1. Choose Dictation, Voice, Record, or Mac text input
Use ChatGPT Dictation when you want to speak a single prompt, inspect the transcription, edit it, and decide when to send it. OpenAI’s current help page identifies this as the microphone icon and explicitly distinguishes it from Voice.
Use ChatGPT Voice when you want a live, free-form conversation in which ChatGPT speaks back. Voice transcripts are conversation records rather than guaranteed verbatim transcripts, so Voice is not the right control when exact prompt wording is the main requirement.
Use ChatGPT Record for a meeting, brainstorm, or voice note that should become a transcript and generated notes. Use macOS Dictation or a dedicated local dictation app when you want speech converted to text before ChatGPT receives anything.
2. Dictate one editable ChatGPT prompt
Open ChatGPT on the Mac, choose the intended conversation, and select the microphone icon in the message composer. Allow microphone access if macOS or the browser asks, speak the prompt, then stop the recording.
ChatGPT returns the transcription as text in the composer. Read it before sending. Correct names, dates, amounts, links, quoted text, negations, and instructions that could change what ChatGPT does. Add files or context only after confirming that the transcript says what you intended.
This review step is the defining advantage of Dictation over a live Voice exchange. The final user message is ordinary text, so you can restructure it, remove accidental disclosures, or abandon it without sending.
- Select the microphone icon in the ChatGPT message composer.
- Speak one message and stop the recording.
- Review and edit the returned transcription.
- Select Send only when the text and destination are correct.
3. Understand Dictation audio retention and training controls
OpenAI says Dictation audio is sent to its models for transcription. The audio is retained for as long as the chat remains in chat history. When you delete the chat, OpenAI says the associated audio clip is deleted within 30 days, subject to security or legal needs and the documented exception for clips already shared and disassociated for training.
For consumer users, OpenAI may train on Dictation audio only when the user has chosen to share audio to improve models. The current control is under ChatGPT Settings → Data Controls: disable Include your audio recordings or turn off Improve the model for everyone. OpenAI says business inputs and outputs are not used for training by default.
These are the Dictation rules. Do not substitute the retention statement for Voice clips or Record audio, because OpenAI documents those workflows separately.
4. Use Voice for a conversation, not exact prompt capture
Select the Voice control when you want to speak with ChatGPT and hear a spoken response. OpenAI describes Voice as a natural, free-form conversation; depending on plan, region, workspace, and app version, the available experience may be Live, Advanced, or Standard.
OpenAI advises using Voice for back-and-forth discussion and Dictation when you want to record a prompt, review its transcription, and send it as text. It also warns that a Voice transcript may not exactly match the spoken conversation.
Voice has its own audio, video, transcript, and model-improvement controls. Read the current Voice documentation for the active mode and workspace instead of assuming that one Dictation retention rule covers every ChatGPT microphone feature.
5. Use Record only when preserved context is the job
ChatGPT Record is a separate workflow in eligible ChatGPT macOS app accounts. It can transcribe and summarize meetings, brainstorms, and voice notes, distinguish speakers, and save generated notes with the conversation or recording.
OpenAI says Record audio files are deleted after transcription, while the resulting transcript and canvas follow the workspace’s conversation-retention settings. Recording other people requires the appropriate consent; OpenAI presents that reminder directly in the workflow.
Record is not a longer Dictation button. It is appropriate when you need a meeting artifact, speaker context, or generated notes. Dictation is appropriate when you need one editable prompt.
6. Fix ChatGPT dictation on a Mac
If the microphone control cannot hear you, open System Settings → Privacy & Security → Microphone and allow the ChatGPT app or the browser you are using. Confirm the intended input under Sound → Input, then close another app or tab that may be holding the microphone.
If the control is missing, update or reload ChatGPT, sign in again, and compare the microphone Dictation icon with the separate Voice control. Account, workspace, region, and app-version differences can affect Voice and Record availability; do not assume every missing microphone feature is Dictation.
If ChatGPT detects the wrong language, review the preferred language in ChatGPT’s Voice or Speech settings and retry in a quiet room. Test macOS Dictation in TextEdit to separate a Mac microphone problem from a ChatGPT account, browser, permission, or network problem.
7. Add IraVoice only for local cross-app text
Use ChatGPT Dictation when sending recorded audio to OpenAI for an editable transcript fits the task. Use Voice for a conversation and Record for meeting-style capture. Those are the native answers.
IraVoice 0.7.1 is relevant on macOS 26+ Apple-silicon Macs when speech recognition should run on the Mac or one held-key text workflow should extend across supported focused fields. IraVoice does not provide a spoken ChatGPT response, record a meeting, preserve ChatGPT audio, or change OpenAI’s data handling after you send the final text.