InvisiWrite started with an argument about a number. A colleague's report came back from a detector at 82% AI — a report she had written herself, at a kitchen table, over two evenings. The number was wrong, and there was no way to see why it was wrong.
AI detection arrived faster than anyone's ability to interpret it. Universities bought it, employers bought it, and both started making decisions on a single percentage with no working shown.
Every figure we produce comes with the sentences that drove it and the reason each was flagged. If you cannot argue with a number, it should not be used to judge you.
Formal, careful, non-idiomatic English gets flagged hardest — which describes almost everyone writing in a second language. We say so in the product rather than hiding it in a support article.
Making a stiff paragraph readable is ordinary editorial work. We build for that, and not for concealing where something came from.
Never used for training, never sold, deleted on the schedule you pick. Business accounts get stricter retention, not looser.
We would rather lose a sale than end up in someone's misconduct hearing, so here is the position in plain terms.
What this is for. Checking how your own writing reads, editing drafts you produced, and understanding a score before someone else acts on it.
What we will not build. Watermark removal, keystroke replay into document editors, or anything else designed to fabricate a record of how work was produced.
Where institutional rules apply. Universities and employers set their own policies on AI assistance, and they differ. Those rules take precedence over anything we say. Read them.
What we will not claim. That any rewrite is guaranteed to pass any detector. Nobody can promise that honestly, and the ones who do are selling certainty they cannot deliver.
Small, and intending to stay that way for a while.
Replace these with real names and photographs before launch. Placeholder team sections are the fastest way to lose a business customer's trust.
Not on a sample. Paste in your own work and see whether the reasoning holds up.
Open the detector