Pithflow

Talk to type in ChatGPT on Windows

ChatGPT does have dictation — the mic in the prompt box transcribes what you say and lets you edit it before sending. What it won't do is work anywhere outside ChatGPT. Windows voice typing (Win + H) will, but it transcribes literally. Below: the steps, and which one suits long prompts.

Get started on Windows 2,000 words per week free · no card

Rated 4.8/5 in our customer survey

The steps, in order

1

Open ChatGPT on your PC

Desktop app or the browser tab — for dictation they're the same text box.

2

Click the prompt box

Dictation types at the cursor, so the box has to have focus first.

3

Hold the key

Ctrl + Space with Pithflow. Win + H if you're using Windows voice typing.

4

Say the whole prompt

Who you are, what you need, what you already tried, what format you want. Take your time — this is where the answer is won.

5

Release the key

That ends the recording, and the cleaned-up text lands in the box.

6

Read it, then send

The prompt waits for you. Adjust a detail, add a constraint, then Enter.

Step 5 is the one that trips people up

Text does not appear while you're talking. The real order is this:

  1. Hold
    Ctrl + Space
  2. Speak
    the whole prompt
  3. Release
    recording ends
  4. It lands
    cleaned-up text

On a long prompt that pause on release is worth having. The paragraph arrives in one piece and in order, rather than word by word with your second thoughts still in it.

What about the microphone ChatGPT already has?

It exists, and it helps to know there are two separate features. Dictation is the microphone in the prompt box: it records, transcribes, and gives you back text you can edit before sending. Voice is a spoken back-and-forth conversation; per ChatGPT's documentation, in the desktop app it's listed for the Plus, Pro, Business, Edu and Enterprise plans.

When the button is missing or picks up nothing, it's nearly always one of three things:

With that fixed, the harder question is the one that isn't technical: why is your prompt so short?

The bottleneck isn't the answer, it's the prompt

Everyone knows a prompt with context beats a one-liner. Most of us send the one-liner anyway, because typing the context is work and a generic answer shows up regardless. Say it out loud and that cost disappears.

Context is the first thing you cut

Typed, you write "draft an email cancelling my service." Spoken, you say who it goes to, how long you've been a customer, what they told you last time, and what tone you want. Those are two different prompts and they come back as two different answers.

You get to think out loud

Ramble, correct yourself halfway through, start the sentence again. None of that reaches the box — the AI cleanup drops filler and false starts. Literal dictation would force you to speak as if you were typing.

The same key outside the tab

ChatGPT's mic lives inside ChatGPT. The answer, though, ends up in an email, a document, or your editor — and that's where you go back to the keyboard if your dictation can't follow you there.

Give it the whole context without typing it

One key in ChatGPT, in your email, and in your editor.

Get started free →

The three options, compared

  ChatGPT dictation Voice typing (Win + H) Pithflow
Works outside ChatGPT No, in there only Yes Yes
Removes filler and false starts No, transcribes what you said No, transcribes literally Yes
Editable before you send Yes Yes Yes
Spoken back-and-forth conversation Yes, that's Voice mode No No
Price Included with ChatGPT Free, built into Windows Free up to 2,000 words/week
Works offline No No No

Two honest concessions there. If what you want is to talk with ChatGPT and have it answer out loud, that's its Voice mode and Pithflow doesn't do it — we put text in a field, we don't hold a conversation. None of the three works offline: Microsoft documents voice typing as using online speech recognition and requiring an internet connection. If you need to dictate with no connection, the tool for that is Windows 11's Voice Access, which recognises on-device.

What changes in the prompt

Nobody speaks a prompt in clean sentences on the first pass. This is the kind of repair the AI cleanup does — an illustrative example, not a measurement:

What you say

"okay so i need you to um — i've got a table of sales by store, and i want like a summary for my manager, not technical, maybe half a page, saying which store went up and which one went down and a guess at why"

What gets typed

"I have a table of sales by store. Write a half-page summary for my manager in non-technical language: which store went up, which went down, and a hypothesis for why."

How dictating in ChatGPT works with Pithflow

There's nothing to connect and no extension to install: Pithflow is a Windows app that types into whatever field has focus. For ChatGPT that means the prompt box, exactly as if you'd typed it. And the same key works two seconds later in the email where you'll paste the answer, or in the document where you'll edit it — you don't switch tools per window.

It's the same mechanic everywhere. If you're weighing up options, the best dictation apps for Windows covers the field, and if your prompts are mostly code and specs, dictation for developers goes into that workflow.

Frequently asked questions

Does ChatGPT already have dictation?

Yes, and it's two separate features. Dictation is the microphone in the prompt box: it records, transcribes, and hands back text you can edit before sending. Voice is a different thing — a real-time spoken conversation — and per ChatGPT's documentation, in the desktop app it's listed for the Plus, Pro, Business, Edu and Enterprise plans.

Then why would I want a separate dictation app?

Because ChatGPT's microphone types into ChatGPT and nowhere else. The answer you get almost always ends up somewhere else — an email, a doc, your editor, a team chat. The moment you leave the tab you're back on the keyboard. Pithflow is the same key in all of those windows.

Why is the microphone button missing, or my ChatGPT dictation not working?

Three usual causes. The browser has the microphone blocked for the ChatGPT site — check the padlock in the address bar. Windows has microphone access turned off for desktop apps — Settings, Privacy and security, Microphone. Or a blocking or privacy extension is intercepting capture; test with extensions off. It also helps not to confuse the two icons: the mic in the composer is dictation, the other button starts a voice conversation.

Does it work in ChatGPT on the web, or only the desktop app?

Both. Pithflow types into whatever field has focus, whether that's the ChatGPT app for Windows or the ChatGPT tab in your browser. To the app it's the same text box.

Does it send the prompt automatically when I stop talking?

No. The text sits in the box and you decide when to send. That's deliberate — on a long prompt you usually want to reread it, fix a name, or add one more constraint before you hit Enter.

Will it get technical terms, library names, and work acronyms right?

You can add your own terms to a personal dictionary and switch on vocabulary packs for your field, so the words you say every day stop coming out mangled. If something lands wrong once, you correct it from the history and it sticks.

Can I mix English and Spanish in the same prompt?

Yes, and in technical prompts it happens constantly — library names, error strings and field names in English inside a Spanish sentence. Switching mid-sentence is handled without touching a language setting.

What happens to my audio when I use ChatGPT's own dictation?

Per OpenAI's Voice Dictation FAQ, ChatGPT keeps dictation audio for as long as the chat stays in your history, and deletes it within 30 days of your deleting the chat. Pithflow sends your audio to its servers to be transcribed and does not store it.

What does it cost?

There's a permanent free tier of 2,000 words per week, no card required. For heavy use, Pro is $9.99 USD per month. Full pricing is on the pricing page.

Want a different key?

Open Pithflow, go to Settings → Hotkey, click the combination you want to replace, and press the new one. It takes effect immediately.

  • Hold to speak — Ctrl + Space by default. Hold it, talk, let go.
  • Hands-free — Ctrl + Shift by default. Tap to start, tap again to stop.

Worth changing if the default clashes with a shortcut you use. A spare function key such as F9 is claimed by almost nothing.

Long prompts without the long typing

The same key in ChatGPT, in email, and in any text field on Windows.

Get started on Windows

2,000 words per week free · no card · cancel anytime