1. Decide which writer workflow you actually need
Live dictation listens now and puts editable text at the insertion point. Saved-audio transcription starts with an existing recording and creates a transcript later. Voice editing changes existing text. Hands-free control operates the interface. A writer may use several of these, but one product claim should not be transferred to every job.
For a book, article, newsletter, or script, identify the system of record first: a Word file, Google document, Scrivener project, Ulysses library, WriterDuet screenplay, or another editor. Then choose a speech path that works in that destination without weakening revision history, structure, collaboration, or export.
- Draft now: live speech becomes editable manuscript text.
- Capture away from the desk: record first, transcribe later.
- Revise existing prose: use explicit editing or rewriting controls.
- Navigate without a keyboard: evaluate Voice Control separately.
2. Test Apple Dictation before buying writer software
Apple says Dictation can enter text anywhere you can type. On current Apple-silicon Macs, the keyboard can remain active while you speak, supported languages can add automatic punctuation, and Dictation has no length timeout, although it stops after 30 seconds without detected speech.
Turn it on under System Settings → Keyboard, inspect the processing disclosure for the active language and configuration, choose a shortcut, and dictate one real passage in the intended editor. Correct names, dialogue, punctuation, paragraph breaks, and homophones before deciding that another tool is necessary.
3. Consider Google Eloquent for the free local Mac baseline
Google’s June 3, 2026 Mac launch describes Google AI Edge Eloquent as a fully offline, 100% on-device desktop app with a customizable hotkey, cross-app dictation, Voice Edit, and local audio or video file transcription. That combines jobs Apple Dictation keeps separate.
Eloquent is therefore the stronger free baseline when a writer needs local file transcription or voice-driven rewriting in addition to live dictation. Its current reviewed product material is English-focused, and a generated or rewritten passage still needs factual, tonal, and manuscript-level review.
4. Use Word or Google Docs when native document commands matter
Microsoft documents Word Dictate for Mac as a Microsoft 365 feature that requires sign-in, a microphone, and a reliable internet connection. It includes spoken punctuation, selection, deletion, navigation, and comment commands. Microsoft says speech utterances are sent to Microsoft for transcription and that the Dictate service does not store the audio data or transcribed text.
Google Docs Voice typing works in current Chrome, Edge, and Safari. Google says the browser controls the speech-to-text service and sends resulting text to Docs. English voice commands can select, format, and navigate document text. Choose either native path when its command set and collaboration model are worth the connected-processing boundary.
5. Treat Scrivener as the manuscript system, not the recognizer
Literature & Latte teaches Mac writers to use the system Dictation path in Scrivener. Speech enters the focused editor document; Scrivener still owns Binder structure, project organization, Composition Mode, document history, and the later Compile operation.
For long projects, create a titled Snapshot before a heavy correction or revision pass. Scrivener says a Snapshot preserves an individual document version and can be compared or restored, but it is not a replacement for a full project backup. Dictation speed is not useful if a chapter cannot be recovered or exported safely.
6. Use Ulysses when sheets, Markdown XL, and export fit the book
Ulysses documents Apple’s system Dictation as its Mac speech-input path. The spoken text enters the editor, while Ulysses keeps the material in sheets and uses Markdown XL by default.
That separation is useful for writers who want speech input without changing the manuscript model. Keep headings, emphasis, links, notes, keywords, goals, library location, and final export as explicit Ulysses operations instead of asking speech recognition to infer the entire publishing structure.
7. Choose the screenplay’s native command layer
WriterDuet’s own Dictate feature runs in a computer web browser, uses the browser’s dictation, and recognizes screenplay line types. Commands such as “Press enter character” and “Press enter action” can move to the next line with the selected screenplay structure.
WriterDuet explicitly says this native Dictate feature is not available in its desktop app, mobile app, or mobile and tablet browsers. A Mac-wide input path can still enter text into a focused desktop field, but it does not inherit WriterDuet’s browser-specific line-type commands.
Final Draft 13 takes a different path on Mac: it uses macOS Dictation and adds an optional English Speech Commands layer for screenplay elements, keystrokes, punctuation, undo, and redo. Choose the screenplay system first, then use the command layer it actually documents rather than transferring commands between products.
8. Do not present Dragon Professional as current Mac software
Nuance positions Dragon Professional v16 for Windows 11 with backward compatibility to Windows 10. It supports live dictation, transcription of existing audio, custom words, commands, and auto-texts.
Those capabilities can make Dragon the better choice for a writer who works on supported Windows hardware and needs a trained vocabulary or deeper command system. They do not make the current product a Mac app. A Dragon-for-Mac search should be answered with that platform boundary before recommending an alternative.
9. Use a two-pass workflow for a book-length draft
Before speaking, open the correct chapter or scene and write a one-sentence intent. Dictate a bounded passage without trying to polish every sentence in real time. Stop, confirm the text landed in the intended document, then fix recognition errors while the intended words are still fresh.
Make structural and stylistic edits in a separate pass. Preserve a Snapshot, revision, version, or backup before large changes, and export a small proof before trusting a full-manuscript pipeline. This keeps speech capture, transcription correction, prose revision, and publication as four reviewable stages.
- Name the scene, argument, or beat before recording.
- Dictate a bounded passage into the correct destination.
- Correct recognition errors before rewriting the prose.
- Version the draft, then revise and export separately.
10. Test dialogue, names, punctuation, and long pauses
Use a sample that resembles the real manuscript: character names, invented places, quotations, contractions, em dashes, numbers, paragraph changes, and one deliberate pause. Generic demo sentences hide the correction work that matters to an author.
Do not compare vendor accuracy percentages measured with different speakers, microphones, rooms, languages, and editing rules. Measure the visible result from the same passage on the same Mac, then count corrections and verify that the destination preserved the intended structure.
11. Separate speech privacy from manuscript storage
Local speech recognition can keep microphone processing on the Mac. It does not make a Word document, Google document, Scrivener project, Ulysses library, WriterDuet project, cloud backup, shared folder, or exported manuscript local.
For an unpublished work, map the complete path: microphone audio, speech model, optional cleanup or rewrite, text insertion, autosave, sync, revision history, collaborators, backups, and export. A privacy claim is useful only when it names the stage it covers.
12. Choose IraVoice only for its narrow writer fit
IraVoice 1.2.2 is a separate option on macOS 13+ on Apple silicon, Windows 10 or 11 (64-bit), and Linux (beta) when held-key dictation with Right Command or Right Ctrl, on-device NVIDIA Parakeet recognition, a 3-day free trial with no account, and a $79 one-time IraVoice Local license match the writing habit. The default model is English; other offline models cover 25+ languages.
IraVoice can learn custom words and expand replacements and snippets, but it does not record interviews, transcribe saved audio, maintain a manuscript library, manage Scrivener or Ulysses structure, create WriterDuet line types, revise a book autonomously, or export and publish it. It types editable text at the cursor in the focused app; the writer and destination app own the manuscript.