The Co-Writing Mindset
A blank page is slow. A page full of generic AI sludge is worse, because now you have to spot the problems, delete them, and start over. The goal of this chapter is to skip both: draft fast and sound like you.
Here's the rule that makes that possible. You bring the argument; AI brings the speed. The moment you ask AI to decide what you think, you've handed over the one thing that makes the paper yours — and you've guaranteed it reads like everyone else's. Use AI to expand points you've already made, restate clunky sentences, and get a rough draft of a section onto the page so you have something to fight with. Editing your own bad first draft is ten times faster than generating a good one from nothing.
So never start with "write my discussion section." Start with your outline (chapter 8), your notes (chapter 6), and your actual claim. Feed those in. Then the AI is a drafting assistant, not a ghostwriter.
Prompts That Produce Your Voice, Not the Default One
Left to its own devices, a model writes in the same beige register: "In today's rapidly evolving landscape," "plays a crucial role," "it is important to note." That voice is detectable, forgettable, and not yours. You override it by being specific about what to write and how it should sound.
Give it your material and a voice spec:
Here is my outline point and three notes from my sources:
[paste point + notes with citation keys]
Draft this as one paragraph (~120 words) for a literature
review. Constraints:
- Make this claim the topic sentence: [your claim]
- Plain, direct academic English. No filler phrases.
- Don't add facts or sources beyond my notes.
- Vary sentence length; avoid starting sentences with
"Furthermore" or "Moreover."
The single most useful trick: give it a sample of your own writing and tell it to match.
Here are two paragraphs I wrote, so you can match my style:
[paste 2 of your real paragraphs]
Now draft the "methods limitations" section in that same
voice, using these bullet points: [your bullets]
Notice what's missing: vague instructions like "make it good" or "academic tone." Those produce the default. Constraints produce control. If you want a deeper toolkit for shaping AI output to a specific style and register, the AI writing & content creation course drills exactly this.
Draft Section by Section, Not All at Once
Asking for a whole paper in one prompt is how you get something that's coherent in tone and hollow in substance. Work in chunks the size of one argument.
A workflow that keeps you fast and keeps the paper yours:
- Talk before you type. Explain the section to the AI in your own messy words first — voice-to-text is fine. Then ask it to organize your explanation into prose. Now the ideas are demonstrably yours.
- Draft one section. Feed the outline point plus notes. Get ~150 words.
- Interrogate it. "Which sentence here is the weakest, and why?" "Where did you smuggle in a claim my notes don't support?" Make the model audit its own draft.
- Rewrite the topic sentences yourself, by hand. They carry your argument and they're the first thing a reader (and an examiner) reads closely. Owning them changes the feel of the whole paragraph.
This loop is faster than it looks because you're never staring at nothing, and never wading through 2,000 words of generic text hunting for the 300 that matter.
The Traps That Get You Caught
Generic prose isn't just boring — increasingly it gets flagged, and worse, it can be wrong. Watch for these.
Invented evidence. If you ask AI to "support this argument," it will happily fabricate statistics, studies, and quotes. Only let it write from notes you supply. Every factual claim and every citation gets verified against the real source — that's chapter 10's whole job, and it is non-negotiable.
Tell-tale filler. "It is worth noting," "delve into," "tapestry," "multifaceted," "in the realm of," sentences that open with "Moreover" three paragraphs running. Search-and-destroy these. They signal autopilot to any reader who's seen AI text — which now includes most of your professors.
Tonal uniformity. Human writing has rhythm: a long sentence, then a short one. A confident claim, then a hedge. AI defaults to evenly-paced, medium-length sentences that never surprise you. Break that pattern manually. Cut a sentence in half. Start one with "But."
Over-hedging. Models pile on "may," "could," "potentially," "it is possible that" until your argument has no spine. If your data shows something, say it shows something.
A quick cleanup pass:
Edit this paragraph. Do not add information. Tasks:
- Delete filler phrases and hedging that weakens the claim.
- Break up any sentence over 30 words.
- Flag in brackets any claim that needs a citation I haven't
provided.
Return the edited paragraph plus a list of what you changed.
Read It Out Loud Before You Trust It
The final gate is your own ear. Read the section aloud. Anywhere you stumble, where a sentence sounds like a press release or a phrase isn't one you'd ever say — rewrite it in your words. This catches what detectors and checklists miss, and it quietly re-stamps the draft as yours.
Do this and the math works in your favor: AI handles the friction of getting words down, you handle the thinking and the final voice. You draft a section in twenty minutes instead of two hours, and it still reads like a person wrote it — because, in every way that counts, one did. For the academic-paper-specific version of this whole loop, the AI research & academic papers course walks through it on real manuscripts.