1. Choose live dictation or saved-file transcription
Use Drafts dictation when you are speaking now and want the result to become a new draft or text at the current cursor. Drafts documents both its custom interface and the standard macOS Dictation hot key.
Use Drafts transcription when the words already exist in an audio or video file. On Mac, Drafts can open that path through drag and drop, File → Import, or Editor → Transcribe. A live microphone session and an imported recording have different source, review, and retention boundaries.
2. Start a new draft or insert into the current one
Choose File → New with Dictation when the spoken text should become a new draft. Choose Editor → Dictate or the Dictate toolbar button when it belongs at the insertion point in the current draft.
Confirm the destination before speaking. Drafts opens ready to edit, so a blank editor does not prove that the correct draft, workspace, tag set, or downstream action has been selected.
- Use New with Dictation for fresh capture.
- Use Editor → Dictate for the current draft.
- Check the cursor and draft before a long session.
- Review the preview before choosing Create or Insert.
3. Treat the preview as temporary until you accept it
Drafts says its custom interface starts listening immediately, displays a preview, and allows pause, resume, and minor corrections. The text remains temporary until Create or Insert completes the session.
That makes the acceptance control a real save boundary. Do not switch apps, close the dialog, or assume an unsaved preview is already in the Drafts database. For important capture, finish a shorter segment and verify the resulting draft before continuing.
4. Set the language and silence timeout deliberately
The custom interface shows the current language or locale and remembers the last selection. Change it before speaking when the next segment uses a different language or regional variant; saved-file transcription does not identify the language automatically.
Drafts can also end a live session after a configured period without text changes. That is useful for hands-free capture, but an unexpected pause can commit the result earlier than intended. Review the active timeout before interviews, thinking pauses, or multi-part notes.
5. Review long sessions and transcription separators
Drafts describes dictation as best suited to shorter snippets. Its guide warns that longer sessions can restart listening, insert an `===` break, and lose a second or two around a breakpoint depending on recognition speed and network conditions.
Saved-media transcription also divides content longer than about one minute into segments because of Apple speech-recognition time limits, inserting the same separator between them. Check every boundary for missing words before running an action, sending text, or treating the draft as a complete transcript.
6. Check permissions and processing without claiming offline
Drafts requests both Microphone and Speech Recognition permission for its custom dictation interface and says the feature uses Apple speech-recognition services. Its transcription guide likewise names Apple speech recognition APIs.
Drafts warns that offline dictation is available only on some newer devices and languages and that a network is often required. Inspect the current Mac, locale, permissions, and Apple disclosure instead of labeling every Drafts session local or offline.
7. Keep capture separate from Drafts actions
Drafts can connect dictation to actions, scripts, keyboard commands, and URL schemes. Those tools can move or transform text after capture, but speech recognition does not validate an action’s destination, template, credentials, or side effects.
Create or insert the text, review it in the editor, and then run the intended action explicitly. That preserves a visible checkpoint between an imperfect transcript and a message, file, task, or automated workflow.
8. Add IraVoice only for a different cross-app requirement
Drafts’ own interface is the better first choice when its preview, locale switching, silence timeout, saved-media transcription, or automation hooks fit the job. IraVoice 0.7.1 instead targets a consistent held-key workflow across supported fields on macOS 26+ Apple-silicon Macs, with speech recognition and supported formatting on the Mac after model setup.
IraVoice does not read the Drafts database, create tags, select workspaces, run actions, invoke URL schemes, import recordings, or transcribe saved media. It supplies text to a supported focused field or falls back to the clipboard; the user still chooses the draft and any downstream action.