The useful way to think about Willow isn't "a better Wispr Flow." It's a different bet about what voice input is for. Most dictation tools treat the keyboard replacement as the whole job: say words, get text, done. Willow's Scribe feature treats that as only half the job — the harder, more valuable half is turning a spoken instruction into finished, context-appropriate writing, the way a good assistant would interpret "tell him 7 works" rather than transcribing it literally into a message that reads like a transcript of a transcript.

That's a genuinely different product thesis, and it's well-supported by the architecture underneath it. Separating ASR from a purpose-trained edit model, and optimizing specifically for edit rate rather than raw transcription accuracy, reflects real thought about what actually frustrates a daily dictation user — it's rarely the occasional misheard word, it's the repeated manual cleanup afterward.

But there's a real tension built into this approach, and it deserves more weight than a single bullet point in a pros/cons list. The more an AI tries to infer intent rather than transcribe literally, the more opportunities it creates to guess wrong about which mode you're in. The documented case of a long-time user — 78 days, over 340,000 words — eventually leaving specifically because Willow interpreted a literal word as an editing command isn't a minor anecdote. It's close to a worked example of exactly the failure mode this review predicted before finding that evidence: context-aware systems trade transcription literalism for interpretive risk, and that tradeoff doesn't disappear just because it usually works.

None of that makes Scribe a bad idea. It makes it a feature that needs continued refinement, and a feature users should understand the shape of before relying on it for anything where literal accuracy matters more than inferred intent — dictating a legal clause, say, versus drafting a casual text reply.

The free tier is the other half of why this review lands where it does. Making real, unlimited core dictation free — not a seven-day trial, not a capped word count, an actual non-expiring free plan — is a genuinely strong move in a category where most competitors gate basic usability behind a paywall. It's good for users. It's also, bluntly, good marketing timed well, and both things can be true at once.

Weigh the company-reported numbers accordingly. 98%+ accuracy and "3x fewer errors than built-in dictation" sound precise. They come from Willow's own testing, including on its own comparison page against a named competitor — informative as a directional signal, not as independently verified fact. The honest framing: Willow is a genuinely interesting, well-architected product from a young, still-proving-itself company, with one standout feature that is simultaneously its best argument and its clearest risk.