thirdraft
Thirdraft vs AI humanizers
An AI humanizer tries to make AI text sound like a person wrote it. Thirdraft tries to make it sound like you wrote it. Those sound close. In practice they're different jobs, and the difference matters if your name is on the result.
What humanizers do
Most humanizers rewrite AI text to strip out the patterns that AI detectors and readers look for. Many lead with detector scores. Lunchbreak's homepage shows a draft scored 94% AI on Turnitin and 98% on GPTZero before its rewrite, and 0% and 2% after. Walter calls itself "the AI authenticity platform" and gives you an authenticity score from its own AI detector after you rewrite. Both also promise writing that sounds like you wrote it. Their pages lead with detection though.
So the test for that kind of tool is a detector score, or "does this read like a human." Any human.
What Thirdraft does
Thirdraft starts from your own writing. You give it a few things you wrote and it learns your habits: how long your sentences run, how you open and sign off, your punctuation, the words you use and the ones you never do. When you rewrite a draft, it picks the samples of yours closest to that kind of writing and rewrites toward them.
The test is whether it sounds like you, judged by you and the people who know how you write. Thirdraft doesn't check detector scores and isn't built to pass them.
That's the test that matters at work, too. An ACL 2025 study found that people who use LLMs a lot spot AI text better than most detectors do. Your coworkers aren't running a detector. They're reading.
Side by side
| Typical AI humanizer | Thirdraft | |
|---|---|---|
| Goal | Make AI text read as human-written, often to lower a detector score | Make a draft read like you wrote it |
| What it targets | Patterns that detectors and readers link to AI | Your own patterns, learned from your writing samples |
| How you judge success | A detector score, or "sounds human" | You, and the people who know your writing, recognize it as yours |
| Learns from your writing | Varies, check each tool | Yes, from samples you add and can delete |
| Em dashes | Often removed for everyone | Kept if you use them, left out if you don't |
| Works inside Claude or Cursor via MCP | Varies, check each tool | Yes: Claude, Claude Code, Cursor, VS Code and other MCP clients |
| Checks AI detector scores | Often, as the headline feature | No |
The em dash problem
The em dash became the best-known AI tell, because chatbots use it constantly. That's rough on people who always used them. Writers who've used em dashes for years now get told their writing looks like AI.
Generic humanizers tend to handle this with one rule for everybody: strip the dashes. That fixes it for people who never used them and erases a real habit for people who did.
Thirdraft follows your samples instead. If you use em dashes, it keeps them. If you don't, it stops adding them. Same goes for semicolons, exclamation marks and the rest of your punctuation.
Which one do you want?
If you need text to get past a detector, Thirdraft is the wrong tool. Better to know that now. If you use AI for first drafts and want the final version to sound like you wrote it, that's what Thirdraft is for. If you'd rather do it by hand first, here's how to make ChatGPT sound like you.
Sources
Questions
Will it pass AI detectors?
Thirdraft isn't built for that, and we don't test against detectors. It rewrites toward how you write. Detectors make their own guesses and disagree with each other a lot, so we don't promise any score. If passing a detector is the goal, this isn't the tool for it.
Is Thirdraft an AI humanizer?
Not in the usual sense. It doesn't aim for generic human writing. It aims for yours, and it needs a few samples of your writing to do it.
Can't humanizers match my style too?
Some say they do. Check whether the tool learns from writing you give it, or applies the same rewrite to everyone. Thirdraft only works from your own samples, and you can see and delete every one of them.
Will it keep my em dashes?
If your samples use them, yes. If they don't, it won't add them.
Does it work inside Claude or Cursor?
Yes. Thirdraft runs as an MCP server, so Claude, Claude Code, Cursor and other MCP clients can rewrite what they wrote for you in your voice. There's a guide for using it on PR descriptions.
Hear your own voice back
Add a few things you've written and rewrite an AI draft. 3 free rewrites a week, no card needed.
Try it free