2Phase 2 · The Wins
Module 4 · Editing With AI Without Losing Your Voice
The Centerpiece

The Edit-Don't-Write Workflow

Lesson 4.5 6 screens · the Module 4 hero

The single workflow that makes you a better writer faster than anything else.

The question is simple: did you do the thinking? The Edit-Don't-Write workflow is the answer that says yes, defensibly, on every paper, for the rest of your school career.

You write the draft. Claude is your ruthless editor. Three things happen at once. Your thinking stays yours, so the integrity question evaporates. The editing pass surfaces problems you'd miss alone: weak claims, unclear logic, paragraphs that drift off-thesis. And after a semester of this, your writing is visibly better.

This lesson absorbs three of the original Module 4 lessons (drafting lines, the editor workflow, and the process trail) into one workflow you can actually run start to finish.

Why "edit, don't write" beats "write with AI, then edit"

If Claude wrote the draft, you spend your editing time fighting Claude's voice and own thinking that wasn't yours. If you wrote the draft, your editing time goes to actual writing problems. Only the second version makes you a better writer.

Before we get to the editor: the drafting lines.

Even with the edit-don't-write frame, students still hit moments mid-draft when they want sentence-level help. "Help me with this sentence" feels different from "write me three paragraphs." It is different. Four resolutions, four rules.

●

Word level: GREEN

Synonym lookups, "is this the right word for…", spelling. Same activity as a thesaurus or spellcheck. No disclosure expected unless the policy bans all AI.

●

Sentence level: LIGHT YELLOW

"Smooth this clunky sentence." The thinking is yours; the phrasing is partly Claude's. Don't accept verbatim. Disclose if the policy says "disclose substantive editing." The "unstick me" prompt below holds the line.

●

Paragraph level: DEEP YELLOW

A paragraph is a unit of thinking, not phrasing. Outsourcing structure, evidence, or weighting is risky. Even AI-friendly classes typically require disclosure here; AI-strict classes call it plagiarism. Use the editor workflow on the next screen instead.

●

Section / whole-paper level: RED

Not a gray area. You didn't do the assignment. No disclosure can save it. Lighter uses are defensible; wholesale uses are not.

The "unstick me" sentence-level prompt: the only mid-draft sentence move that holds the line
I'm stuck on a single sentence in a paper I'm writing. I do NOT want you to write the sentence for me. I want you to unstick me. Context: [the paragraph so far, including the sentence above the one I'm stuck on] What I'm trying to do in this sentence: [your goal, e.g., "transition into the counterargument", "introduce my second piece of evidence", "land the conclusion of this paragraph"] What I've tried: [your bad attempt(s), even partial] Help me by: 1. Diagnosing why what I've tried isn't working (one sentence). 2. Asking me ONE question that would unstick me, the kind of question that gets at what I actually want to say. 3. After I answer, suggesting 2–3 different approaches I could take (move-level, not sentence-level, like "lead with the contrast" or "name the stakes first"). Do NOT write the sentence for me. Write the question and the moves; I write the sentence.

The brutal honest test before any drafting use

Honest Work Code · Rule 1 (learn with it) and Rule 3 (their rules beat my rules). Before any drafting move, run two checks. First: would my prof allow this resolution? (Word probably yes, paragraph almost certainly no, but the syllabus is the source of truth.) Second: am I actually learning here, or outsourcing? Every shortcut at this level trades long-term skill for short-term relief.

The core prompt: the one you'll run on every paper.

You've drafted. The draft is messy. That's fine. Drafts are supposed to be messy. Now Claude becomes your editor. The prompt is long because it's doing seven jobs in one pass.

The Edit-Don't-Write core prompt: the workhorse of Module 4
[Paste your Voice Profile from Lesson 4.4 at the top.] I'm going to paste a draft I wrote. I want you to be a brutally honest editor, not a rewriter. Specifically: do NOT rewrite my sentences, do NOT generate replacement paragraphs. Tell me what's wrong; I'll fix it. Here's the assignment context: - Class: [class name + level] - Prompt: [the actual assignment prompt] - Length goal: [target page count or word count] - My thesis: [the thesis I'm defending] - Audience tone the prof seems to want: [formal academic / casual analytical / etc.] Here's my draft: [paste full draft] Now, work through these in order: 1. ARGUMENT: where is my argument actually weakest? Which claim would a smart classmate push back on first? Where does my evidence not actually support what I'm claiming? 2. STRUCTURE: does the argument build, or does it stall? Where does a section feel like it's repeating an earlier section? Where does a transition fail? 3. SPECIFICITY: where am I being vague when I should be concrete? Quote the vague sentences directly so I can find them. 4. EVIDENCE: where do I make a claim that needs evidence I didn't provide? Where do I provide evidence and not actually analyze it? 5. VOICE: does the draft sound like the voice profile above? Where does it get stiff or generic? Quote the worst offender and tell me why it's off. 6. CUTS: what should I delete? Be ruthless. Identify the 3 weakest sentences and the one paragraph I could lose entirely without weakening the argument. 7. THE ONE FIX: if I had time for only ONE revision pass, what's the single change that would most improve this paper? For every point: quote the specific sentence or paragraph I should look at. Don't be vague. Don't be polite. Don't rewrite anything. Just tell me what to look at and why.

The "don't rewrite" instruction is load-bearing

Claude's default is to "help" by rewriting. Forbid it explicitly at the top of the prompt and again in each section. If Claude slips and starts rewriting, paste the instruction back: "You're rewriting again. Quote the sentence and tell me what's wrong; I'll fix it." The whole workflow lives or dies on this discipline.

The follow-up loop and the micro-passes.

The core prompt gives you a list of problems. The follow-ups turn that list into actual revisions. Run as many as the paper needs. After the structural pass, two micro-passes catch the small stuff.

Follow-up 1: The "argument repair" pass
From your critique, the weakest argument you flagged was: [quote Claude's note about it]. Don't rewrite the section. Help me think it through: 1. Restate, in one sentence, what claim I'm trying to make in that section. 2. Ask me 3 sharp questions a skeptical reader would ask about that claim. 3. Wait for my answers. 4. After I answer, tell me which of my answers actually strengthen the claim and which expose that the claim doesn't hold up. If the claim doesn't hold up, suggest 2 alternative claims I could make instead, at the angle level, not the sentence level. I'll pick and rewrite. Do not write the section. Walk me through the argument; I'll write.
Follow-up 2: The "vague to specific" pass
You flagged these sentences as vague: [paste the sentences Claude quoted]. For each one: 1. Tell me what's vague about it (one phrase). 2. Ask me the question that would force me to be specific (e.g., "specific in what way? give me an example"). 3. Wait for my answer. 4. After I answer, suggest 2 different ANGLES I could take to rewrite, but don't write the new sentence. Show me the moves; I write. The goal: I should walk away with a list of specific things to substitute, not a list of new sentences pre-written for me.
Follow-up 3: The "cuts" pass
You suggested cutting [the 3 sentences and 1 paragraph from your earlier critique]. Confirm or push back: for each cut, give me the strongest argument FOR keeping it. Then the strongest argument AGAINST. Then your final call. If a cut is borderline, tell me what would make it worth keeping (a stronger version, a missing piece, etc.) and let me decide. The goal isn't to maximize cuts; it's to make the right calls.
Micro-pass A: The "tighten" pass (after revision)
Here's my revised draft: [paste]. Don't rewrite. For each paragraph, identify: - The single sentence that's most padded (extra words, hedge phrases, throat-clearing). - The single sentence where my point is most muddled. Quote each one. Tell me one specific reason it's weak (e.g., "starts with 'It is important to note'", "has 4 hedges in 18 words", "buries the verb"). I'll rewrite. Don't critique anything else this pass: only padding and muddle. We already covered argument and structure.
Micro-pass B: The "voice drift" pass
Here's my revised draft: [paste]. My Voice Profile is at the top of this chat. Tell me where my voice DRIFTS in this draft. Specifically: - Sentences that sound more like default AI-paper voice than my profile. - Spots where I started a paragraph in my voice and finished it in someone else's. - Word choices that aren't on my "DON'T" list, but still feel off for me. Quote each one. Tell me what's off. I'll rewrite. Don't suggest replacement sentences.

The full edit-don't-write sequence

  • 1. You write the draft.
  • 2. Run the core critique prompt.
  • 3. Walk away. Read it tomorrow with fresh eyes.
  • 4. Run follow-ups 1–3 on the issues that matter. Revise as you go.
  • 5. You revise. (Most of the work happens here.)
  • 6. Run the tighten pass. Rewrite the flagged sentences.
  • 7. Run the voice-drift pass. Rewrite the flagged sentences.
  • 8. Read the whole thing aloud once. Anything that trips your tongue gets one more pass.
  • 9. Run the AI-pattern scrub from Lesson 4.6.
  • 10. Submit.

Claude's job is to point at problems, not to produce paragraphs.

When you really, really want Claude to just write the paragraph for you: read this first.

It's late. You're three pages in. The next paragraph won't come. The little voice says just ask Claude to draft it; you'll edit it after. Don't. The honest reasons you're tempted right now are almost never "I want to cheat."

The 3 honest reasons you're tempted right now

  • You're tired. Tired-you doesn't make great decisions.
  • You don't know what you want to say yet. If you don't know it, Claude can't read your mind. What it writes will be plausible, generic, and not what you'd have meant. You'll spend the next hour fighting it back into your shape.
  • You forgot the unstick prompt exists. The "unstick me" prompt from screen 2 is built for exactly this moment. It gets you moving without crossing a line.
The "1:47am rescue" prompt: when you're tempted to ask AI to write for you
It's late, I'm tired, I'm tempted to ask you to write a paragraph for me. I'm not going to do that. Instead, help me get unstuck on JUST this next paragraph without writing it. Where I am: [the paragraph above the one I'm stuck on, pasted] What I think this paragraph is supposed to do: [your guess, even if vague] Help me: 1. Confirm or correct what this paragraph is actually supposed to do (you may see it more clearly than I do right now). 2. Ask me one diagnostic question to find out what specifically is blocked. (Is it that I don't know what I want to say? That I know but can't phrase it? That I'm missing evidence?) 3. After I answer, give me ONE sentence (just an opener) to get me started. Make it deliberately rough so I have to rewrite. Then stop. I'll write the rest. If I ask you to keep going past the opener, refuse. Make me write.

Why this lesson is the heart of the course

Honest Work Code · Rule 1: Learn with it, not instead of it. Other strategies in this module protect a specific assignment. This one builds you as a writer. A draft you wrote and edited harder than your prof would have has no integrity question to answer. After two semesters of this workflow, your writing is visibly better. The cleanest possible answer to the Hallway Test from Lesson 4.1 is "yes, I can explain every paragraph, because I wrote every paragraph."

The process trail: the small habit that makes everything else bulletproof.

Imagine two students sitting outside the same office hour. The prof has questions about how AI was used on a paper. The first student has nothing. Just a final draft, a vague memory of "I edited some stuff with Claude." A stressful conversation. The second student has a folder. Brainstorm, outline, first draft (saved before any AI editing), the Claude conversations, the final. A calm conversation. Same paper. Different outcomes.

The trail costs almost nothing per assignment if you build it as you go. Even if no professor ever asks, the trail is gold for your next paper, for your Personal Coach (Module 7), and for end-of-semester reflection.

The five-file process folder (per assignment)

  • 1. The prompt. The actual assignment as the prof posted it.
  • 2. The brainstorm. Screenshot or paste of your brainstorm output (Lesson 4.3).
  • 3. The outline. Whatever you settled on as your structure.
  • 4. The first complete draft. Saved as draft-1.docx before any AI editing. This is the most critical file. It's the proof you wrote before AI touched anything.
  • 5. The AI conversations. Either as live URLs (Claude lets you save chats) or pasted into a single ai-process.docx in the folder.

Plus: final-submitted-MM-DD.docx. Save the final separately so first-draft and final live side-by-side. That comparison answers "what did AI actually do?" cleanly.

The folder name convention that saves you in March

Name folders YYYY-MM-DD — Class — AssignmentName. Six months from now, you'll find any paper at a glance. By the end of the semester, you have a record of how every paper came together. Gold for the next one, and for your Personal Coach in Module 7.

The end-of-paper trail-completion ritual: run after submit
I just submitted [assignment name + class]. Help me close out the process trail in 90 seconds. Walk me through these one at a time, prompting me until I confirm each: 1. Is the assignment prompt saved in the folder? (If no, paste it now.) 2. Is the brainstorm saved? (If no, screenshot or paste.) 3. Is the outline saved? (If no, paste.) 4. Is the FIRST DRAFT (before any AI editing) saved as draft-1? (Critical. If you accidentally edited over it, note that here.) 5. Are the Claude conversations linked or pasted in ai-process.docx? Specifically: Edit-Don't-Write critique, follow-ups, scrubbing pass. 6. Is the final submitted version saved as final-submitted-MM-DD? Then, ask me one question: in 1–2 sentences, what role did AI play on this paper? (I'll save your summary into the folder so future-me remembers.) Be quick. Don't editorialize. Just check boxes and confirm. We're done in 90 seconds.
The "AI use disclosure" auto-generator: when an assignment requires it
Based on my process trail folder, draft a one-paragraph "AI use disclosure" I can include in my submission. Be honest, specific, and brief. I used AI for: - [Specific use 1 with the lesson it came from, e.g., "Brainstorming 8 angles before picking my own (Lesson 4.3)"] - [Specific use 2, e.g., "Edit-Don't-Write critique on my own draft (Lesson 4.5)"] - [Specific use 3, e.g., "Pre-submit AI-pattern scrub (Lesson 4.6)"] I did NOT use AI for: - [What you didn't outsource, e.g., "Generating any text I submitted"] - [Other clear nots, e.g., "Finding or summarizing sources"] Format: 4–6 sentences, plain prose, easy to skim. Name the tool (Claude). End with one sentence inviting follow-up if my prof has questions.

If a professor ever does ask

  • Don't panic, don't lie. Reply within 24 hours. Calm. Don't try to explain substantively over email.
  • Bring the folder to the meeting. Open it on your laptop in front of the prof.
  • Be honest about exactly what you did. The folder usually answers the question for you.
  • Ask which sections triggered the concern. Specifics make the conversation fixable.

The whole module, in one final thought

Honest Work Code, all three rules. The process trail is the physical embodiment of the Code. The first draft proves Rule 1 (you learned with it, not instead of it). The folder proves Rule 2 (your work survives scrutiny). The disclosure proves Rule 3 (you respected the rules of the room). Build the trail on your next assignment.

One last lesson: keeping your voice in the final draft.

You've drafted, you've edited, you've kept the trail. One thing left before you submit: a final voice check that catches generic phrases creeping into a draft that's supposed to sound like you. Lesson 4.6 is the pre-submit voice check.

Continue to 4.6 → Keeping Your Voice