What we think we’re actually doing.
Most AI text today is technically correct and emotionally hollow. It uses six words where two would do, opens with “in today’s rapidly evolving landscape,” and leans on the same handful of intensifiers in the same handful of places. People — even people who can’t name the tell — can usually sense it.
HumanizeMyWriting rewrites that text so it reads like a person wrote it. Not because the user is trying to fool anyone, but because most of the time the underlying ideas are theirs — the AI was just scaffolding. The product exists for the gap between “I outlined this” and “I’d be proud to put my name on this.”
What this is for
We built the humanizer for writers, students, marketers, ESL speakers, and anyone whose first draft (AI-assisted or not) doesn’t yet sound like them. We think those use cases are obviously legitimate — that’s how editors and ghostwriters have worked forever.
What we’re honest about
Some users will pass humanized AI text off as their own original thinking in places where original thinking is explicitly being evaluated. We can’t prevent that, and we don’t pretend the detection arms race is a misunderstanding — it’s real, and we’re participants in it. If you’re submitting work to a course or publication that prohibits AI-assisted writing, this product probably isn’t for that submission.
What we won’t do
- Train models on your prose — ever. Not the humanizer, not the detector, not the voice analyzer.
- Sell, trade, or share your text with anyone. The structured logs we keep contain word counts, latency, and model name — never prose.
- Build features that target specific institutions’ detection setups. We’re making writing read more naturally; we’re not in the business of name-by-name detector evasion.
- Market the product around dishonesty. The landing copy says “bypass every detector” because that’s a real outcome of writing that reads like a human wrote it — not because we want you to deceive someone you owe authentic work to.
The bet
Detection software has a structural ceiling. As AI text gets better, it converges on the distributional middle of human writing — and you can’t reliably flag “sounds like a competent human” as AI without flagging actual competent humans too. Eventually the distinction will matter less than whether the work is good and the author endorses it. Until then, we’re shipping a tool that’s genuinely useful to writers and transparently uncomfortable for graders who relied on detectors as a shortcut.
See also: privacy for what we store; terms for what you and we agree to.