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 keep dictating for years. 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. Speaking runs around 130 words per minute against roughly 40 for typing. That difference is waiting on the other side of a week of small steps.
Day 1 — Dictate one message. That is all.
Install AirTypes, sign in, and set a hotkey you can reach without looking. 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.
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 the hotkey, speak, release. The text lands at your cursor in whatever app has focus. There is no window to switch to, no transcript to copy, no paste step. Once your hands know that rhythm, everything else in this plan is just repetition.
Two setup details that pay for themselves immediately: pick a hotkey that does not collide with your editor or browser shortcuts, and 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.
Day 2 — Move every reply to voice
Today, one rule: every chat message, comment, and short reply gets dictated instead of typed. Slack, Discord, email replies of two lines, code review comments, Jira 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 teaches 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.
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.
AirTypes ships six on-device accuracy tiers, from the fastest small models up to the largest Whisper tiers. The trade-off is straightforward: bigger models are more accurate on accents, domain vocabulary, and noisy rooms, and they take a little longer to process each utterance. Everything runs locally either way, so the only cost of a higher tier is time and CPU, never privacy.
Spend ten minutes today dictating the same three sentences of your real vocabulary at two different tiers. Whichever one gets your jargon right without feeling sluggish is your tier. Set it and stop thinking about it.
If you want the deeper explanation of what changes between tiers, see our guide to offline speech recognition and how on-device engines work.
Day 5 — Let AI clean it up
Raw dictation is faithful to what you said, including the "um", the false start, and the sentence you abandoned halfway. Cleanup fixes that — and this is where dictation stops being transcription and starts being writing.
With My Agent, your speech can be routed through your own AI provider (OpenAI, Anthropic, or a local Ollama model with your own key) before the text is injected. You speak casually; a polished, correctly punctuated version lands at your cursor.
Set up two profiles today and no more:
- Clean-up — fixes filler words, punctuation, and capitalisation, keeps your wording. This becomes your default.
- Professional email — turns a spoken brain-dump into a structured, polite message. Use it for anything a client or manager reads.
Two profiles you actually use beat eight you have to think about. You can add more once these two are automatic. The full pattern is covered in the voice-to-AI-prompt guide.
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 cognitive task than producing something polished from an empty page, and dictation is the fastest way to get to "rough".
People consistently report the same thing at this step: the draft takes a fraction of the time 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.
Day 7 — Read your stats, lock the habit
Open the AirTypes dashboard and look at the week: words dictated, transcriptions, estimated time saved. 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 "you dictated 6,000 words this week".
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 and code comments. For researchers it is notes. One workflow, permanently, is what a habit is made of.
If you are on the 7-day free trial, this is also the day it ends. The decision is small — AirTypes is $3.99 a month — but make it against your own numbers, not a feeling.
The four mistakes that stall people
- Correcting mid-sentence. The biggest one. Speak the whole thought, then edit once. Mid-sentence correction is what makes dictation feel slower than typing.
- 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.
- 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.
- Building eight AI profiles on day one. Configuration is fun and feels productive. It is not the habit. Two profiles, then use them for a week.
What happens after week one
The pattern we see in long-term users is that usage increases between week two and week eight, which is unusual — most utility apps drop off sharply once novelty fades. Dictation goes the other way because each new context you try it in (terminal, design tool, notes app, AI chat) works the same way, so the habit generalises instead of staying local to one app.
Around week three, 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" — dictating prompts to ChatGPT, Claude, and Gemini, drafting in a second language, working with your hands off the keyboard when your wrists need a break.
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?
Three to five days of daily use for most people. The awkwardness of day one is about learning to speak in complete thoughts, not about the software.
Do I need to speak slowly?
No. Whisper-based models are trained on natural conversational speech, so a normal pace transcribes better than slow, over-enunciated speech. Full sentences help; slow words do not.
What should I dictate first?
Short, low-stakes text — a chat reply, a commit message, a comment. Save documents for day six.
Is offline dictation as accurate as cloud dictation?
For clear speech on a decent microphone, on-device Whisper tiers are comparable, and the larger local tiers handle accents and technical vocabulary well. The trade-off is processing time, not accuracy — and your audio never leaves your machine.