From Voice Memos to Structured Knowledge

The problem

I record voice memos on walks. Five minutes, sometimes ten. Inside any one of them I'll touch three projects, a marketing idea, and a decision I've been chewing on for a week.

Then the file sits there. Titled by date. Buried under the next memo, and the one after that.

A week later I remember I said something useful about a project direction. I can't remember which recording. I scrub through four of them and give up.

The memos aren't the problem. The lack of any index across them is.

What's actually in there

I took ten consecutive memos from two weeks of recordings and read them as one document instead of ten.

Seven distinct threads showed up. Not seven memos — seven topics that wove through multiple recordings across multiple days:

A project idea mentioned Monday, refined Wednesday, given a deadline Friday
A pricing thought that started as a throwaway line and turned into a decision by the second week
A health goal that kept surfacing as a footnote to work planning

None of those lived in a single memo. They were spread across recordings, devices, and weeks. Each memo on its own looked like rambling. All ten read together had a clear shape.

The shape hiding in stream of consciousness

Voice memos feel chaotic but they aren't. When I talk through my day I'm doing four things in roughly the same order:

Set context — "Okay, today is Tuesday and I want to talk about..."
Explore — branching, comparing, weighing
Decide — "I think the move is to..."
Assign next steps — "So the game plan is..."

That's an outline in a hoodie.

The workflow

The whole thing takes about ten minutes for a week of memos. Four prompts, one LLM, one folder of markdown files at the end.

Step 1: Pull your recent memos

talkie memos

You get a list with titles, durations, word counts. Grab the past week or two.

Step 2: Read them as a batch, not one by one

This is the part most people skip. Don't read each memo in isolation. Read all of them together and look for threads — ideas that appear more than once or that connect to each other.

In my ten memos:

Thread	Appeared in
API redesign	2 memos
Mobile app concept	3 memos
Marketing strategy	4 memos
Product launch prep	5 memos
Pricing model	2 memos
Onboarding flow	3 memos
Health and sustainability	1 memo (but kept coming up afterwards)

Step 3: Generate topic files

Each thread becomes its own file. Not a summary — a working document with the core idea, how it evolved, open questions, and links to the other topics it touches.

Here's what one looks like:

mobile-app.md

#Mobile App

"The mobile app could be this really focused, voice-first

experience — just the core features, none of the desktop baggage."

##Why it exists

Desktop users want a different thing than mobile users. This acknowledges that.

##Open Questions

-How much of the core engine should be shared vs rebuilt?

-Is this a separate product or a companion app?

##Connects to

-Product Launch — needs to ship before the launch push

-Marketing Strategy — could be the breakout narrative

-Onboarding Flow — first-run experience matters more on mobile

Step 4: Generate daily notes as anchors

Each day's memos become a daily note that links into the topic files. This preserves the timeline so I can trace how an idea went from a throwaway line to a concrete plan.

Step 5: Open it in Obsidian

The [[wikilinks]] light up. Graph view shows which ideas cluster, which ones are orphans, and which days produced the most connections.

Two weeks of memos end up looking like this:

Knowledge Graph — 7 topics from 10 voice memos

Center hub Topic nodes Daily notesHover to explore connections

Not seven recordings. Seven ideas, tracked across time, linked to each other.

File structure

topics/
  API Redesign.md
  Mobile App.md
  Product Launch.md
  Marketing Strategy.md
  Pricing Model.md
  Onboarding Flow.md
  Health & Sustainability.md

daily/
  2026-03-15.md  →  links to 3 topics
  2026-03-14.md  →  links to 4 topics
  2026-03-10.md  →  links to 3 topics
  2026-03-09.md  →  links to 3 topics

The prompts

These are the four I use. Paste your transcripts alongside each prompt. Works in Claude, ChatGPT, whatever you have.

Prompt 1: Extract topics

The big one. Feed it all your transcripts at once.

Prompt 1 — Extract Topics

Here are my voice memo transcripts from the past two weeks. Read them all as a single stream of consciousness¹ and identify the distinct topics, themes, or threads that appear across multiple memos. For each topic: - Give it a clear name - Write a 1-2 sentence summary - Note which memos it appears in - Flag whether it's an idea, an active project, or a decision³ Don't summarize each memo individually. I want cross-cutting themes, not per-memo summaries. [paste all transcripts here]

Prompt 2: Generate topic files

Once you have a topic list, this writes the Obsidian-ready file for each one.

Prompt 2 — Generate Topic Files

For each topic below, create an Obsidian markdown file with: - YAML frontmatter: tags, status¹ (idea/active/thinking/ongoing), first_mentioned date - A section with the core idea, using a direct quote from the transcripts where possible - An "Open Questions" section with unresolved threads - A "Related" section with [[wikilinks]] to other topics from the list Keep the voice natural — these should sound like me thinking, not a report. Use my actual words from the transcripts. Topics: [paste topic list] Source transcripts: [paste transcripts]

Prompt 3: Generate daily notes

Anchors the topics to specific days.

Prompt 3 — Generate Daily Notes

For each day that has memos, create a daily note in Obsidian format with: - YAML frontmatter: date, source memo IDs, tags - Bullet-point summary of what was discussed that day - [[Wikilinks]] to the relevant topic files wherever a theme is mentioned - Direct quotes for anything particularly clear or decisive³ Group multiple memos from the same day into one daily note¹. Keep it scannable — someone should be able to read a daily note in 30 seconds and know what that day was about. Memos: [paste transcripts grouped by day] Topic files: [list your topic names]

Prompt 4: Generate the index

A master file that ties the topics together.

Prompt 4 — Generate Index

Create an INDEX.md for an Obsidian vault with: - A table of all topics with status and first-mentioned date - A timeline of daily notes with a one-line summary of each - [[Wikilinks]] to everything This is the entry point¹. Someone opening this vault for the first time should understand the full landscape in 60 seconds². Topics: [list topics] Daily notes: [list dates with summaries]

Tips

Batch. Five to ten memos at once. Single-memo summaries miss the threads that cross recordings, which is the whole point.
Don't clean the transcripts. Leave the "um"s and "uh"s. The LLM handles them, and they tend to cluster around the parts where I'm actually thinking.
Run it weekly. Sunday evening, pull the week, run the prompts, update the vault. It works as a ritual better than as a one-off.
Let topics merge. After a few weeks you'll spot topics that should fold into each other. That's a signal — the thinking is converging.

Why this works

Most knowledge systems fail because they ask you to be organized at the moment of capture. Pick a folder, write a title, add tags. That's friction, and friction kills capture.

Voice memos invert it. Capture is messy and fast. Organization happens later, in batch, by a model that's good at exactly that.

What's next

I want this to be one command — pull transcripts, run prompts, write the vault. Until then, the manual flow above takes ten minutes and gets you most of the way.