ChatGPT Speech to Text: Dictate Prompts Instead of Typing

ChatGPT speech to text comes in two forms. Dictation turns a recording into text you can edit before sending, and Voice is a spoken conversation where ChatGPT talks back. Both live inside ChatGPT. If you also want to talk instead of typing in Claude, Gemini, your email and Slack, a dictation app like LumeVoice gives you one shortcut for all of them.

ChatGPT's dictation and Voice, explained#

OpenAI's own rule of thumb: use Voice for a live back-and-forth, and use dictation when you want to record a prompt, check the text and then send it (ChatGPT Voice help).

Dictation: ChatGPT's built-in voice typing#

1. Click or tap the microphone icon in the message box. In the ChatGPT desktop app, select Dictate in the composer.
2. Say your prompt.
3. Stop recording. The transcription appears in the box, where you can edit it before you send.

In June 2026, OpenAI began rolling out a new speech-to-text model for dictation on all plans. Two things to know: it only works inside ChatGPT, and OpenAI keeps the dictation audio for as long as the chat stays in your history (Voice Dictation FAQ).

Voice: a spoken conversation#

On the web, select the Voice icon in the prompt window; in the mobile apps, it's in the message bar. ChatGPT listens, answers out loud, and adds a transcript to the chat when you're done. In the desktop app, Voice also works across Chat, Work and Codex for Plus, Pro, Business, Edu and Enterprise users, depending on rollout and workspace settings.

Voice has usage limits that depend on your plan. As of October 2026, Free accounts get limited access and Plus includes 3 hours of Voice in Chat per rolling 24 hours. OpenAI also notes that Voice transcripts aren't verbatim. Voice is for thinking out loud, not for producing text you'll reuse.

Why talk to ChatGPT instead of typing?#

Speed is the obvious reason. In a Stanford, University of Washington and Baidu study, speech was about three times faster than typing on a phone keyboard.

The better reason is context. OpenAI's prompting advice starts with being clear and specific and giving the model enough context. Typed prompts tend to be short, because every extra sentence feels like work. Spoken prompts tend to include the background you'd give a colleague, and that's what gets a useful answer the first time.

Here's a prompt as you might say it:

What you say: "Um, I need a short, polite email to my landlord. The heating has been broken for, uh, two weeks, actually three weeks. I've called twice and, you know, nobody came. I want a repair date this week, in writing. Keep it under 150 words and don't sound angry."

And what LumeVoice types into ChatGPT:

What lands in the box: "I need a short, polite email to my landlord. The heating has been broken for three weeks. I've called twice and nobody came. I want a repair date this week, in writing. Keep it under 150 words and don't sound angry."

Few people would type all of that. Saying it takes under half a minute.

How to dictate ChatGPT prompts with LumeVoice#

LumeVoice types into whatever text field has your cursor, so it works on chatgpt.com in any browser and in the ChatGPT desktop app on Mac and Windows.

1. Click into ChatGPT's message box.
2. Hold fn (Globe) on a Mac or Ctrl + Win on Windows, and talk through what you need.
3. Let go. A cleaned-up prompt appears in the box: filler words gone, self-corrections applied, spoken lists formatted as lists.
4. Read it, then press Enter. LumeVoice never sends anything and never reads ChatGPT's answer aloud. It only writes.

For a long prompt, double-press the key for hands-free mode, talk for as long as you need, and press once to finish. On your phone, use the Lume keyboard in the ChatGPT app on iPhone, or the floating bubble over your keyboard on Android.

How to dictate a prompt ChatGPT can act on#

A good spoken prompt has the same parts as a good typed one. Saying them in this order helps:

1. Start with the outcome. "I need a one-page plan for..." tells ChatGPT what done looks like.
2. Give the background. Who it's for, what happened, what you've already tried. This is where talking beats typing.
3. List your constraints out loud. "Three rules: one, under 200 words; two, no jargon; three, end with a question." LumeVoice turns that into a numbered list ChatGPT can follow.
4. Name the tone and the format. OpenAI suggests describing tone with plain words like formal, friendly or serious. Then say the format: a table, five bullet points, a draft email.
5. Fix mistakes by talking. "Make it 300 words, actually 200" leaves only the 200 in your prompt.
6. Read before you send. Check names and numbers, then press Enter. If the answer misses, dictate a follow-up rather than starting over.

Your custom dictionary helps here too: add product names, people and jargon once, and they come out spelled right in every prompt you dictate.

ChatGPT dictation vs Voice vs LumeVoice#

As of October 2026:

When ChatGPT's own dictation is enough#

If ChatGPT is the only place you want to talk instead of type, its built-in dictation may be all you need. It's free, it's already there, and a stray "um" rarely matters in a prompt. If you want to think out loud and hear answers, Voice does something LumeVoice doesn't: it talks back.

LumeVoice makes sense when the habit spreads beyond one app:

• You use more than one AI tool. The same key works in Claude, Gemini, Perplexity and Microsoft Copilot.
• The text is for people, not just a model. Emails, Slack replies and docs need the filler and false starts gone, and that's where cleanup matters most.
• You want your words, not ChatGPT's. Often the quickest route is to dictate the message yourself instead of asking ChatGPT to write it. We compare the two approaches in speech to text vs AI writing.

The short version#

ChatGPT's dictation writes prompts and Voice holds a conversation, both inside ChatGPT. LumeVoice writes your words in ChatGPT and everywhere else. Your audio isn't stored after processing, it's never used to train models, and your dictation history stays on your device.