Claude Code /voice: test the words and the workflow

Start with the terminal voice example

The 19 March 2026 video tests Claude Code voice input against ChatGPT speech-to-text and WhisperFlow. Luke uses his Scottish accent as a practical challenge and reports a daily preference for WhisperFlow. That is his comparison and experience, not proof that every accent or microphone will produce the same ranking.

1. Follow the terminal test at 0:18 and use the exact command named in the description: /voice.
2. Dictate a short instruction whose intended wording you can record for comparison.
3. Inspect the transcribed words before submitting them as a coding instruction, especially names, numbers and technical terms.

Separate raw transcription from usable dictation

At 0:58 the video discusses raw dictation, then demonstrates WhisperFlow at 1:40 and ChatGPT as a free alternative at 2:00. Accurate words and a useful final instruction are different results. Punctuation, false starts and the placement of the text can still affect how easily the next application understands your request.

1. Repeat the same spoken passage across the compared tools rather than changing the task for each one.
2. Note transcription mistakes separately from edits needed to turn the text into a readable instruction.
3. Check the final text before pasting or submitting it; fluent cleanup should not quietly change your meaning.
4. For the WhisperFlow route, test the description’s paste-anywhere-on-Mac use case in the application where you actually need the text.

Choose for the place you work

WhisperFlow history and statistics appear at 3:26, followed by the verdict at 3:57. Terminal-only input and system-wide dictation have different reach. A tool can transcribe well but still be inconvenient if you constantly need the result somewhere else. The source’s free-option discussion reflects the recording date, not verified current plan eligibility.

1. Review available history to compare the text you spoke with what was ultimately inserted.
2. Record the corrections, extra copying and context switches needed for one complete task.
3. Choose the tool that produces an accurate usable instruction in your actual workflow, and recheck current access terms before relying on a free route.

The takeaway

The useful comparison measures a completed dictation task, not only the apparent accuracy of the first transcription.