Improve an AI response

Make feedback specific

When Claude or ChatGPT returns a draft that misses the mark, stay in the same conversation. Point to the part you want changed, explain the desired result, and say what should remain. Whisper One enters your feedback as text; the assistant applies it after you submit.

A three-part follow-up

1. Click the assistant's message field and press Fn.
2. Identify the target: the opening sentence, the second paragraph, the tone, or a particular claim. Quote a short fragment if the target is ambiguous.
3. Say the change and the boundaries. Press Fn again, check the transcription, and submit.

Something to say

Revise only the opening paragraph of your last draft. Start with the decision we need from the reader. Keep the three project facts and the closing question unchanged. Make the opening no longer than two sentences.

This is easier to evaluate than “make it better.” Expected result: the assistant revises the named part. Compare the new version with the original to confirm it kept the facts you named.

When the answer keeps drifting

Ask for one change per message. If it introduced an invented date, say that the date is unknown and request its removal. If you need a different structure, describe the headings or bullet count. For long feedback, use multiple recordings under 30 seconds before submitting.

Speaking “delete that sentence” into an ordinary text field can insert those words literally. These instructions are intended for Claude or ChatGPT, not a promise of system-wide voice editing commands.