Exactly as spoken: Capture Your Wording Before Choosing a Writing Style

Sometimes the useful output is your own wording: a reminder, a rough paragraph, or an idea you want to keep developing yourself. Exactly as spoken gives you a direct transcription workflow without a separate AI formatting pass. It is also a helpful starting point when you are trying to understand why a draft differs from what you said.

The distinction matters because speech recognition and writing style are different stages. Recognition turns audio into text. A formatting model can then reshape that text according to a style. Exactly as spoken bypasses that second stage; cloud speech processing still requires a connection and available credits.

Select the direct transcription workflow

• Open Styles and select Exactly as spoken as the active style.
• Open its Rewrite and check Speech transcription for an available speech model.
• Confirm Text formatting shows No formatting. This style does not use a text formatting model.
• Use Today's Start listening to record a short thought with the selected style, then stop to process it.
• Read the result before using it. Recognition can still introduce incorrect words or punctuation.

For dictation into another app with the global shortcut, set Styles → Default style to Exactly as spoken for unmatched destinations, or link that destination to Exactly as spoken under Linked apps or Linked websites. A linked match takes priority over the default. Selecting a style card alone does not determine the style used by global recording.

An input such as “I might move the launch to Friday, depending on whether the review is finished” should retain the uncertainty and dependency. Review “might” and “depending on” as carefully as the date. A transcript that loses either can change a tentative plan into a commitment.

Understand the optional text changes

No formatting does not mean a character-for-character record of the audio. Speech recognition itself produces written text, including its interpretation of punctuation. Words replacements may also affect the result.

Voice commands can be enabled for Exactly as spoken, and the app-wide filler setting can remove supported hesitation tokens. These options are separate from AI formatting and default off for fresh settings. Save the settings that fit your writing workflow.

If you want to evaluate recognition alone, keep optional commands and filler removal off for that test. That makes it easier to spot whether an unexpected change came from recognition or an enabled transformation.

Know when to add formatting

Direct transcription is a good fit when you want to edit the structure yourself, preserve a rough draft, or capture text for another writing tool. A formatted style is useful when you repeatedly need a defined shape, such as short paragraphs or a structured brief.

If the result is wrong, first check the microphone, Speech language and difficult vocabulary. Changing a formatting instruction will not recover an inaudible name. If the words are right but the layout takes repeated manual work, try custom formatting rules.

See the Exactly as spoken instructions and the speech and formatting model guide for the two-stage workflow.