"Sounds like a bot" is the one complaint that kills outreach quietly. NudgeLink attacks it from both sides: system defenses you get for free, and a voice-learning loop that only works if you feed it.
What the system already does
- No templates, no merge fields — every draft is written from scratch for one person, from their actual profile and posts.
- Anti-repetition — recent openers and closers on your account are tracked and avoided, so your messages don't share one detectable groove.
- Artefact blocking — drafts containing model tells (placeholders, meta-text, refusal phrasing) are rejected before they ever reach your queue, and an AI-likeness score watches drafts continuously.
What only you can do
- Give it your real writing. In Settings → Voice, paste a handful of your own posts, DMs, or emails — your voice profile is extracted from how you actually write, not from a description of it.
- Edit drafts. Every edit is stored as a tone sample that feeds every future draft. Cut anything you would never say out loud, shorten sentences, use your natural sign-off — the system learns the difference between its draft and your version.
- Stay in review mode until drafts stop needing edits. The approval queue is the training loop; auto mode too early locks in a voice that isn't yours yet.
A fair test
Read a draft aloud. If a sentence makes you wince, edit it — that wince is exactly the signal the tone-sample loop needs. Most clients find drafts converge on their voice within a few dozen edited messages.