On a Windows PC, voice to text for ChatGPT, Claude or Gemini comes down to one gesture: a dictation shortcut that writes into the message box, whatever the app. Windows has a free one (Windows + H), which works online and cuts out on silence; an app like Ondula does the same thing offline, with no limit, in every assistant at once.
You have three paragraphs in your head for ChatGPT: the background of the project, what you've already tried, the tone you want back. At the keyboard, that shrinks to two lines, and the answer misses the point. Out loud, the three paragraphs come out in a minute.
In short
- To write a long prompt by voice, you need dictation that fills the text box, which you read over before sending. An assistant's voice mode is a conversation, not dictation.
- On the assistants' side, the offer is patchy: as of 09/10/2026, the Gemini app for Windows had no dictation, and the Mac one only spoke English.
- Windows + H works in every assistant, but it needs the internet and stops after a silence.
- Ondula types at your cursor in any assistant, offline, free and with no limit. With Cloud, its AI modes tidy up a prompt you said in one go, and "Free" continues from what's already typed.
Voice mode vs dictation: why they're not the same thing
Voice mode, when an assistant offers one, turns it into someone you talk with: you speak, it answers, the exchange goes on. Handy with your hands in the dough, for a short question. For a real prompt, it's the wrong tool: you don't see your message before it's sent, you can't fix a misheard name, you can't paste the document to analyze right below.
Dictation, on the other hand, writes what you say into the input box. You read it over, fix a word, add an attachment, then send when you're ready. That's what people searching for "voice to text for ChatGPT" are after: talking instead of typing, without losing control of the text. And when a tool's dictation fails, they search for that too: "dictate chatgpt not working" was among the Google suggestions we collected on October 20, 2026.
What do the assistants themselves offer for dictating on a computer?
Each company is moving on its own. Here's what we were able to document, with the date we checked:
| Assistant | What exists | Limit found |
|---|---|---|
| Gemini for Mac | Dictation in any app (Fn key), since 07/29/2026 | English only; Mac only |
| Gemini for Windows | No dictation (09/10/2026) | |
| Gemini in Chrome | Dictation in any web field announced as "coming soon" with Gemini 3.5 Transcribe (08/26/2026) | No date |
| Claude Code | /voice command since March 2026, according to third-party pages (checked on 10/20/2026) | Online; unavailable with an API key, Bedrock or Vertex AI; only in Claude Code |
For ChatGPT and Claude on the web or in their apps, check what your version offers. The practical takeaway doesn't change: built-in dictation stays with its own assistant. If you move from ChatGPT to Claude during the morning, then to the assistant in your code editor, none of them follows you.
How to dictate into ChatGPT with Windows + H, with nothing to install
Windows voice typing writes into any field, so into ChatGPT, Claude or Gemini, in the browser or in their apps.
- Click in the assistant's message box.
- Press Windows + H. A small panel opens with a microphone.
- Speak. The text appears as you talk.
- Read it over, fix it with the keyboard, send.
Two limits weigh on long prompts. Voice typing requires the internet: Microsoft says it goes through its online services. And it cuts out after a silence, "by design" according to Microsoft: you search for your sentence for two seconds, and you have to start it again. For three lines, that's bearable; for a two-minute brief where you think as you talk, it breaks your train of thought. If it won't start at all, our Windows voice typing troubleshooting guide goes through the causes.
One dictation tool for every assistant
Ondula doesn't know about any assistant in particular, and that's the point: you press a shortcut (F9 by default, changeable), you speak, your words show up live in the live window, then the text is pasted at your cursor. Whether that cursor is in ChatGPT in Edge, Gemini in Chrome or the Claude app makes no difference.
What matters for a long prompt:
- No cut-off on silence. The shortcut starts dictation, another press stops it. Go quiet while you remember a name: nothing moves. The only cap is two hours per dictation offline.
- Offline, free, unlimited. Transcription happens on your PC, on the graphics card whatever its brand or on the processor, with no account.
- Fixable before sending. A misheard word can be reworked in the live window; dictation waits.
- Nothing gets sent for you. By default, Ondula pastes the text without pressing Enter: you read it over, then you send. A "Press Enter afterwards" option exists for those who want the opposite.
One honest caveat: Ondula only protects the voice step. The text you then send to ChatGPT, Claude or Gemini goes to that assistant, under its own terms.
AI modes: your prompt put in order before it's pasted
When you speak, you backtrack, you correct yourself, you add a constraint at the end. An assistant can cope with that, but a clean brief gets better answers. That's the job of AI modes, included with Cloud.
A shortcut opens a modes wheel around your mouse. Each mode has a name, an icon, its own shortcut, and above all an instruction you write yourself. An example "Brief" mode: "Put my dictation in order: context, request, constraints, format of the answer. Don't add anything." You dictate in one go, and the ordered version shows up as a preview: Confirm (Enter) pastes it, Cancel (Esc) discards it, Retry lets you dictate a comment ("shorter", "keep the list of files") in the same mode. You can also click into the answer to fix it by hand.
The "Free" mode, always on the wheel, does something else: it continues from what's already typed, without repeating it. You started a prompt at the keyboard and pasted an excerpt from a contract; you dictate "add that I want three options ranked by risk", and the sentence slots in where you left off. A mode can also attach the window you had in front of you, so it knows what you're responding to.
Modes go through our servers, which is why they're part of Cloud: €5.99 a month, cancel anytime, with 3,000 modes and 20 hours of online audio each month. The trial lasts 14 days, no credit card required. Details on the pricing page.
How to dictate a long prompt that gets a good answer
When you speak, you naturally give more context than at the keyboard. The order that works: the situation, what you want, what to avoid, the format of the answer.
A tricky letter, in ChatGPT:
"My landlord is keeping part of my security deposit for paint that was already damaged when I moved in. I have the move-in inspection report with photos. Help me write him a reply that's firm but polite, cites the inspection report and asks for a refund within fourteen days. No threats, no legal jargon. One page at most."
Reviewing a document, in Claude:
"I'm going to paste the minutes from yesterday's site meeting. Don't rewrite them. Just list the decisions made, who has to do what, and the items left without an owner. If something is ambiguous, ask me instead of guessing."
A few habits make the difference:
- Dictate first, paste second. The brief by voice, then the document below it with the keyboard.
- Think out loud. "No, actually, make it a neutral tone": the assistant follows the correction.
- End with a method instruction. "Before answering, ask me any questions you need" avoids generic answers.
- Turn your standing instructions into voice shortcuts. Say three words and Ondula pastes "answer in plain English, no bullet points, one paragraph". Shortcuts and the dictionary are part of the free tier.
- If you'd rather think in your own language, translated dictation writes English directly while you speak Spanish, German or Polish, offline.
For technical prompts (terminal, commits, jargon), our article on voice typing for developers goes into detail.
Pitfalls to know about
The mic is already taken. If the assistant's voice mode or a video call is using the mic, dictation may hear nothing. Close one before opening the other.
The open-plan office. Dictating a brief in the middle of your coworkers bothers everyone. A headset with the mic close to your mouth lets you speak softly.
The PC that struggles. On a modest laptop, Cloud's online dictation is lighter on the machine than offline dictation.
Frequently asked questions
How do I turn on voice to text in ChatGPT on a PC?
With nothing to install: click in the message box and press Windows + H, which needs the internet and stops after a silence. To dictate at length, or offline, a dictation app like Ondula writes into ChatGPT just as it does everywhere else.
Does Gemini have dictation on Windows?
Not in its app as of 09/10/2026, unlike the Mac app, which is limited to English. Google has announced dictation in any web field in Chrome, with no date. In the meantime, Windows + H or Ondula types into Gemini like into any other page.
Is my voice sent to OpenAI, Anthropic or Google?
Not by Ondula: offline, your voice is transcribed on your PC, and only the text you send goes to the assistant. Online, the audio is used only for transcription and is then kept by neither Ondula nor its provider.
To try it, download Ondula for free from the Microsoft Store: offline dictation has no limit.
