4.6 — Keeping Your Voice in the Final Draft | AI for Real Life

Even when you wrote it, it can still read like AI.

Here's the trap most students don't see coming. You did the work. You wrote your own draft. You ran the Edit-Don't-Write workflow from Lesson 4.5: Claude critiqued, you revised. You feel good. Then your prof reads it and gets that same wary look they get when they think a paper was AI-generated.

What happened: Claude's critiques contain phrasings you absorbed. The polishing pass you did to make sentences "clean and clear" scrubbed out the personal, off-rhythm sentences that read as human. The result: a paper you 100% wrote that sounds like AI wrote it. This lesson is the pre-submit check that fixes that.

About AI detectors: they're not reliable, and that's not the point

Detector software has high false-positive rates, especially on non-native English writers and on polished writing of any kind. Don't write to fool a detector. Write to sound like a person who thought about this.

The 2026 dead-giveaway list.

What follows is the working list of AI-style patterns that current professors and TAs flag in 2026. It will keep evolving, but the categories don't. Words, sentence patterns, paper-level structure.

🪦

Phrases & words to ban

Words: delve, multifaceted, intricate, nuanced (especially stacked together), underscores, tapestry, navigate the landscape, plays a crucial role, stands as a testament.
Openers: "In today's society," "In the modern world," "Throughout history," "It is important to note that…"
Connectives: sentences starting with Moreover, Furthermore, In addition, In conclusion.

🔁

Sentence patterns

The 2026 signature: "It's not just X — it's Y." Tortured perfect parallelism. Em-dashes used as commas more than 3 times in the whole paper. Paragraphs that end by summarizing themselves ("the summary sandwich"). Hedge stacking ("perhaps, in some sense, one could argue…"). Bullet-list-itis where prose belongs.

🧱

Structural defaults

Intros that announce the plan in three bullet sentences ("This paper will… It will then… Finally, it will…"). Uniform paragraph length and structure across the whole paper. Reflexive "balanced perspective" hedges in places where the prompt didn't ask for both sides. "In conclusion" at the start of the closer.

The Pre-Submit Scrub: run on every paper.

Run this after you've revised everything else. The point is to make Claude tell you which sentences are making YOUR writing sound like AI writing.

The Pre-Submit Scrub

[Voice Profile at top.] Scan this draft for AI-style patterns. Don't rewrite anything. Just flag, quote the exact sentence and tell me what's wrong with it. Specifically check for: WORDS: delve, multifaceted, intricate, nuanced (when stacked), underscores, tapestry, navigate the landscape, plays a crucial role, stands as a testament, in today's society, in the modern world. OPENERS: any sentence starting with Moreover / Furthermore / In addition / It is important to note / In conclusion. Any intro paragraph that announces "this paper will." PATTERNS: any "It's not just X, it's Y" construction. Em-dashes used more than 3 times in the whole paper. Three-or-more-item lists with perfect parallel structure. Paragraphs that end by summarizing themselves. VOICE DRIFT: any sentence that sounds more like generic AI-paper than my voice profile. Output: a numbered list. For each flag, give me (1) the exact sentence quoted, (2) which pattern it triggered, (3) one sentence on why it's a problem. DON'T propose replacements. I'll rewrite each one in my voice. Be aggressive. False positives are fine. I'd rather over-scrub than miss something.

What to do with the flagged list

Open your draft. Rewrite each flagged sentence in your own words. Do NOT ask Claude to rewrite them. Rewrite them yourself, and aim for shorter, more specific, and friend-conversation-ready. If you can read the new sentence aloud and it sounds like something you'd actually say, you've fixed the issue.

The deeper fix: rewrite for human-ness, not just to scrub patterns.

Scrubbing patterns is a defensive move. The offensive move is rewriting toward sounding human. Two quick passes that matter more than any single word swap.

Three quick passes for human-ness

1 · Say it out loud. Read each paragraph aloud. Any sentence that doesn't fit your normal speaking rhythm gets rewritten.
2 · Keep the structurally-odd sentences you naturally wrote. When you draft, you'll sometimes write a sentence that's structurally odd (not incorrect): a short one, an unusually long one, one that starts with "And." Don't edit those out just because they break a pattern. They're part of how you actually write, and they're a big part of why the paper reads as yours.
3 · Specific over general. Replace adjective stacks ("complex, multifaceted, intricate") with one specific noun ("paradox," "knot," "trade-off"). Fewer words, more meaning, harder to fake.

The "abstraction → specific" pass

Here's my draft: [paste]. Don't rewrite. Find the 5 most abstract sentences in the paper, sentences that sound profound but don't actually say a specific thing. Quote them. For each, ask me a single concrete question that would force me to commit to a specific claim instead of an abstraction. Examples of the kind of question I want: "What specifically?" "Says who?" "When?" "Whose example?" "What would change if this weren't true?" I'll answer your questions and rewrite the sentences myself. Don't write anything for me; just point me at the abstractions.

Why this scrub matters

Honest Work Code · Rule 1: Learn with it, not instead of it. Scrubbing AI-style patterns isn't only about how the paper looks. It's about confirming you actually did the work.

Phase 2 continues: Module 5: Research, Reading & Critical Thinking.

Module 4 is the integrity flagship. Module 5 is the research flagship. The hallucinated-citation trap, the source-checking workflow, speed-reading dense PDFs, synthesizing across sources, and using Claude as a sparring partner that pushes back instead of agreeing. Same Honest Work Code, applied to research instead of writing.

Continue to Module 5 →