# Dictation
> Dictate instead of typing: the mic in the chat composer turns your speech into the prompt live. Where to find it, how it works and what to do if it fails.
Source: https://endless.app/docs/transcripts/dictation
Transcripts turn a recording into a document. **Dictation** goes the other way: you talk, and
the words appear in the composer as the prompt you were about to type. The product puts it in a
single line — *"Talk instead of type — Select the mic in the composer and dictate your prompt."*
Nothing is saved as a transcript, nothing lands in your assets. It is a way of writing.
## Where the mic is
When a chat composer is **empty**, the send button is a microphone, labelled **"Start
dictation"**. Type anything and it turns back into the send button — so the mic is always
exactly where your thumb already is, and never in the way of sending.
The big composer you land on, and any chat you open.
The composer on the **Social radar** page has it too — it's the same
composer with a recency control attached.
**The generate composers don't have it.** On [Image](/docs/create/image),
[Video](/docs/create/video) and the generate bar on a [canvas](/docs/create/canvas), the
button is a run button rather than a mic, and it stays a run button whether or
not you've written a prompt.
## Dictating
### Select the mic
The first time, your browser asks for permission to use the microphone. The composer switches
to a recording bar straight away: a cancel **✕**, a pulsing red dot, and a running timer.
### Talk
Your words appear in the composer as you speak. The sentence you're still saying shows as
provisional text and firms up as you pause; each finished chunk is added to the message for
real, announced as *"Dictation added to message"*. Anything already in the composer is kept —
dictation adds to your draft rather than replacing it.
### Stop, or cancel
The send button becomes a stop button while you're recording — select it, or press **Enter**, to
finish. Endless catches up on the last thing you said and then hands the composer back to you.
To throw the recording away instead, select the **✕**, or press **Escape**. Nothing is added.
### Read it before you send it
Dictation writes the prompt, it doesn't send it. Pressing Enter while recording stops the
recording — it doesn't send the message — so you always get a look at the text first. Edit it
like anything you typed, then send.
## Worth knowing
A single dictation runs for at most five minutes and then finishes itself,
keeping what it heard. For anything longer, dictate again — or record it and
make a [transcript](/docs/transcripts) instead.
While recording, **Enter** confirms and **Escape** cancels. Both keep their
normal meanings the moment the recording ends.
Stopping keeps the text; cancelling drops it, including the part you could
already see. There's no undo for a cancelled dictation.
Dictation isn't billed by the minute the way a transcription is, and it
doesn't need any particular plan.
## If it doesn't start
The composer says what went wrong rather than failing quietly.
| What you'll see | What to do |
| -------------------------------------------------------------------------------------- | ------------------------------------------------------------------------------------- |
| *"Microphone access is blocked — allow it in your browser settings"* | Permission was denied for this site. Re-allow the microphone and select the mic again |
| *"No microphone found — connect one and try again"* | No input device is available to the browser |
| *"Dictation isn't supported in this browser — type your message instead"* | The browser can't do live audio. Use a current desktop or mobile browser |
| *"Couldn't start dictation — try again"* or *"Dictation failed — try again"* | A transient problem starting or holding the session. Select the mic again |
Dictation needs a working microphone and a live connection — it transcribes as
you speak rather than recording a file and sending it. If the connection drops
mid-sentence, what had already been added to the composer stays.