For a live spoken exchange instead, use ChatGPT Voice. Dictation is useful when you want to speak the input but inspect the words that will become your request.
Record one request with a clear stopping point
Start with a short task you can check. In this fictional example, a facilities coordinator wants an email draft about an inventory item:
Draft an email asking whether we can reserve eighteen folding chairs for reference K-27. Do not confirm a booking. Ask for collection instructions.
Record the request, finish the capture using the displayed control, and wait for its transcription. The text should be available for editing before you send it. Recording audio and submitting the finished text are separate parts of this flow.
Inspect the words that could change the task
Compare the returned text with what you intended. For the example, three details deserve special attention:
| Detail | Intended wording | Why a mistake matters |
|---|---|---|
| Quantity | eighteen chairs | Eight chairs would produce the wrong request. |
| Reference | K-27 | A different code could identify another item. |
| Instruction | Do not confirm a booking | Losing “not” changes a question into a commitment. |
Edit the prompt directly if one of those details is wrong. For a reference code that must be exact, typing it is often simpler than repeatedly recording the whole request.
Then send the corrected text. Read the resulting email to confirm that it asks about availability and collection rather than announcing a reservation. Correct input helps define the task; it does not remove the need to review the answer.
Understand what happens to the recording
Dictation sends the audio to OpenAI for transcription. Its audio is retained while the chat remains in your history. Deleting the chat generally deletes the associated clip within 30 days, subject to security or legal exceptions and previously shared, disassociated training audio. Dictation retention
Consumer audio may be used for training if you chose to share it. In Settings > Data controls, disabling Include your audio recordings, or turning off Improve the model for everyone, stops future audio sharing across dictation and voice chat. A training choice and deletion are separate controls. Audio-sharing settings
Choose permitted material before recording, especially in a shared room. Finish with the corrected prompt and a checked answer, rather than relying on how clearly you remember speaking.