Voice-Driven Code Review: Talk Through the Diff
Reviewing a pull request is mostly explaining your thinking. You read a diff, you form an opinion, and then you slow to a crawl typing that opinion into a comment box. Voice-driven code review flips that. You read the change, you talk through it, and the words appear cleaned up and ready to post.
Key takeaways
- Review comments are prose, not syntax, which is exactly what dictation handles well.
- System-wide dictation drops your spoken note straight into GitHub, an editor, or an AI chat.
- On-device processing keeps private code and your voice on your Mac, never on a server.
- You can also narrate a diff to an AI agent and ask it to double-check the change.
Why review is the perfect job for your voice
Dictating code line by line is a bad idea. Syntax, brackets and variable names do not survive being spoken. But a code review is not code. It is a running commentary: "this loop reruns the query on every iteration, pull it out," or "the null check here is redundant now that the caller guarantees a value." That is plain English reasoning, and prose is where voice shines. Most people speak around three to four times faster than they type, so the friction between having a thought and recording it drops sharply.
The result is that you review at the speed you think. You keep your eyes on the diff instead of ping-ponging between the code and the comment box, and you tend to write fuller, more human feedback because talking is cheaper than typing. If you already do other kinds of coding by voice on your Mac, review is the lowest-risk place to start, because a stray word in a comment never breaks a build.
How the workflow fits together
The pipeline is short. Your microphone captures the spoken note, an on-device model transcribes it locally, an on-device AI pass cleans up the filler and punctuation, and the finished text lands in whatever field your cursor sits in, whether that is a GitHub thread or your editor.
Because the dictation works system-wide, it does not care whether you are in a browser tab, a native app, or a terminal-based review tool. It types where your cursor is. That same trait is what lets you talk to an AI assistant with your voice during review, which we will come back to below.
Setting it up in four steps
Install an on-device dictation app
Download BlaBlaType for macOS and grant the accessibility permission it needs to type into other apps. The speech models run locally, so nothing is uploaded.
Pick a comfortable shortcut
Set one keyboard shortcut to start and stop dictation. Push-to-talk works well for review: hold the key while you speak, release when you finish the thought.
Add your project vocabulary
Drop your service names, class names and jargon into the custom dictionary so terms like your repo names and internal acronyms transcribe correctly instead of turning into guesses.
Open a diff and start talking
Put your cursor in the review comment box, hold the shortcut, and describe the issue. Release, glance at the cleaned text, and post. Repeat down the file.
Two ways to talk through a diff
There is more than one shape to this workflow, and they compare cleanly. The first is straightforward dictation: you are the reviewer, you speak your own comments, and the app just types them. The second is conversational: you narrate the change to an AI coding agent and ask it to sanity-check your reasoning or flag things you missed.
| Approach | What you do | Stays on device | Best for |
|---|---|---|---|
| Dictate comments | Speak your own review notes into the comment box | Yes | Human review of a teammate's PR |
| Narrate to an agent | Describe the diff to an AI and ask it to check the change | Depends on the AI | Reviewing your own work before you push |
| Type it all | Write every comment by hand | Yes | Very short, one-line changes |
For the second approach, the dictation stays local, but remember the AI you talk to may not. If you point your prompt at a hosted agent such as one configured in Cursor, the text you send leaves your machine. If you want the whole loop private, pair local dictation with a locally run model through something like Ollama, so both your voice and your code stay on the Mac. Either way, the dictation layer never uploads your audio.
Review at the speed you think
Talk through any diff and get clean, punctuated comments in any app, with every word processed on your Mac. No card needed for the trial.
Download for macOSThe privacy question you cannot skip
Code review means talking out loud about a private codebase, sometimes one under an NDA. That makes the difference between cloud and on-device dictation more than a preference. A cloud tool records your microphone, ships the audio to a server, and transcribes it there. In a review workflow that audio includes function names, architecture decisions, and whatever you say about the change. On-device dictation removes that exposure entirely: the model runs on your Mac, so the audio and the transcript never leave it.
This is the same reasoning that applies to any spoken work you would rather keep to yourself, from client notes to a private voice journal. If you want the full picture of where your words can end up, we broke it down in a separate piece on whether Mac dictation is private. For review specifically, the on-device rule is simple: nothing spoken about the code should touch a server you do not control. You can see how BlaBlaType handles this on the pricing page, where the on-device guarantee applies to every plan.
Frequently asked questions
What is voice-driven code review?
Voice-driven code review means reading a diff on screen and speaking your feedback aloud instead of typing it. A dictation app turns your spoken notes into clean review comments, so you narrate reasoning at the pace you think rather than the pace you type.
Is dictating code review comments private?
It depends on the app. Cloud dictation uploads your audio to a server, which can be a problem for private repositories under an NDA. BlaBlaType runs speech recognition entirely on your Mac, so your voice and the code you talk about never leave the device.
Can I talk through a diff to an AI coding agent?
Yes. Because system-wide dictation types wherever your cursor is, you can speak a prompt straight into a chat with an AI agent like Cursor or a local model, describing what the diff should do and asking it to check the change.
Does voice review work in GitHub and my editor?
Yes. On-device dictation that works system-wide types into any text field, so it drops your spoken comment into a GitHub review box, a pull request thread, or the comment field in your editor without copy and paste.
Is voice really faster than typing review notes?
For prose feedback, usually yes. Most people speak around three to four times faster than they type, and review comments are mostly plain-language reasoning rather than code, which is exactly what dictation handles well.