Turn rambling notes or a voice memo into clean structure

Paste the raw mess into a plain chat, name the output shape you want, and get something you can act on.

What you'll have when you're done

A working method for turning any raw input (a voice memo transcript, a page of rough meeting notes, a scattered brain dump) into a usable document: decisions captured, action items owned, open questions visible. Plain chat, no setup, runs in under five minutes.

The voice memo you haven't touched

You recorded a voice memo on the drive back from a client meeting. Nine hundred words of you thinking out loud, some of it brilliant, some of it filler, some of it three things said twice. Your notes app has a transcript. You need a clean action list by morning and you haven't opened the file once.

Or you have a page of shorthand from a strategy session. Arrows, fragments, names with asterisks next to them, two timelines that may or may not be the same timeline. You know what it means right now, but in 48 hours you won't.

Or you sat down and typed everything in your head for six minutes because a coach told you to. The result is 400 words of disconnected threads and you have no idea how to read it.

The problem is not the mess. The problem is that "clean this up" is not an instruction. It's a request for a paraphrase, and that's exactly what you'll get back: a slightly shorter version of the same mess, restructured but still shapeless.

The fix is to name the output shape before you send. The model has no idea whether you need action items, a decision log, or a one-pager. Tell it the form you want out, and it reshapes the entire response around that.

What you need first

• A Claude or ChatGPT account. If you are still setting up, your first task is basic access, nothing else.
• Your raw text. The messier the better for your first test. A voice transcript, rough notes, or a brain dump you typed in one go.
• A browser. Plain chat only. No terminal, no plugin, no project required.

Step by step

The core move

Open a new chat. Paste your raw notes. Before you hit send, add one line that names what you want out. That line is the whole skill.

Without it: you get a compressed version of what you put in.

With it: you get a structured document shaped around your actual need.

Five ready-to-use asks

These cover most situations. Pick the one that matches what you need before you paste:

• "Organize this into themes with bullets under each."
• "Pull out every action item, who owns it, and any deadline."
• "Give me a one-paragraph summary, then the open questions."
• "Turn this into a one-pager: decision, context, options."
• "What did I actually decide here, and what is still unresolved?"

None of these are long. The value is in the specificity, not the word count. You are telling the model what role the output plays in your day, and it builds the whole response around that.

By input type

Different input shapes call for different asks. Here is a tailored prompt for each:

Voice memo transcript: "This is a voice memo transcript. Ignore filler and repetition. Pull out the decisions, the action items with owners, and the open questions."

Rough meeting notes: "These are rough notes with shorthand. Organize into what was decided, what needs follow-up, and who is doing what."

Scattered brain dump: "This is a brain dump with unrelated threads mixed together. Group into themes, label each, list bullets under it. Do not add content I did not write."

The last line on the brain dump ask ("Do not add content I did not write") is worth adding whenever your input is sparse. The model will try to be helpful by filling gaps. Sometimes that fills in correctly. Sometimes it smooths over something that was actually unresolved. You want the structure; you do not want invented content.

The refine step

The first output is a draft, not a final. Once you have structure, it takes one more prompt to tighten it:

• "Merge the overlapping bullets in the second theme."
• "That theme is actually two different things. Split it."
• "The action items are buried. Pull them to the top as a separate section."

This is faster than trying to get a perfect structure on the first pass. Broad structure first, then one round of refinement. Per Anthropic's and MIT Sloan's prompting guidance, a second focused prompt on a specific section almost always outperforms a more complex first prompt trying to do everything at once.

How you'll know it's working

The output uses your actual content. Your names, your decisions, your open questions. If the response sounds like generic meeting-notes advice that could have come from a model that never read your input, the ask was probably too vague. Add one concrete constraint ("name the specific people every time they appear") and run it again.

You act on it within five minutes of opening the file. You send the action items to the people who own them. You forward the decision log to whoever was not in the room. You paste the open questions into the next meeting agenda. The structure reached the work.

You stop dreading the transcript. The voice memo you would have ignored for three days gets handled the same afternoon. The 900 words you recorded driving collapse into eight bullets you can actually send.

When it breaks

It smoothed over a gap. The model works only from what you gave it, but it will sometimes read across an ambiguity and pick a side without flagging that it did. Read the output against your original before you share it with anyone. Check that the decisions attributed to specific people are actually theirs. A quick scan against the source is faster than a correction email.

Your shorthand got misread. Acronyms, nicknames, and internal names can get interpreted wrong. "JB" might come back as someone's full name, or as two different people depending on where it appeared. Correct the labels on first use: "JB is Jordan Briggs, VP of Ops" added to the top of your next paste fixes it for that session.

The structure is a draft you own. The model is scaffolding. It is doing the first pass so you are not staring at raw material when you are tired. But the structure it hands back is a starting point, not a finished artifact. You edit it, you own it, you send it. That is the right relationship with the output.

Level up

Once this habit is running, the natural constraint becomes context. You paste the same intro every time: who the people are, what your shorthand means, what the meeting was about. That is friction you should not carry.

Claude Projects let you load your standing context once: your team's names and roles, your standard shorthand, the recurring meeting types you take notes in. Every conversation inside that project already knows the setup. You paste the raw notes; you skip the intro.

From there, the intermediate move is tying the output to your actual goals. A structured list of action items is useful. A structured list of action items mapped to your quarterly priorities and labeled by who should own each one is how you get from raw dump to a prioritized week. That workflow needs a Project with your goals loaded. This one does not. Start here.