Speak 500-word prompts directly into the ChatGPT input box, in any browser or the desktop app. StarWhisper is a local-first voice-to-text hotkey for Windows that works in ChatGPT, Claude, Gemini, Perplexity, Cursor, and any text field.
From install to dictating directly into ChatGPT in under five minutes.
Download StarWhisper from starwhisper.ai or the Microsoft Store. The installer takes about a minute. On first launch, allow microphone access. The free plan covers 500 words per day, which is plenty for several long prompts.
Navigate to chatgpt.com in Chrome, Edge, Firefox, Brave, or any browser. Or open the official ChatGPT desktop app for Windows. The dictation flow is identical for both. Start a new conversation or continue an existing one.
Place the cursor inside the message text input at the bottom of the conversation. This is the field labeled "Ask anything" or similar. StarWhisper types into whichever Windows text control currently has focus, so the cursor needs to be in the right spot before you start.
Press and hold the global dictation hotkey. The default works for most setups, and you can rebind it in Settings if you prefer something else. The StarWhisper icon shows recording state so you know your microphone is live before you start speaking.
Dictate the entire prompt at the pace you would naturally speak. Long instructions, context, examples, constraints, requested output format, all of it. Pause where sentences end. Whisper handles punctuation automatically. You can speak in any of 96 supported languages.
When you release the hotkey, StarWhisper transcribes locally and pastes the result into the ChatGPT input box. Read it, fix any words Whisper misheard, add a clarification if needed, then hit Send. Your prompt only reaches OpenAI when you press Send, exactly as if you had typed it.
Specific advantages for users who write long, structured prompts.
The prompt arrives as text in the input box, where you can read it, restructure it, add bullets, paste in code, and refine before sending. Voice mode commits the moment you finish speaking.
ChatGPT, Claude, Gemini, Perplexity, Mistral, You.com, Poe, OpenRouter chat, any browser-based AI receives dictated text identically. Same for Cursor, VS Code, JetBrains, Slack, Discord, Notion, Word, Gmail.
Hold the hotkey, dictate at your pace, release when finished. Multi-paragraph prompts with examples and instructions arrive complete. No need to stay inside a voice-mode conversation window or worry about ChatGPT cutting you off.
Audio is processed on your PC with Whisper running locally. Your speech is not uploaded to any third-party transcription service before it reaches ChatGPT. The text only goes to OpenAI when you decide to press Send.
500 words per day on the free plan covers several long prompts a day. Pro is $10 per month or $80 per year for unlimited dictation across all your daily writing, not just ChatGPT.
Dictate prompts in your native language and ask ChatGPT to respond in any language you prefer. Useful for content creators, translators, and anyone whose thinking is faster in one language than another.
The longer you use ChatGPT, the longer your prompts get. A casual user writes "summarize this article". A power user writes a multi-paragraph brief with role assignment, context, examples of good and bad output, constraints, target format, and a list of edge cases to handle. That prompt is 300 to 800 words. Typing it takes five to fifteen minutes. Speaking it takes one to three.
The speed gap matters more than it sounds. Prompt quality is the single biggest variable in ChatGPT output quality. When typing a long prompt feels slow, you cut corners, leave out context, skip the examples that would have steered the response. When dictating is fast, you include everything. The model gets a better brief, the output gets better, and the back-and-forth gets shorter. StarWhisper is built to make this loop quick.
Dictation is also less fatiguing for repeated work. Anyone who runs ChatGPT all day, content marketers, copywriters, developers, founders, support engineers, ops people automating workflows, knows that the cumulative wrist load of typing prompts adds up. Switching to voice for the input side cuts that load roughly in half.
OpenAI has its own voice mode for ChatGPT. It is a great product for a different use case. Both deserve a clear comparison.
| Capability | ChatGPT Voice Mode | StarWhisper dictation into ChatGPT |
|---|---|---|
| Best for | Conversational back-and-forth | Long structured prompts, edit-before-send |
| Output you get | Spoken or text reply, in a voice session | Text in the prompt box you can refine |
| Works with Claude/Gemini/Perplexity | No | Yes, identical flow |
| Works in Cursor, VS Code, Word, Notion | No | Yes, any text field |
| Audio handling | Streamed to OpenAI | Processed locally in default Local Mode |
| Subscription | Requires ChatGPT Plus or Team | Free plan covers daily dictation, Pro $10/mo |
| Languages | Supported set is smaller | 96 via Whisper |
Voice mode is great when you want to chat with ChatGPT like a person. Dictation into the prompt box is better when you want to write a careful, detailed brief, edit it, and only then send.
Dictate the brief for a 1,500-word article. Speak the angle, the target audience, the three subtopics, the call to action, and the brand voice notes. Edit the dictated brief, send to ChatGPT, get a draft. Repeat for outlines and rewrites. For more on this, see voice-to-text for content creators.
Dictate the description of a refactor in plain English, paste the existing code, ask ChatGPT or Claude for the change. Or dictate test cases as natural-language descriptions. Works equally well in Cursor and VS Code, both of which are just text inputs to StarWhisper.
Dictate a long question with all the relevant context, sources, and constraints you would otherwise summarize. Get a more grounded answer because the model has the full brief from the start.
Dictate strategic prompts on a walk or commute (with a headset mic on Windows). Edit when back at the desk. Send. This is how a lot of strategy work happens in 2026.
This is a frequent and reasonable question. StarWhisper Local Mode runs Whisper on your own CPU or GPU. The audio is captured by your microphone, processed in memory on your device, and converted to text without any network call. Nothing is uploaded anywhere during transcription. The text that StarWhisper hands off to ChatGPT's input box is the same text you would have typed.
When you then press Send in ChatGPT, your text prompt reaches OpenAI's servers, which is no different from typing manually. If your concern is OpenAI seeing the prompt content, dictation does not change that. If your concern is a third-party transcription service receiving your raw audio, Local Mode addresses that completely.
There is an opt-in Cloud Mode for cases where you want maximum accuracy on hard audio. It uses the OpenAI Whisper API. It is never enabled by surprise, the choice is visible in the StarWhisper UI, and you can stay on Local Mode permanently if that is what you prefer.
With a few sessions, the workflow becomes natural and the speed gain over typing is large enough that most users do not go back to keyboard-only prompts.
StarWhisper is a system-wide hotkey for Windows. The dictation surface is "whatever text field has focus right now." That means the same flow you use for ChatGPT works for:
One install, one hotkey, every text input on the OS gets voice-to-text. That is the practical reason regular ChatGPT users adopt it for more than just ChatGPT after a few days.
Same hotkey, different surfaces.
Overview of every way to use voice in the ChatGPT workflow.
Long-form writing inside Docs using the same hotkey.
Capture notes, meeting recaps, and pages by voice in Notion.
How writers use voice to draft AI prompts and full posts.