Two different pressures, one shared cause
The detection-clause pressure is new and getting more common: clients have been burned by contractors handing back lightly-edited AI output, so some now write explicit AI-detection requirements into contracts, sometimes with vague or unrealistic standards attached (see below). The reputational pressure is older and more fundamental: a copywriter's actual product is a distinct voice applied to a client's problem, and if your drafts read as interchangeable with what any assistant would produce unedited, you're competing on price against that assistant, which is a fight you'll lose.
Both problems trace back to the same habit: using AI assistance to speed up drafting (reasonable, common, not something to feel bad about) without a deliberate check afterward for whether the output still sounds like you, with your actual sentence rhythm and phrasing choices, rather than the assistant's default register.
Building the check into delivery
NeverPrompted runs entirely in your browser, nothing you paste ever leaves your machine, which matters for client work under NDA. It checks for a statistical AI watermark, in case a drafting tool you used applied one, and separately scores how far the draft's style sits from a reference corpus of real, varied prose: the specific tells, uniform sentence length, stock transitions, an em dash standing in for a comma, that make copy read as assistant-smoothed rather than crafted.
Where something's flagged, you get a proposed rewrite and a diff, so you're editing your own copy with the tool's help, not handing the client something the tool wrote on its own. That distinction matters for your invoice as much as your conscience: you can genuinely say you reviewed and approved every change. Set the rewrite strength lighter for a client's brand voice that's already close to what you want, or heavier for a rushed first draft that needs more work.
What to tell a client who wants a guarantee
Some briefs now ask for copy that's "guaranteed" against a named AI detector, and it's worth knowing why no honest freelancer, or honest tool, can promise that outright: detectors vary, some check for statistical watermarks applied by a specific AI vendor at generation time, which nobody outside that vendor holds the key to defeating, and others make style-based guesses that can misfire on genuinely human writing too. What you can promise, honestly, is that you checked the copy yourself, reduced the specific AI-style tells you have control over, and can show the client a dated record of that check.
That's what the Pro plan's PDF report is for: a dated record you can attach to a delivery, showing you ran a specific check before handing work over. Free gives you unlimited checking plus a weekly rewrite budget, workable for occasional projects; Pro removes the weekly limit, upgrades the rewriting model, and adds that report, which is the more useful tier once detection-clean delivery is a standing part of your client contracts rather than an occasional request.
Wherever this page describes a result: a detected mark is not proof of authorship, and an absent mark is not proof of human authorship. NeverPrompted's on-device rewrite can reduce detectable evidence but cannot guarantee defeating a vendor's undisclosed watermark, on any tier.