# How to Make AI Writing Sound Human: The De-AI Playbook

> Kill the em dash and the dozen quieter AI tells. The de-AI playbook for tweets and posts, plus why symptom-scrubbing never fully works.

URL: https://voicemoat.com/blog/make-ai-writing-sound-human

## Key facts
- Published: 2026-07-30
- Category: AI and Voice
- Read time: 19 minutes
- Author: VoiceMoat team
- Publisher: VoiceMoat
- Topics: Making AI writing sound human; AI writing tells and detection; Em dashes as an AI giveaway; Voice-matched AI drafting for social media; De-AI editing workflow for tweets and LinkedIn posts

## Key points
- The em dash is the single most recognizable AI tell right now, but it is one symptom of many: removing it alone does not make a draft read as human.
- When you tell ChatGPT to stop using em dashes, you must name a replacement (comma, period, or the words 'and' and 'but'), or the model substitutes semicolons and colons and the rhythm still reads as machine-made.
- A dozen quieter tells survive after the dash is gone: the 'it's not X, it's Y' construction, cliche verbs, symmetrical three-item lists, bold-term formatting, predictable emoji, and eerily uniform sentence rhythm.
- Manual de-AI editing works on one post and quietly fails across a posting cadence: you cannot hand-edit rhythm and cliche out of 20 posts a week and stay consistent, so the tells leak back in.
- VoiceMoat is an AI social media tool for Twitter and LinkedIn (create, schedule, engage, analyze, grow) whose audenAI writing partner learns your own past posts first, so drafts start in your voice instead of arriving as average-AI text you then de-AI by hand. Disclosure: VoiceMoat is our own product.

## Related articles
- Why Your AI Tweets Sound Generic or Robotic (And How to Fix It): https://voicemoat.com/blog/why-ai-tweets-sound-generic
- Why Every AI LinkedIn Post Sounds the Same (and How to Break the Pattern): https://voicemoat.com/blog/why-ai-linkedin-posts-sound-the-same
- Why every AI draft you write sounds the same: https://voicemoat.com/blog/why-ai-drafts-sound-the-same
- How to train AI to write in your voice: the cross-model guide: https://voicemoat.com/blog/how-to-train-ai-to-write-in-your-voice
- How to make ChatGPT write tweets in your voice in 2026 (and the faster way): https://voicemoat.com/blog/make-chatgpt-write-tweets-in-your-voice-2026

## FAQ
Q: Does ChatGPT sound human when writing?
A: It can, but not by default. Out of the box ChatGPT writes toward the statistical average of everyone's text, which is why the tells cluster: over-frequent em dashes, the 'it's not X, it's Y' construction, cliche verbs, symmetrical lists, and eerily even sentence rhythm. With strong custom instructions, persona prompting, and manual editing you can push a single draft to read as human, but the effort has to be repeated on every post because the model has no persistent memory of your voice.

Q: Can everyone really tell my post was written by ChatGPT?
A: Often, yes, and faster on short posts. A tweet has so few words that any single tell, like one em dash or one balanced triad, makes up a large share of the text and stands out. Readers now carry an unconscious AI-detector that fires on the dash, the cliche verbs, and the uniform rhythm. Removing the em dash helps but is not enough on its own, because the dozen quieter tells survive after the dash is gone.

Q: How do I use AI for Twitter growth without sounding robotic?
A: Never ship the raw draft. Strip the em dashes and name a replacement (period, comma, or 'and'/'but'), kill the 'it's not X, it's Y' pattern and the cliche verbs, break the symmetrical lists, vary your sentence length hard, and add one specific detail only you would know. That works per post. If you publish at volume, hand-editing every post stays consistent only if the tool drafts in your voice to begin with, so there is far less to correct.

Q: Why does my AI writing have no personality?
A: Because a general model is built to predict the most likely next words across everyone's writing, so its safest output is the middle of all of it, and the middle has no personality by definition. Personality lives in the specifics a model does not have: your rhythm, the words you never use, the exact number from your week, the joke only you would make. You add those back by hand, or you use a model conditioned on your own past posts so they are there from the first draft.

Q: Are em dashes proof that something was written by AI?
A: No. Many careful human writers use em dashes, and treating the mark as proof of AI punishes good writing. The actual tell is over-frequency: a default model reaches for a dash in almost every paragraph, where a human uses one occasionally. On public social posts it is still worth removing, because perception moves faster than fairness, but the dash alone is not a reliable signal and removing it alone does not make a draft read as human.

Q: Is there a tool that writes in my voice so I do not have to de-AI every post?
A: Yes. VoiceMoat is an AI social media tool for Twitter and LinkedIn whose writing partner, audenAI, trains on your own past posts and models your rhythm, vocabulary, and punctuation, so drafts start in your voice instead of arriving as generic AI text you then edit by hand. It is free to start (paid plans from $35/mo), and disclosure: VoiceMoat is our own product. audenAI suggests, you decide.

Source: https://voicemoat.com/blog/make-ai-writing-sound-human
Site overview: https://voicemoat.com/llms.txt. AI usage policy: https://voicemoat.com/ai.txt.