Start here
Paste a model draft. It names the tells a model left behind and rewrites around them, without changing a fact.
It is important to note that we should leverage this opportunity to delve into a robust, seamless, and scalable approach.
We should look at this properly. Here is an approach that holds up.
The draft is set in a monospace and the rewrite in a reading face. That is the whole product in one picture: the facts survive, the machine shape does not.
The 60-second path
- Open the app. No account needed for the first three rewrites a day.
- Paste a draft, pick a register — Slack, email, skip-level, LinkedIn.
- Read the tells report beside it. Every flagged span is a thing a model does and a person mostly does not.
- Press rewrite. Compare the two, then take what you want.
What it does not do
- It does not add facts, and it does not remove them. If a fact is missing from your draft, it will be missing from the rewrite. The fact audit beside the result is there so you can check that claim rather than trust it.
- It does not store your drafts. Text posted to the rewrite endpoint is held in memory for the length of the request. See Privacy.
- It does not defeat AI detectors, and it is not sold as doing that. It removes the patterns that make prose read as machine-written. Those overlap with what detectors look for; the overlap is not a guarantee, and nobody can honestly offer one.
- It is one voice, one engine, one meter. Every client — web, extension, curl —
hits the same
POST /api/humanize.
Where to go next
| You want to | Read |
|---|---|
| Rewrite something well | Rewriting a draft |
| Understand a flagged span | Tells and the report |
| Score a draft without rewriting it | Detect |
| Make it sound like you specifically | Voice profiles |
| Call it from your own code | HTTP API |
| Rewrite inside Gmail or LinkedIn | Chrome extension |
| Know what a plan gets you | Plans and quota |