when each section makes the next one possible. Reader can't evaluate your results until they trust your method. Can't care about your method until they believe the question matters. That dependency chain is your order.
Sequence your sub-claims by what depends on what, not by what you researched first or what you find most interesting. Then pressure-test the order:
Here is my ordered list of sub-claims. For each one, answer:
what must the reader already accept before this point lands?
Flag any point that assumes something I haven't established yet.
If the tool flags a forward dependency — point 4 leaning on something you don't prove until point 6 — you have a reorder problem, and it's far cheaper to fix now than after you've drafted 2,000 words. Treat the AI as a logic checker here, not a writer. It is genuinely good at spotting "you asserted X but never set it up," and that single catch can save you a brutal revision pass.
Build the Section-by-Section Skeleton
Now map your argument onto real sections. The standard shapes exist for a reason: IMRaD for empirical work, a thesis-and-evidence build for humanities and argumentative essays. Pick the one your field expects, then bend it to your argument rather than the reverse.
For each section, write three things: its job (one phrase — "establish the gap this paper fills"), its key points (your sub-claims and evidence, in order), and its exit (the idea that hands off to the next section). A section without a clear job is a section you'll waffle through.
Push the detail down to the paragraph. A genuinely draftable outline has a line per intended paragraph — topic sentence plus the source or data behind it. That granularity is what lets you draft fast later (Chapter 9), because you're never staring at a heading wondering what goes under it.
Turn this section into a paragraph-level outline.
For each paragraph give me: a one-line topic sentence, the
note or source it draws on, and what it sets up next.
Mark any paragraph where I have a topic sentence but no
supporting note.
Those unsupported paragraphs are your honest to-do list. Either you find the source, or you cut the claim. Don't draft over a hole and hope nobody looks.
Make AI Attack the Outline
Your outline looks finished. That's exactly when to try to break it. AI is most useful here as an adversary, because you've stopped being able to see your own gaps.
Run three distinct passes — don't blend them into one vague "review this":
Act as a skeptical reviewer for [field]. Read this outline.
1. Where is the logic weakest — which claim is least supported?
2. What obvious counterargument do I never address?
3. What would a reader expect in this kind of paper that's missing?
Be specific. Don't praise anything.
The counterargument pass matters most. A paper that never names its strongest objection reads as naive, and reviewers pounce on it. If the tool surfaces an objection you can't answer, that's not a flaw in the outline — it's the section you didn't know you needed.
One caution: AI loves to suggest more. More subsections, more "you could also discuss." Resist. A tight paper that proves one thing beats a sprawling one that gestures at five. When the tool proposes an addition, ask whether it serves your one-sentence thesis. If not, it's noise wearing the costume of thoroughness. The skill of cutting AI's suggestions is worth as much as the skill of prompting for them — the AI for Academic Research course drills this judgment if you want the reps.
Lock It Before You Draft
A finished outline passes a simple test: you could hand it to a labmate and they could draft any section without asking you what you meant. Check yours against that bar.
Concretely, you should be able to point to:
- One thesis sentence, unchanged through every pass.
- Three to five sub-claims, ordered so each earns the next.
- Sections with a stated job and a clean handoff.
- A paragraph line for every intended paragraph, each tied to a source or a marked gap.
- The counterargument, named and given a home.
If any of those is fuzzy, fix it now. Outlining is the cheapest place in the whole process to change your mind — moving a paragraph is a keystroke, rewriting three pages around a relocated argument is an evening you won't get back.
When all five hold, you don't have notes anymore. You have a blueprint. The next move is to fill it in your own voice, fast, without sounding like the machine that helped you check it.