I explain the same thing on Looms constantly. Now each one becomes a searchable doc.
The workflow
- Copy the Loom transcript
- Ask Claude to structure it into steps
- Add headings and a short intro
- Publish to your help center
Result: A polished help article from a 5 minute video.
The prompt
Turn this transcript into a help article with a short intro, numbered steps, and a troubleshooting section. Transcript:
[PASTE]A note on what follows. The workflow, the prompt and the result above come from the member. The notes below (why the prompt works, what to watch for, how to adapt it) were written with AI help.
Why this prompt works
Asking for a troubleshooting section separately is what stopped the article being a transcript with headings. In a recording I say things like "if it does not show up, refresh", and left alone the model buries that mid-step where nobody finds it at the moment they need it. Pulling those asides into their own section is most of the value. Numbered steps sound obvious but they force a decision the transcript never made: what counts as one step. I talk in run-on sentences and the numbering is where the structure actually gets invented. The short intro is there because our old docs opened with three paragraphs of context and support kept sending people to the middle of the page anyway.
What to watch for
- Anything you did on screen without narrating is gone. I record a second pass now, talking through the clicks I skipped.
- Filler becomes instruction. "So I usually just, you know, hit save" turns into a confident step about a button that may not be labelled Save.
- UI labels get approximated. It writes what the button probably says rather than what it says.
- My tangents survive. If I went off about a related feature for 30 seconds, it becomes a section.
How to adapt it
- Tie it to reality:
Use exact UI labels only if I said them out loud. Otherwise write [CHECK LABEL]. - For screenshots, add
Mark where a screenshot is needed with [SCREENSHOT: what it should show]. - If the recording rambles, add
Ignore anything not part of the main task, even if I spent time on it.
What good output looks like
I read only the numbered steps and try to do the task from them. If I have to remember something from the video, the doc is not finished. Troubleshooting should contain the specific failures I mentioned, not invented ones. Any [CHECK LABEL] left in is fine, that is the doc telling me where to look, and I would rather have five of those than one confidently wrong button name.