How to use voice input in Cursor: shortcuts and fixes

Cursor's voice input is dictation: you click the microphone icon in the chat input, speak, and your words appear as text that you review and send. In the IDE you can also press Cmd Shift Space, and in the Agents Window you hold Ctrl M while you talk, according to Cursor's prompting docs. Everything below comes from Cursor's own docs, as of November 6, 2026.

How to use voice input in Cursor

Dictation works in the chat input of an agent. Pick the way that matches where you are working.

1. Open an agent chat in the IDE, the Agents Window or Cursor for iOS.
2. Start dictation. Click the microphone button in the chat input, or use the shortcut for your surface (next section).
3. Allow microphone access if your computer, browser or phone asks.
4. Speak your prompt. Cursor's docs suggest speaking naturally and including technical details such as file and function names.
5. Read the transcription, fix anything that was misheard, and press send yourself.

The button has a different label on each surface, which matters if you are hunting for it in a tooltip:

The labels and shortcuts are from Cursor's voice chat docs, which compare the two features in one table. The docs write the IDE shortcut with the Mac key names. They don't list a Windows or Linux equivalent, so check your keyboard shortcut settings if Cmd is not a key you have.

The shortcuts

In the IDE, Cmd Shift Space toggles dictation: press it to start, press it again to stop. The docs use the word "toggles" for it.

In the Agents Window it works like a walkie-talkie. You hold Ctrl M for as long as you are talking, and let go when you are done.

Cursor lists no shortcut for voice chat, the other voice feature. That one only starts from a button.

Voice input is not voice chat

Cursor has two voice features, and the names get mixed up. Cursor says dictation used to be called Voice Mode, which adds to the confusion.

Voice chat is the realtime call that Cursor added on November 6. It needs a signed-in account, has a daily limit and is a separate thing from what this guide covers. We cover it in Cursor adds voice chat to its agents. If you only want to speak a prompt instead of typing it, you want dictation.

Voice input not working: fixes from the docs

Cursor's docs have no troubleshooting page for dictation. These are the problems the docs do describe, and what they say to do.

The microphone is blocked

If you denied microphone access earlier, Cursor shows "Microphone blocked. Allow it in your system or browser settings." On iOS the message is "Microphone blocked. Allow it in Settings." Turn microphone access back on for Cursor there, then try again. On a Mac, that means the operating system's privacy settings for your microphone; in a browser, the site settings for cursor.com.

The microphone button is missing in the IDE

The docs tell you in the voice chat FAQ that the IDE hides the voice chat button when dictation is turned off. That implies a setting that turns dictation off, but the pages we read don't say where it is. If the button is missing, and none of the other cases apply, Cursor's advice is to update the app.

The shortcut does nothing

Check which surface you are in. Cmd Shift Space is documented for the IDE, and Ctrl M, held down, for the Agents Window. Neither is documented the other way around. Cursor's docs don't say whether you can rebind them, so look for them in your keyboard shortcut settings before assuming a conflict with another app.

The button you see starts a call, not dictation

The waveform button is voice chat. The microphone button next to it starts dictation. If Cursor starts talking back, you pressed the waveform.

What the docs don't say

Several things people search for are not answered in Cursor's documentation: which languages dictation understands, whether it works in the terminal or in inline edit, whether the audio is processed on your computer or on a server, and what the setting for turning dictation off is called. Anything you read elsewhere on those points is not from Cursor. Cursor's docs only say that Privacy Mode gives zero data retention for voice chat, and they say nothing about dictation either way.

Write your prompts the way the docs suggest, with file and function names spoken out, and read the text before you send it. Dictation puts words in the box and nothing more, so a mistake in the transcript becomes a mistake in the prompt.