SpeakribeDictation for ChatGPT
ChatGPT has its own voice mode, but it takes over the conversation: you talk to it, it talks back. Speakribe does something narrower and often more useful: it types what you say into the prompt box, and then you stay in a normal text chat, free to edit before you send.
Sure, want me to summarize it in three bullet points?
Summarize this contract in three bullets and flag anything unusual.
Why it helps here
The prompts that get the best answers are the long, context-heavy ones, and those are precisely the ones nobody wants to type. Built-in voice mode solves the wrong half of that: it makes the conversation spoken, when what you wanted was just to stop typing.
You still get to edit
The text lands in the prompt box, not in a live conversation. Read it, fix a word, add a line, then send. That is the whole advantage of text chat over voice mode.
Same shortcut everywhere
Browser tab, desktop app, or a different assistant entirely. Speakribe types into the focused field either way, so you are not learning per-app voice features.
Your prompt is transcribed locally
The transcription step happens on your Mac. What you then choose to send to ChatGPT is between you and them.
Things people say into ChatGPT
- “Summarize this contract in three bullets and flag anything unusual.”
- “Rewrite this email so it's warmer but still says no.”
- “I'm comparing two database options for a small app. Here's the context, then ask me what's missing.”
- “Turn these rough notes into a structured agenda for a 30-minute meeting.”
Setting up for ChatGPT
Nothing here is specific to ChatGPT except the last step, which is the one worth reading. ChatGPT already binds a microphone button and its own voice mode, so the point is to leave those alone and put Speakribe somewhere they aren't.
Pick a shortcut ChatGPT isn't using
Speakribe listens for one global key combination. A modifier-only binding like holding both Option keys is the safest choice here, because ChatGPT's web app already claims a lot of single-letter shortcuts and a modifier-only chord can't collide with any of them.
Click into the prompt box first
Speakribe types into whatever field currently has focus. That means the prompt box has to be focused before you start speaking, not after — if you speak with the page focused but no cursor in the box, the text has nowhere to land.
Speak the whole prompt in one go
Hold the shortcut, say the entire thing including the background you would normally skip, release. Long is the point: this is the input length that makes ChatGPT's answers better, and the reason you stopped typing it was effort, not judgment.
Decide whether it should press Enter for you
There is an auto-submit setting, off by default, that sends the message as soon as the text lands. For ChatGPT specifically, leave it off. The whole reason to dictate into a text box rather than use voice mode is the beat between finishing and sending, where you reread and fix.
One dictation, start to finish
A real prompt, dictated in one take of about twenty seconds. Typed, it is close to a minute of work, which is exactly why the version people actually type is shorter and vaguer than this.
I need help deciding something. We have a Next.js app on Vercel and the build has crept up to about eight minutes, most of it in one step that regenerates a few hundred static pages. I've already tried caching the fetch layer and it barely moved. Before I go and split the build, tell me what I'm likely to be missing, especially anything that only shows up once the page count keeps growing.
I need help deciding something. We have a Next.js app on Vercel and the build has crept up to about eight minutes, most of it in one step that regenerates a few hundred static pages. I've already tried caching the fetch layer and it barely moved. Before I go and split the build, tell me what I'm likely to be missing, especially anything that only shows up once the page count keeps growing.
- "Next.js" and "Vercel" came through with their capitalisation intact. Product names common on the web usually do. An internal codename usually does not, which is what the custom-words list in settings is for.
- The pause after "I need help deciding something" became a full stop rather than a comma, which is the call you would have made too. Sentence boundaries are inferred from how you speak, so a run-on delivery produces a run-on sentence.
- The filler at the start survived. That is deliberate — nothing is being rewritten behind your back, so what you get is what you said, and trimming it is a two-second edit you make with your eyes on the text.
What it won't do in ChatGPT
It won't drive ChatGPT for you
Speakribe types text. It cannot open a new chat, switch models, attach a file or click Send by voice. If you want to talk to an assistant rather than write to one, that is what ChatGPT's own voice mode is for, and this is not a replacement for it.
No spoken formatting or editing commands
Saying "new paragraph", "bullet point" or "scratch that" puts those words in your prompt. Structure comes out of how you speak and from the edit you make afterwards, not from command words.
The prompt box has to be in front
Text goes to the focused field. Dictating while a different window is in front sends your prompt to that window instead, which is a mildly alarming way to learn how focus works.
Terminal is a partial exception
macOS lets an app block keyboard monitoring — Terminal's Secure Keyboard Entry does exactly that, as do password fields. Speakribe falls back to a second mechanism there, but a modifier-only shortcut is the reliable choice if you move between ChatGPT and a terminal a lot.
Questions people actually ask
- How is this different from ChatGPT's voice mode?
- Voice mode is a spoken conversation. Speakribe just types for you: your words appear in the prompt box and the chat stays text, so you can revise before sending.
- Does it work in the browser and the Mac app?
- Both. Speakribe types into whichever field has focus, so it makes no distinction.
- Can I dictate in a language other than English?
- Yes, over 100 languages, including Hebrew, and you can switch mid-sentence without changing a setting.
- Can it send the prompt automatically when I stop talking?
- Yes. There is an auto-submit setting that presses Enter once the text lands, and you can change which key it sends if you prefer Cmd+Enter. It is off by default, and for ChatGPT we would leave it off: the pause before sending is where you catch the thing you said badly.
- Is my prompt sent anywhere before it reaches ChatGPT?
- No. Transcription runs on your Mac and the text goes straight into the prompt box; nothing is uploaded in between. Whatever you then send to ChatGPT is sent by you, to them, on your decision.
Speakribe in other apps
Ready to stop typing?
One payment. Every future update included.