Getting Started

Your First 7 Days With Voice Dictation: The Habit Plan That Makes It Stick

The way to make voice dictation stick is one small step a day for a week: one message on day one, every reply on day two, one full email on day three, and so on until speaking is your default. Almost nobody quits dictation because it does not work; they quit in the first week, while typing is still automatic. With AirTypes, Plain is free and offline, so there is no clock on the week, and Polished cleans up your words when you want it, at 120 minutes of speech for every dollar.

This is what you are building toward. You hold a hotkey in your email client, talk the way you normally talk, and let go:

You say

"ok so um the meeting is moving to tomorrow at ten, client is not available, and uh send him the invoice again"

→
Plain types (free, offline)

"The meeting is moving to tomorrow at ten, client is not available, and uh send him the invoice again."

Five seconds of speech, typed at your cursor, transcribed on your own computer. The stray "uh" is the honest cost of free, offline dictation; on day five you will see what Polished does with the same sentence. None of this is hard. What is hard is the habit, and that is what the next seven days are for.

Why week one decides everything

Typing is not a skill you think about. You have been doing it for years, and the path from thought to text runs through your fingers without conscious effort. Dictation asks you to build a second path, from thought to speech to text, and for the first few days that path is slower simply because it is new.

This is the whole problem. Not accuracy. Not setup. The gap between "this is obviously faster in principle" and "this is faster for me right now" is about three to five days of ordinary use. People who cross it tend to keep dictating. People who judge the tool on day one usually go back to typing and conclude the technology is not ready.

So the plan below is deliberately small. Each day asks for one thing, and each thing is short enough that you cannot really fail at it. Commonly cited figures put speaking at around 130 words per minute and typing at roughly 40. That difference is waiting on the other side of a week of small steps.

There is no deadline here. AirTypes Plain is free with no time limit, so if day three takes you three days, that is fine. The seven days are about the habit, not a countdown.

Day 1: Dictate one message. That is all.

Install AirTypes on macOS or Linux, create the free account it asks for at first launch, and pick a Plain voice model. On a Mac, allow Microphone and Accessibility when asked; without Accessibility, AirTypes can hear you but cannot type into other apps. Then dictate exactly one thing: a Slack reply, a text message, a comment on a pull request. Something under thirty words that you were going to write anyway. On Windows, where the desktop app is still in development, you can follow the plan with Dictate in your browser: click the mic, speak, and paste the Polished text. It uses credit after the first free minute.

Do not open a blank document and try to dictate an essay. That is the single most common day-one mistake and it fails for the same reason a blank page fails when you type: the problem is composition, not input.

The mechanic to learn today: hold Ctrl + Shift + Space, speak, release. The text lands at your cursor in whatever app has focus. There is no window to switch to and no transcript to copy. Once your hands know that rhythm, everything else in this plan is repetition.

Two setup details that pay for themselves immediately. First, make sure the hotkey does not collide with your editor or browser shortcuts; change it in Settings if it does, and switch to Press to Toggle if holding three keys feels awkward. Second, use a headset or the built-in mic in a quiet room rather than a laptop mic next to a fan. Everything about accuracy starts with clean input audio. If you need help with any step, how to use AirTypes walks through setup screen by screen.

Day 2: Move every reply to voice

Today, one rule: every chat message, comment, and short reply gets dictated instead of typed. Slack, Discord, two-line email replies, code review comments, ticket updates.

Short bursts are the fastest way to build the habit because the feedback loop is tight. You speak for four seconds, you see the result, you adjust. Twenty of those in a day teach you more than one long dictation session.

You will notice something on day two: you start speaking in complete sentences. That is the actual skill being built. Not enunciation, not speed: thinking one whole sentence ahead of your mouth. It arrives on its own after a few dozen reps.

Day 3: One email, one take

Pick one email you owe someone and dictate the entire thing in a single pass. Do not stop to fix a word. Do not restart a sentence because you phrased it awkwardly. Speak it end to end, then read it once and edit.

This is the highest-leverage exercise in the whole week. Correcting mid-sentence is the habit that makes dictation feel slow. It forces you to switch between speaking mode and editing mode every few seconds, and that switch costs more than the errors do.

Speak first, edit once. Almost everyone who says "dictation is not faster for me" is correcting mid-sentence. Press to Toggle helps here: you press once, talk through the whole email with your hands free, and press again.

Day 4: Tune your accuracy tier

By now you have a feel for where the transcription is strong and where it slips, usually on names, product terms, and technical jargon. Today is the day to change the model tier rather than change how you speak.

Plain runs OpenAI's Whisper on your own CPU, and you choose between two tiers in Settings: Swift (~78 MB), the fastest, and Crisp (~148 MB), the one we recommend. The trade-off is straightforward: the bigger model is more accurate on accents, domain vocabulary, and noisy rooms, and it takes a little longer to process each utterance. Plain runs locally on both, so the only cost of Crisp is time and CPU, never privacy. Both run on a 4 GB machine with no GPU.

Spend ten minutes today dictating the same three sentences of your real vocabulary on both tiers. Whichever one gets your jargon right without feeling sluggish is your tier. Set it and stop thinking about it. If neither gets your jargon right, that is a job for Polished, which you try tomorrow.

For the deeper explanation of what changes between tiers, see Whisper model sizes explained and how offline speech recognition works.

Start your week with Plain, free

AirTypes Plain is free, offline and has no time limit. Hold a hotkey, speak, and your words are typed in any app on macOS or Linux.

Download AirTypes

Day 5: Let Polished clean it up

Plain is faithful to what you said, including the "uh" in the middle and the sentence you abandoned halfway. For your own notes that is fine. For a message a client or your manager reads, you may want it cleaner. That is what Polished is for, and today is a good day to try it.

Polished sends the recording to AirTypes to transcribe and clean it up. A large speech model transcribes it, then an AI removes filler words and false starts and fixes grammar and punctuation, keeping your meaning, numbers and names. Here is the same sentence from the top of this post:

You say

"ok so um the meeting is moving to tomorrow at ten, client is not available, and uh send him the invoice again"

→
Polished types

"The meeting is moving to tomorrow at 10:00. The client is not available. Send him the invoice again."

Switch modes on the Home screen. Your first minute of Polished is free, then it is prepaid credit: $5 buys 600 minutes of speech, credit never expires, and there is no subscription. When credit runs out, Polished pauses and Plain keeps working. You add credit from inside the app; a payment page opens in your browser.

The habit rule for today: use Polished for messages other people read, keep Plain for everything else. Two modes you understand beat constant switching. Two more things worth knowing:

  • It works in the language you speak. Plain is English only. Polished handles English, Hindi, Bangla, Japanese, German and most other spoken languages, so try yours. The Always translate to English setting lets you speak your language and get English typed.
  • It has a guard. AirTypes checks that the rewrite keeps every number you said and stays close to your length; if it drifts too far, you get your own words instead. With translation on the guard checks numbers only, so check names before you send.

If you dictate prompts to AI tools, the same split applies. See dictating prompts to ChatGPT, Claude and Gemini.

Day 6: Dictate a first draft

Today, take the longest piece of writing on your plate (a document, a spec, meeting notes, a report section) and speak the first draft. All of it. Badly, if necessary.

The point is not that the draft will be good. The point is that editing something rough is a fundamentally easier task than producing something polished from an empty page, and dictation is the fastest way to get to "rough".

Two things usually surprise people at this step: the draft takes less time than they expected, and it is longer than what they would have typed. Both are normal. Speech has less friction, so more of what you were thinking actually makes it onto the page. Plain is a good fit for drafts, since you will edit anyway; if the draft is going straight to someone, Polished saves you the cleanup pass.

Day 7: Read your stats, lock the habit

Open the Account tab in AirTypes and look at the week: transcriptions, words dictated, estimated time saved, and the activity chart. Numbers matter here because the improvement is invisible day to day. You do not feel the four minutes you saved on an email, but you notice a weekly word count in the thousands.

Then make one decision and stop there: pick the single workflow that stays on voice permanently. For most people it is chat replies or email. For developers it is commit messages, code comments and prompts to coding agents. For researchers it is notes. One workflow, permanently, is what a habit is made of.

There is nothing to cancel and nothing that expires today. Plain stays free. If Polished earned its place on day five, keep it for the messages that matter; if not, you have lost nothing.

The four mistakes that stall people

  1. Correcting mid-sentence. The biggest one. Speak the whole thought, then edit once. Mid-sentence correction is what makes dictation feel slower than typing.
  2. Starting with the hardest text. Do not learn dictation on a legal document or a technical spec. Learn it on Slack messages where the cost of an error is zero.
  3. Bad audio input. A laptop mic in a room with a fan, or a headset with the boom pointed at your chin, will make any speech engine look bad. Fix the input before you blame the model.
  4. Tweaking settings instead of talking. Trying both model tiers, every hotkey and every mode on day one feels productive. It is not the habit. Pick a tier, pick a mode, and use them for a week.

What happens after week one

Dictation tends to spread rather than fade, because each new place you try it (terminal, design tool, notes app, AI chat) works the same way. The habit generalizes instead of staying stuck in one app.

A few weeks in, most people stop thinking about the hotkey at all. That is the finish line. From there, the interesting question stops being "is dictation faster" and becomes "what else can I do now that speaking is my input": prompting ChatGPT, Claude, and Gemini by voice, talking to your coding agent, drafting in your own language with Polished, or working with your hands off the keyboard when your wrists need a break (see dictation for RSI).

None of that is available in week one. All of it is available in week four, and the only thing standing between the two is seven small days.

FAQ

How long does it take to get used to voice dictation?

Most people are comfortable within three to five days of daily use. The first day feels awkward because you are learning to speak in complete thoughts, not because the software is inaccurate. By the end of week one, dictation usually feels faster than typing for messages, emails, and first drafts.

Do I need to speak slowly for voice to text to work?

No. Modern Whisper-based models are trained on natural conversational speech, so speaking at your normal pace produces better results than slow, over-enunciated speech. What helps is speaking in full sentences rather than stopping and restarting mid-phrase.

What should I dictate first?

Start with short, low-stakes text: a chat reply, a commit message, a comment. Short bursts teach you the hold-to-speak rhythm without the pressure of composing a long document by voice. Save documents for day six.

Is offline voice dictation as accurate as cloud dictation?

For clear speech on a decent microphone, on-device Whisper models are close. In AirTypes, Plain runs on your device and uploads no audio; Polished uses a large speech model on AirTypes servers when you need more accuracy on accents and technical vocabulary, want cleanup, or speak a language other than English.

Do I have to pay to finish the seven days?

No. AirTypes Plain is free with no time limit, so the whole plan works without paying anything. Polished is optional prepaid credit for day five and beyond if you want the cleanup.

Get AirTypes, free, and start day one

Plain is free and offline with no time limit, so your seven days run at your pace. Polished is optional prepaid credit when you want your words cleaned up.

Download for macOS and Linux