Willow Voice

Dictation has been almost good enough for about thirty years, and the reason it never stuck is not the transcription.

It is that the text always appeared somewhere else. You spoke into an application, corrected the punctuation, then copied the result into the place you actually needed it — and a three-step process loses to typing, every time, no matter how fast the middle step gets.

Willow Voice's answer removes the steps rather than speeding them up. Put your cursor where you want the words, hold a key, speak, and the formatted text is already there.

Willow Voice: Dictation That Sticks

• It works in any text field — Slack, Gmail, Cursor, Notion, iMessage — rather than in an application you copy from.
• Hold a hotkey and speak; formatted text appears where the cursor already was, with no switching or cleanup.
• Punctuation and filler words are handled for you, because nobody speaks in punctuation.
• Willow Scribe turns rough spoken intent into polished writing rather than only transcribing it.
• Mac and iOS, with the site stating 100,000+ professionals and a dedicated security page.

Why Has Dictation Never Worked Before?

Willow describes a three-step interaction: place your cursor in any text field, press a hotkey and speak naturally, and perfectly formatted text appears where the cursor is.

The important word is *any*. Slack, Gmail, Cursor, Notion, iMessage — anywhere you would normally type, which means dictation stops being a destination and becomes an input method.

That is the difference from every previous generation. A dictation application is somewhere you go; an input method is something you use without thinking about it, and only the second becomes a habit.

The second half is the formatting. Speaking naturally means speaking without punctuation, with restarts and hesitations and the occasional abandoned sentence, and older dictation faithfully transcribed all of it — leaving you to clean up text you could have typed correctly in the first place.

Handling that automatically is what makes the output usable without editing, and it is the claim to test hardest. The gap between raw transcription and something you would actually send is where every previous attempt at this failed.

The contrast with a meeting notetaker is worth drawing. Wave captures what other people said; this produces what you meant to write, which is a different product entirely.

Dictation applications are somewhere you go. An input method is something you use without thinking. Only the second ever becomes a habit.

Willow Scribe Turns Intent Into Writing

Willow Scribe is described as AI writing from your intent: describe roughly what you want to say, and it produces polished writing in your voice, in place.

That is a meaningfully different product from dictation, and worth separating. Dictation writes down what you said; Scribe writes what you meant, which is a much larger claim and a much larger risk.

The case for it is real. Most people know what they want a message to convey and lose ten minutes constructing sentences, and the constructing is not where the value is. Describing the intent in twenty rough seconds and getting a draft is a genuine reallocation of effort.

The risk is equally real and specific: the output is your writing with your name on it, sent to someone who will hold you to it. A dictated message you spoke is yours by definition; a generated one from a rough description is yours only if you read it.

The practical rule follows. Use Scribe for the messages where the words do not matter much and read every one before sending, and dictate the messages where the words do. That division is easy to make in the moment and easy to forget.

Working In Every App Is The Whole Point

Willow states it works in every app on Mac and iOS, and the practical significance of that is worth spelling out with examples.

The applications listed are a fair sample of a working day: Slack for internal conversation, Gmail for external, Cursor for code, Notion for documents, iMessage for personal. A dictation tool that covered three of those and not the others would be useless, because the habit never forms if it sometimes does not work.

Cursor is the interesting inclusion, because prompting an AI is a category the site names specifically. Prompts are long, conversational and awkward to type — you are describing what you want in prose — which makes them close to an ideal case for speaking.

That is arguably the strongest use of the tool and one nobody predicted three years ago. The volume of prose people now type into AI tools is enormous, and it is exactly the shape of writing that speaking suits.

The constraint is physical rather than technical. Dictation requires a room where you can speak, which rules out open-plan offices, shared spaces, trains and anywhere a colleague is on a call — and that is a substantial fraction of where work happens.

It is also the exact inverse of a narration tool. Narakeet turns your text into speech; this turns your speech into text, and only one of them removes work from a working day.

• Use it for prompts and messages first, where speaking suits the writing.
• Read anything Scribe produces before sending it, since it is your words legally.
• Give it a week — the habit forms around day four or not at all.
• Check accuracy on your own product names and jargon before relying on it.
• Accept that it needs a room where you can talk, which excludes a lot of offices.

Building A Dictation Habit In A Week

Willow's own framing is three seconds to start, a habit in a week, and the second half is the harder claim.

New input methods fail on muscle memory rather than on capability. Typing is automatic; reaching for a hotkey is a decision, and decisions lose to habits until they become one.

1. Pick one context to start with — AI prompts are the easiest, since they are long and nobody reads them but the model.
2. Use it there exclusively for three days, even when typing would be faster, to build the reflex.
3. Note what it gets wrong on your own vocabulary, and check whether product names and jargon come out correctly.
4. Add a second context in week two — Slack messages are the natural next step.
5. Keep Scribe separate from dictation in your head, and read everything it writes before it goes out.
6. Judge after two weeks by whether you reach for it without deciding to, which is the only test that matters.

Where Speaking Beats Typing, And Where It Does Not

Being specific about this saves a fortnight of finding out, because the answer is not uniform across kinds of writing.

Speaking wins for anything conversational and long: messages, emails, AI prompts, first drafts, anything where you already know what you want to say and the typing is pure transcription of your own thoughts.

It also wins on volume. Speech runs at roughly three times typing speed for most people, which is why the four-times-faster claim on the site is plausible for the right kind of writing rather than obviously inflated.

Typing wins for anything structured or precise. Code, formulas, anything with symbols, anything where you are thinking as you write rather than transcribing a finished thought. Writing is also a thinking process, and speaking a paragraph you have not worked out produces a paragraph nobody has worked out.

The honest summary is that dictation is excellent at getting known thoughts out of your head and poor at helping you have them. Which is fine — most of what anyone types in a day is the first kind.

For messages where you would rather generate than dictate, a purpose-built writer such as WriteMail.ai sits in the same gap from the other side.

Willow Voice: Platforms And Reach