This article is primarily about the Claude.ai feature called “projects,” but his example project is about using Claude to cleanup the nasty, incoherent speech-to-text output that Word and similar programs spit out. I’ve tested the commands involved in the free version of Claude, and they work reasonably well. I still went back and reworded and expanded a bunch of stuff after I used it, but it was nice having something that would reliably cut out the nonsense words and repetitions and add some kind of punctuation. Here are the commands; just copy and paste into Claude’s chat window, and upload a short document with your latest chunk of dictation. As always with AI, check the company’s privacy and training policies before feeding it anything of a personal or sensitive nature.
Given a raw transcription, generate a cleaned-up version by following these guidelines:
- Provide your response in an artifact.
- Remove filler words (e.g., “uh”, “mm”, “like”) while preserving the core content.
- Correct mistranscribed words. Provide a brief summary of corrections made.
- Reorganize ideas into a more logical flow, creating coherent paragraphs.
- Maintain the speaker’s voice, style, and word choices as much as possible. Use their original sentences when they are clear and make sense.
- Use active voice predominantly. Only use passive voice when absolutely necessary.
- Preserve the speaker’s core word choices and distinctive expressions.
- Retain all technical terms, proper nouns, and content-specific vocabulary.
- Improve grammar moderately without making the language overly formal.
- Use {curly brackets} to indicate significant ambiguities or suggest clarifications.
- Address repetitions: Remove or consolidate repeated phrases or ideas while preserving the speaker’s emphasis where appropriate.
- Ensure proper punctuation and capitalization throughout the transcription.
- Complete or clarify incomplete sentences where possible, using {curly brackets} for significant additions.
Example output format:
Cleaned Transcription:
[Insert cleaned transcription here]
Corrections Made:
- Original: [mistranscribed word] Corrected: [correct word]
- Original: [mistranscribed word] Corrected: [correct word](List all corrections)
Please review the corrections and provide feedback if any are incorrect. I’ll revise the transcription based on your input.

5 thoughts on “How to Use Claude.ai for Cleaning Up Transcriptions”