Question · Voice and AI
How do you dictate with AI?
To dictate with AI, put your cursor where the text should go, start dictation in an app with an AI editing step, and speak. The app turns your voice into a raw transcript, and then a language model can remove filler words and add punctuation before the text lands on the page.
Updated October 17, 20264 min read8 cited facts, each traced to its source
To dictate with AI, you use the same steps as any dictation: place the cursor, start dictation and speak. What is different is the second stage. Many AI dictation apps send the raw transcript through a language model that removes filler words, adds punctuation and applies your spoken corrections. Before you start, check three settings: whether the AI edit is on, where your audio is processed, and whether the app reads text on your screen. The guide to AI dictation explains each stage, and the overview of speech to text AI places it among related tools.
Dictate with AI in five steps, after checking three settings
These steps work for most AI dictation apps. The settings come first, because they decide what the app does with your words.
- Choose the editing mode. Some apps let you turn the AI edit off. Superwhisper's Voice to Text mode, for example, outputs raw transcribed text with no language model involved 1. Pick the mode that matches the text: raw for records, edited for drafts.
- Check where the audio goes. Cloud apps send your audio to a server before text comes back. Wispr Flow's privacy page says dictated audio goes to its servers over an encrypted connection, TLS 1.2 or later 2. Local options can keep the audio on the device. Superwhisper's Free plan uses local Whisper models without AI processing, and the company says nothing dictated leaves the device 3.
- Check the screen setting. Some apps read text from the app you are typing into to get names and jargon right. Wispr Flow's privacy page says that with context awareness off, the app transcribes without reading anything on your screen 4. Decide whether that trade is worth it before you turn it on.
- Place the cursor and start. Put the insertion point in the text field where the words should go, then start dictation and speak.
- Speak in whole sentences, then proofread. Read the output before you send it, and check names, numbers and any sentence the AI rewrote.
The AI edit removes filler words, but it can also change what you meant
The main gain is the cleanup. Wispr Flow's home page says the app automatically removes filler words, adds punctuation and formats your writing as you dictate 5. Microsoft describes similar editing in Fluid dictation on Copilot+ PCs, which it says corrects grammar, punctuation and filler words as you speak 6.
The cost is the risk of over-correction. In a 2025 INTERSPEECH paper, Yamashita and colleagues say that language-model correction methods which rely mainly on the text, and ignore phonetic cues, lead to over-correction: the model changes words that were already right 7. Read the edited text against what you said, especially when the exact words matter.
It depends on the text, the privacy you need and the language
Three cases change the answer.
- Verbatim text. For a record, a transcript or a quote, turn the AI edit off, because an edit can change words that were already right 7.
- Sensitive text. If the content should not leave your device, choose a local option and confirm it in the app's documentation before you rely on it. A cloud app's privacy page is the place to check the audio route.
- Languages other than English. On iPhone, Apple says that with Apple Intelligence in iOS 27, dictation uses an on-device model to improve spelling, punctuation and capitalization, and that this is available in English on named iPhone models 8. Test your language before you count on the AI pass.
| Setup | Where the AI edit runs | Where audio goes |
|---|---|---|
| Superwhisper Free | No AI edit: local Whisper models only 3 | Stays on the device |
| Wispr Flow | Cloud processing, with the AI edit built in 5 | Sent to the company's servers 2 |
| iOS 27 dictation with Apple Intelligence | On-device model, English on named models 8 | Not stated in Apple's note |
The table shows how the options differ. It is not a ranking, because the right setup depends on your text and your privacy needs.
What to do: start with plain dictation, then add the AI edit where it helps
If you dictate short messages or notes, the built-in tools on your computer or phone are often enough, and you do not need a separate app. Add an AI dictation app when you want the cleanup, or when you need the app to fix names and jargon that the built-in tool gets wrong.
When you do add one, keep the AI edit on for drafts, turn it off for records, and use a local option for anything sensitive. Yap, which we make, is an on-device dictation app for Mac and Windows, so it falls in the local group of that comparison. You can check the current options on our download page.
For the other questions people ask, see how to dictate a prompt, why AI dictation is slower, whether AI dictation changes your words, whether AI dictation reads your screen and whether AI dictation can run locally. For the terms used here, see filler words, large language model, on-device speech recognition and cloud dictation.
Evidence
Every number on this page, with where it comes from. Open a fact to see the exact words in its source.
- 1.
- 2.
- 3.
- 4.
- 5.
- 6.
- 7.In the 2025 INTERSPEECH paper, existing language-model error correction for speech recognition relies mainly on the text, ignores phonetic cues, and that leads to over-correction, meaning the model changes words that were already right.
Natsuo Yamashita et al., arXiv (accepted by INTERSPEECH 2025), 2025 · Original
- 8.
Go deeper
Related questions
- Can AI dictation run entirely on your own computer?
- Does AI dictation read your screen for context?
- Is AI dictation more accurate than built-in dictation?
- Which AI models do dictation apps use?
- Why is AI dictation slower than built-in dictation?
- Are AI summaries of voice recordings accurate?
- Are smart speakers still popular?
- Can AI detectors flag dictated text?