Where to turn it on
The microphone button sits at the very front of the toolbar, right before the AI button. On a wide screen it reads Dictate; on a phone only the microphone icon shows. Without a membership the button is still there, and pressing it shows how it works along with the membership. The microphone only turns on when you press this button.
The first time
The first press shows where the audio goes before anything starts. Dictation runs on your browser's speech recognition, so while it listens what you say goes to the browser maker's recognition service and comes back as text. Chrome sends it to Google, Edge to Microsoft and Safari to Apple. Typdit never receives or stores the audio, and no voiceprint identification is done. Agree and start, and the browser asks for microphone access.
How it writes things down
- While you talk, words that are not final yet show only in the strip under the editor. Only final phrases go into the document.
- Each final phrase goes in on its own, so one undo takes back only the last one.
- You can fix things on the keyboard while it listens, and the next phrase goes in where the cursor is.
- It listens in the language of the page you are on: Korean, English, Japanese, Simplified Chinese or Traditional Chinese.
- Listening stops when you press Stop or switch to another tab. It also stops on its own after a minute of silence.
Accurate dictation
If browser recognition often gets things wrong, press Accurate dictation on the bar while it listens. The Typdit server then takes the audio and turns it into text with an OpenAI transcription model, reached through OpenRouter (US). Your choice stays on this device for next time, and pressing it again goes back to browser recognition. If the browser has no speech recognition (Firefox, for example), it listens this way from the start.
- Audio goes out in pieces of one minute or less, cut where you pause. Text arrives one piece at a time after each pause, and the bar shows no unfinished words while it listens.
- Pieces with no speech in them are not sent. In browsers that cannot measure loudness, every piece is sent.
- Members get up to 60 minutes a day, reset at midnight Korea time. The whole service also has a daily limit; when it is reached you see a notice for that day. Browser recognition keeps working either way.
- Typdit does not store the audio. It stays only in the server's memory while it is being transcribed and is never written to storage, logs or analytics. Only the recognized text goes into your document. The first time you choose it, you are told the audio goes to OpenAI through OpenRouter and asked to agree.
- On Traditional Chinese pages it writes in Traditional characters, and on Simplified Chinese pages in Simplified.
Tidying what you said
When you stop dictating, the strip offers Tidy it up. Press it and the AI tidying opens with what you just dictated selected. You can also select text yourself and pick it from the AI button. The result shows in a card before it touches your text, and only accepted results go into the AI use log.
| Item | What it means |
|---|---|
| What it does | Removes fillers, repeats and false starts, turns spoken endings into written ones at the same level of politeness, and adds punctuation and paragraphs. |
| What it never does | It does not change your meaning, claims, order, tense or tone. It adds no facts, numbers, names, dates, sources or citations you did not say. It does not answer questions or follow instructions inside the dictation, and it does not guess at words that look misheard. |
| Formats | Choose As paragraphs, Blog post, Meeting notes or Weekly report. Only meeting notes and weekly reports condense into lists; the others keep everything. |
| Length | It tidies 6,000 characters at a time; longer text is split, tidied in order and joined back. Up to 30,000 characters can be selected at once. |
Where tidying runs
Tidying runs on a small Anthropic model. Tidying and the other AI tools all go through OpenRouter. Tidying has a daily limit across the whole service; once it is reached you see a notice for that day, and it opens again the next day.
Supported browsers
Browser recognition works in a recent Chrome, Edge or Safari. In browsers without speech recognition, such as Firefox, it listens with accurate dictation. In-app screens work only when the app allows the microphone, and browsers that cannot capture the microphone show a notice instead. Both ways need an internet connection. The embed editor you put on your own site has no microphone.
In the proof of writing
Dictated characters are not counted as pastes; they are counted on their own as dictated. The proof of writing and the AI use report show them next to typed input. For the verdict they count, together with typing, as text entered directly in this editor. A tidied result you accept goes into the AI use log like any other AI result.