# Stop Sounding Like ChatGPT: An Editing Checklist · ScriptGrain

> A concrete, pass-by-pass checklist for de-flavouring AI-influenced text while keeping your voice, with before-and-after examples for every item.

Canonical: https://scriptgrain.com/blog/stop-sounding-like-chatgpt

# Stop Sounding Like ChatGPT: An Editing Checklist

*By Jack Stovell · 2026-09-10 · Guides*

Here's the thing: your draft doesn't have a ChatGPT problem. It has an editing problem. The model gave you flat, competent, evenly-spaced sentences, and you kept them because they weren't wrong. Fair enough. But "not wrong" is a low bar, and it's not the same as sounding like you.

This isn't a piece about catching AI text or gaming a detector. That's a different, worse game, and I won't play it here. This is about editing back to your own baseline, the way you actually write when nobody's smoothing you out. If you want to see that baseline in numbers, run a sample through the free style tool at scriptgrain.com/tools/writing-style-analysis. It's free, no card, and it'll show you your own sentence-length variance, contraction rate, the lot. Then use the checklist below.

Pass one: vary your sentence lengths on purpose

AI-drafted text loves the medium sentence. Fifteen to twenty-five words, over and over, like a metronome. Humans don't write like that. We write a long one, then three words, then another long one.

Before: "The new pricing model was designed to be more transparent for customers, and it also aimed to reduce the administrative burden on the finance team, which had been a persistent complaint for several quarters."

After: "The new pricing model was meant to be clearer for customers and lighter on finance's admin load. Both problems, one fix. Finance had been complaining about this for months."

Read your paragraph back. If every sentence is roughly the same length, you've got a metronome problem. Break one in half. Let another run long.

Pass two: cut the inflated-significance phrases

"It's important to note." "This represents a significant shift." "In today's fast-paced world." None of these say anything. They're throat-clearing dressed up as insight, and readers skim straight past them.

Before: "It's important to note that this change will significantly impact how teams collaborate going forward."

After: "This changes how teams collaborate."

Say the thing. Don't announce that you're about to say the thing.

Pass three: replace stock transitions with your own connectives

"Furthermore." "Moreover." "In conclusion." These are the connective tissue of a model trained to sound neutral everywhere at once. You don't talk like that, so don't write like that. Use "so", "and to be fair", "that's why", "except". Whatever you'd actually say if you were explaining this to a colleague over coffee.

Before: "Furthermore, the data suggests that customer retention improved after the update."

After: "And retention got better after the update, which, to be fair, nobody expected."

Pass four: restore your contractions

This one's fast and it matters more than people think. AI drafts often sit at a weirdly formal contraction rate, not because the model can't do "don't" or "it's", but because it hedges toward neutral. If you naturally say "wasn't" and the draft says "was not", fix it. Read the sentence aloud. If it sounds like a memo, it's still full of uncontracted verbs.

Before: "We did not anticipate that the rollout would take this long."

After: "We didn't think the rollout would take this long."

Pass five: delete the wrap-up paragraph

You know the one. It restates everything you just said, adds "in conclusion" or a synonym for it, and gives the reader nothing new. AI text does this constantly, like it's worried you weren't paying attention. Cut it. If your ending doesn't land a new point or a sharper version of your opening one, it's dead weight.

Before: "In conclusion, these changes to the workflow will improve efficiency, reduce errors, and ultimately benefit the whole team."

After: (nothing. the piece just ends on the last concrete point you made)

Pass six: reintroduce your specifics

This is the big one, and it's the one people skip because it takes actual memory, not just editing. AI drafts default to vague nouns: "a client", "a recent study", "significant improvement". You know things it doesn't. You know the client was called Dave and the meeting ran over by forty minutes. Put that back in.

Before: "A client recently mentioned that the process felt much smoother."

After: "Dave, from the Leeds office, told me the new process saved him about twenty minutes a week. That's the whole pitch, really."

Don't invent numbers to sound precise, that's its own kind of fake. Only use the ones you actually know.

Pass seven: read it aloud

Last pass, always. Not skimming with your eyes, actually reading it out loud, in the room, to nobody. Every phrase that makes you stumble, every sentence you have to re-say to get the rhythm right, mark it. That's where the AI flatness is hiding. Your mouth catches what your eyes miss.

Why this matters more than detection

None of this is about beating a detector. Detectors are a symptom-checker, not the point. The point is that flat, generic prose reads badly whether or not a machine flagged it, and readers notice before any tool does. Editing toward your own measured baseline, the sentence lengths you actually use, the contraction rate you actually have, gets you writing that sounds like you again. That's what the free analysis is for: not a pass/fail stamp, just your numbers, so you know what "sounds like you" actually measures.

Run through these seven passes on your next draft. It's slower than publishing straight out of the box. But at the end of the day, that's the whole job: making it yours before it goes out under your name.

[More from the ScriptGrain Journal](https://scriptgrain.com/blog)