use case · chatgpt
Good answers need context, and context is a lot of typing. Speak at around 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.
Put the cursor in the ChatGPT box — chatgpt.com in any browser, or the ChatGPT desktop app — press Ctrl ` , and talk. Your words appear right there as ordinary text: read it over, fix a name, cut a sentence, then send it yourself. Nothing reaches ChatGPT until you decide.
That is also what sets dictation apart from talking with ChatGPT out loud. A voice conversation is a conversation; dictation gives you a written prompt you still control — one you can edit, paste into a different chat, or keep for later.
Thinking out loud rarely comes out as a tidy prompt. Switch on the Prompt voice mode and a spoken stream of thought is reshaped into a clear, structured request before it lands. It runs on the built-in local model at no cost, or on an Ollama server you already have; Pro can send the rewrite through your own OpenAI account instead.
An error dialog, a spreadsheet formula, a chart that looks wrong, a paragraph in a contract. Describing it takes a paragraph; showing it takes one gesture. Press Alt ` , drag over just that part of the screen, circle the number or arrow the button while you ask about it, and paste into ChatGPT. The annotated screenshot and your transcribed question arrive together — no save, no find-the-file, no attach.
The full walk-through, including what works in ChatGPT, Claude, Gemini and Copilot: paste a screenshot into ChatGPT and ask out loud →
It is not a plugin or a browser extension. CroppickVoice types wherever your cursor is, so the same shortcut works in the next tab, the next app and the next AI — Claude included. Switch chats, switch tools; the button stays where it is.
No. CroppickVoice types into whatever ChatGPT you already use, the free version included. It never talks to OpenAI on its own; the only optional connection is Pro's cloud polish, through an OpenAI account you add yourself.
No. Speech becomes text on your device, and what reaches ChatGPT is the text you send it, nothing more. Audio is never uploaded to anyone, and the app tracks nothing about you.
Yes. The on-device model catalog covers the majority of languages across Europe, Asia and the Americas. Your language is detected automatically, or you can pin it in Settings.
Yes. It types at the cursor in any app, so a browser tab and the desktop app behave the same: put the cursor in the message box and press the shortcut.
Free, no account with us, and it works offline after a one-time model download. Related: dictating to Claude, asking AI about your screen and all use cases.