use case · chatgpt
Talk to ChatGPT. Show it your screen.
Good answers need context, and context is a lot of typing. Speak at ~150 words a minute instead of typing at 40 — and when the question is about something you can see, send the thing itself, marked up, with your words attached.
dictate your prompts
Put the cursor in the ChatGPT box — browser or desktop app — press Ctrl ` , and talk. Your words appear right there as text you can read over, fix, and send when you're ready. That's the difference from a voice conversation: ChatGPT's spoken mode is great for chatting out loud, while dictation gives you a written prompt you still control before it goes anywhere.
Rambled a bit? The Prompt voice mode turns a spoken stream of thought into a clear, structured request before it lands — free on the built-in local model or your own Ollama server, and Pro can run it through your own OpenAI or Claude account instead.
show it, don't describe it
Some questions are really about a thing on your screen: an error dialog, a line in an invoice, a chart that looks off, a form that won't submit. Describing it in words is the slow way.
- 1
Press Alt `
Drag over just the part that matters — recording starts as you do.
- 2
Mark it up while you talk
Circle the number, arrow the button, say what you need to know about it.
- 3
Paste into ChatGPT
The annotated screenshot and your transcribed words arrive together — no save, no find-the-file, no attach, no typing it all up.
private by default
Your speech becomes text on your device — audio isn't uploaded to anyone, and the app tracks nothing about you. What reaches ChatGPT is exactly what you chose to paste there, nothing more. The full network-use table →
Windows 10/11 today — macOS and Linux builds are on the way. Also see dictating to Claude and speech to text for Windows.