What voice coding means now
“Voice coding”, or programming by voice, covers two different habits.
- Speaking syntax. You say the symbols and keywords, and a specialist tool turns them into code. It takes practice, because you learn a spoken vocabulary.
- Speaking intent. You tell an AI assistant what you want in normal sentences. The assistant writes the code. You read it, run it and ask for changes.
The second is why many people now look up how to code by voice. If you can explain a problem to a colleague, you can explain it to ChatGPT, Claude or Cursor. Talkativ is built for that second habit. It is a dictation app, not a voice-coding tool.
Why prompts suit speaking
A good prompt has context, the goal, the limits and what you have already tried. That can run to a few hundred words. Saying it the way you would explain it to a colleague is an easy way to get all of it down.
For more on writing prompts by voice, see dictating prompts. For commit messages, comments and docs, see voice typing for developers. This guide covers the workflow around the AI assistant.
A simple workflow
- Click into the chat box in ChatGPT, Claude or Cursor.
- Press Ctrl + Space on Windows, ⌥ Option + Space on a Mac and say the task. Give the goal first, then the files or error, then the limits.
- Say what a good answer looks like. “Keep the change small and explain what you altered” is an easy line to add.
- Press again to stop. Your words appear a moment later, with punctuation.
- Read the prompt once before you send it. Fix any misheard term.
- Review the code the assistant returns. Run it. Voice does not replace testing.
What to say in a prompt
Four short parts keep a spoken prompt clear. First, the goal in one sentence. Second, the context: the language, the framework, the file or the error message. Third, the limits, such as “do not change the public interface”. Fourth, what you want back, such as code only, or code with a short explanation.
Describe error messages in your own words rather than reading them out character by character, then paste the exact text in afterwards if the assistant needs it.
Get technical words right
Dictation writes what it hears, so a library or an internal service name can come out wrong the first time. Add the terms you use most to Vocabulary: framework names, product names, teammates. Vocabulary is on every plan, including Free. It is a word list, not voice training.
With Advanced transcription formatting switched on in Settings, a spoken list such as “firstly, secondly, finally” becomes a numbered list, which suits step-by-step prompts. It is on every plan, and off until you switch it on.
Rewriting a prompt with Genius Mode
On the Max plan, Genius Mode changes text you have highlighted when you tell it what to do. Highlight a rough prompt, press Ctrl + D on Windows, ⌥ Option + D on a Mac, say something like “make this shorter” or “turn this into a bullet list”, and press again. It listens for up to 30 seconds. It is meant for prose, so don’t use it to edit code you have not reviewed.
Limits worth knowing
- No spoken commands for code, such as “open bracket” or “new line”.
- It needs an internet connection and does not work offline.
- It can mishear technical terms. Read what it typed.
- Words appear after you stop, not while you speak.
- Press once to start and again to stop. The shortcuts are fixed.
- There is no phone app.
Where Talkativ fits
Talkativ is a dictation app for Mac and Windows. For voice coding it handles the part that is ordinary language: prompts, explanations, comments and messages. It does not write syntax for you. Free is 2,200 words a week with no card, so you can try it on a few real prompts. See Talkativ for Windows or Talkativ for Mac for the details.
The same shortcut works in the chat around the work too: our guide to dictating in Teams and Slack covers that. If you want a plain explanation of what sits behind this kind of tool, read what is AI dictation.