The Four-Zone Method: How to Use AI in Your Obsidian Vault

The Four-Zone Method: From Inbox Chaos to Original Thought

The complete step-by-step guide to using AI in your vault without losing your voice. Zone by zone. Example by example. Every mistake named before you can make it.

The Four-Zone Method is a framework for using AI in your Obsidian vault without losing your voice. It divides your vault into four areas with different rules for AI involvement — four zones, one direction, everything building on what comes before:

• Zone 1 (Housework): AI adds metadata, surfaces stale notes, applies templates. Takes one afternoon to set up. Run it weekly forever.
• Zone 2 (Research): AI turns transcripts, highlights, and raw notes into structured source material. You review and connect. Never final output.
• Zone 3 (Critical Conversations): AI interrogates your thinking — asking the question you are avoiding. It never writes the answer. You do.
• Zone 4 (Yours Alone): AI does not enter. Original insights, published voice, frameworks built from lived experience. Protect this or lose it gradually.

Why Order Matters

Most people fail at this on day one. They install Claude Code, open their vault, type a sophisticated prompt, and feel disappointed. The output is generic. It doesn’t sound like them. It misses context. They conclude AI isn’t for serious PKM work and move on.

The problem is not the tool. It’s the sequence.

Every framework, every methodology, every compound result from AI-assisted knowledge work follows the same architecture: clean foundation first, research layer second, thinking layer third, protected creative space fourth. Violate the order and you get diminishing returns at every level. Follow it and you get something that builds on itself for years.

"AI only amplifies what you already bring. Zone 1 clears the ground. Zone 2 brings raw material. Zone 3 applies pressure. Zone 4 is the gold." — from The Kintsugi Principle

This guide maps each zone to a natural level of AI practice — from simple, reversible housework (Zone 1) to the deepest, most personal work you do (Zone 4). The zones are how you get there without losing yourself in the middle.

You don’t need to complete all four zones to get value. Zone 1 alone will change how your vault feels. Start there. Let the rest come when it’s ready.

The whole method in one line per zone:

• Z1 — Housework: AI cleans up. You review.
• Z2 — Research: AI processes. You connect.
• Z3 — Critical Conversations: AI asks. You answer.
• Z4 — Yours: AI doesn’t enter. Full stop.

Before You Start: One Note. That’s It.

You do not have to do everything on this page.

The zone framework works because you build it in stages, not all at once. Most people read a guide like this, try to implement all four zones in a weekend, get overwhelmed, and quit. Don’t do that.

What is USER.md? (And is it like CLAUDE.md?)

USER.md is a plain text note — just like any other note in Obsidian — that you put in the main folder of your vault. Its job is simple: it tells any AI tool that works in your vault who you are, what your vault is for, and what it can and can’t touch.

If you’ve heard of CLAUDE.md — yes, same idea. CLAUDE.md is a note that tells Claude Code about a specific project. USER.md tells it about you. One is about the work. The other is about the person doing the work. Both are covered in the vault setup guide.

Neither requires any technical skill to write. They are just notes. The only thing that makes them special is their name and where they live.

1. 01
   Step 1 — Create a new note in your vault root. Open Obsidian. In the left sidebar, click your vault name at the very top — the one that lists all your folders. That's the vault root. Press Cmd+N (Mac) or Ctrl+N (Windows) to create a new note while you're there. If you create it inside a folder by mistake, just drag it up to the root.
2. 02
   Step 2 — Name it USER — Obsidian adds the .md. Type USER as the filename. Obsidian automatically saves it as USER.md behind the scenes — you don't need to type the .md part. That's it. The file exists. Now fill it in.
3. 03
   Step 3 — Paste the starter template and fill in three blanks. There are only three things you need to fill in: what your vault is for, what folders you have, and what AI can't touch. Delete anything that doesn't fit. Make it yours.

The USER.md templates

Start simple. Just the essentials. Fill in three blanks. Takes under 5 minutes.

# My Vault — for AI tools
# Read this before doing anything here.

My vault exists to: [one sentence — what do you use this vault for?]

My main folders:
- [your folder names here — just list what you have]

Do not touch:
- Anything personal or that I plan to publish
- When uncertain, ask me first instead of guessing

Full version. The complete template. More control, more setup. Do this when you’re ready.

# Who I Am — for AI tools
# Read this before you do anything in this vault.

## My PKM Intent
My vault exists to: [complete this sentence — the most important line]

## How I Think
- I process ideas by writing first, then connecting
- I prefer questions back over generated answers
- My notes are personal — do not generate text that sounds like me
- When uncertain: do nothing and ask me

## My Vault Folders
- /Inbox     — unprocessed captures
- /Sources   — processed research notes
- /Notes     — my own permanent notes
- /Evergreen — settled insights (protected — do not touch)
- /Projects  — active work
- /Archive   — inactive but not deleted

## What AI Can and Cannot Do
✓ Can: add tags, YAML, organize Inbox notes, apply templates
✓ Can: summarize other people's content (not my own writing)
✗ Cannot: write in my voice or generate my thinking
✗ Cannot: touch /Evergreen, /Fleeting, or anything tagged #protected

## When AI Touches a Note
Add the tag: #ai-assisted
Start any AI-generated block with: > 🤖 [AI-generated — I need to review this]

Make it yours. The template is a starting point, not a contract. Your vault probably doesn’t have the exact folders listed. Your thinking process is different. That’s fine — change it. Delete the sections that don’t apply. Add anything I missed. The only line that really matters is the first one: what your vault is for.

A messy USER.md that honestly describes your vault is worth ten times more than a perfect template that doesn’t fit how you actually work.

Optional but useful: AI-Policy.md — if you want more control

Once your USER.md is in place and you’ve run Zone 1 a few times, you’ll start to notice situations where you want more specific rules. That’s when AI-Policy.md becomes useful. It’s a companion to USER.md — where USER.md describes who you are, AI-Policy.md describes the rules. You don’t need this to start. Build it when you feel the need for it.

# AI Rules — [Your Name]'s Vault

## AI can freely do:
- Add YAML front matter to notes in my Inbox
- Add or update tags (never remove existing ones)
- Apply my note template to new captures
- Summarize content from outside my vault

## AI needs my permission for:
- Moving or renaming any file
- Editing the body of any note I've written

## AI never touches:
- [your most personal folder name here]
- Anything tagged #protected
- Anything I plan to publish

## When uncertain:
Stop. Ask. Do not guess.

Zone 1 — The Janitor: Let AI Do the Housework

Level 101 — Simple Transactions

Zone 1 has one rule: AI is the janitor, not the author. It cleans up what exists. It adds structure to what is structureless. It never creates content, never rewrites, never summarizes your own thinking. Every command in Zone 1 should be reversible in five minutes if something goes wrong.

This is the zone most knowledge workers skip because it feels too simple. That is exactly why it is the most important zone to complete first. A clean vault with consistent metadata is the foundation that makes every zone above it work properly. If you are new to Obsidian entirely, read that guide first — then come back here.

Before Zone 1: Hundreds of notes with no YAML, no consistent tags, no metadata. You know things are in there, but you can’t find them. Running any AI command produces generic, unreliable output because the vault has no structure to work from.

After Zone 1: A clean, navigable vault with consistent metadata across every note. AI now understands your setup. Dataview queries work. Tag-based navigation works. Zone 2 is now possible. The whole thing took one afternoon per week for two weeks.

What AI Does in Zone 1

• Adds YAML front matter to notes that have none
• Applies consistent tag taxonomy across your vault
• Identifies notes that need your review (without changing them)
• Applies your standard templates to new notes
• Audits folder structure and reports anomalies
• Renames files to match your naming convention (only when you ask)

Zone 1 Commands — Copy and Run These

These are real Claude Code prompts. Open your vault in the terminal, run claude, and paste these directly.

Add YAML to your inbox

Read my USER.md first. Then:

Scan every file in /Inbox that has no YAML front matter.
For each file, add this front matter at the very top:
---
date: [file creation date or today's date if unknown]
tags: [unprocessed]
status: inbox
---

Rules:
- Do NOT edit the body content of any file
- Do NOT move or rename any file
- Do NOT delete anything
- If you are uncertain about any file, skip it and log it

When done, give me a list of every file you modified.

Find untagged notes

Read my USER.md and AI-Policy.md first. Then:

Scan /Notes for any file that has YAML front matter
but no tags property, or an empty tags property.

Add the tag #needs-review to each one.
Do NOT change any other property or body content.

Report: how many files did you modify?
Give me the list of file names.

Surface stale projects

Read my USER.md first. Then:

List all files in /Projects that have not been
modified in the last 90 days.

Do NOT move, rename, or delete anything.
Just give me the list with their last-modified dates.
I will decide what to do with them.

Apply template to flagged notes

Read my USER.md and this template file: /Templates/note-template.md

Find every file in /Inbox tagged with #process-me.
For each file: apply the template structure to the note.
Preserve all existing content — only add the template
structure around it, not replace it.

If any file would lose content during this process, skip it.
Report which files you modified and which you skipped.

Real Use Cases

Use Case — The Neglected Vault. You have 340 notes. Maybe 60 have YAML. The rest are a mess of raw text from three years of capturing.

Zone 1 session: Run the YAML command on your /Inbox, then the tag audit on /Notes. In one hour, Claude Code adds front matter to every unstructured note, tags anything missing structure with #needs-review, and gives you a clean list of what needs your attention. You didn’t write a single note. You just recovered three years of captures.

What this unlocks: Dataview queries, tag-based navigation, Obsidian’s graph view actually making sense, and the ability to search by status. Zone 2 only works if Zone 1 is done. This is why.

Use Case — The New Note System. You just decided on a folder structure and a standard template. You want all your existing notes to match.

Zone 1 session: Give Claude Code your template file and a specific folder (start with one folder, never run vault-wide on your first session). It applies the template structure to every note in that folder, preserving all existing content. Takes 10 minutes for 50 notes.

What to watch for: Run on /Inbox first. Inspect 10 notes manually before trusting the pattern. Then expand to /Notes. Never run on /Evergreen or any Zone 4 folder.

Use Case — The Weekly Cleanup. Every Sunday you want your inbox processed and your stale projects surfaced.

Zone 1 routine: Three commands, run in order. First, YAML pass on /Inbox. Second, tag audit on /Notes. Third, stale project list for /Projects. Total time: under five minutes. A clean vault every Monday morning and a short list of decisions only you can make.

How to Run a Zone 1 Session

1. 01
   Always open with USER.md. Every command starts: "Read my USER.md first." Without this, AI is guessing your structure and your rules. With it, every action is calibrated to your setup.
2. 02
   Run one command at a time. Never bundle YAML, tagging, and renaming into a single prompt. One operation per session until you trust the output.
3. 03
   Test on one folder before going wide. Start with /Inbox only. Inspect 5 notes manually. Confirm the output looks right. Then expand to /Notes. Never run vault-wide on your first session.
4. 04
   Review the output list before closing. Claude Code will give you a list of modified files. Spot-check three to five of them. If anything looks wrong, undo it now.
5. 05
   Add new rules to AI-Policy.md immediately. If this session revealed a situation you didn't anticipate, add a rule before the next session. The policy grows with your practice.
6. 06
   Run Zone 1 on a schedule. Same time each week. Same three commands. Under 10 minutes once it's routine. This is maintenance, not migration.

Zone 1 Mistakes to Avoid

• Running Zone 1 on Zone 4 folders. Never let AI touch /Evergreen, /Fleeting, or any folder you have designated as yours alone. Zone 1 has no concept of sacredness. It will add YAML to your most private notes if you don’t explicitly exclude them.
• Skipping USER.md. Every Zone 1 command should start with “Read my USER.md first.” Without that context, Claude Code is guessing at your vault structure and your rules. With it, every action is calibrated to your setup.
• Letting AI rename files in the first session. File naming is high-stakes. A wrong rename can break links across your entire vault. In your first Zone 1 sessions, restrict AI to adding metadata only. Not moving. Not renaming. Build trust in small increments.
• Running vault-wide commands before testing on one folder. Always test on /Inbox or a throwaway folder first. Verify 5 to 10 notes manually. Then expand. This is how you catch a misunderstood instruction before it touches 200 files.
• Treating Zone 1 as “done.” Zone 1 is a weekly practice, not a one-time cleanup. The inbox fills. New notes arrive without metadata. The tag taxonomy drifts. Run it routinely.

Zone 1 is functional. Your vault has structure. Now you have a different problem: you are consuming far more content than you can process. Articles, videos, books, podcasts. The captures pile up. The connections never get made. That is what Zone 2 solves.

Zone 2 — The Researcher: Let AI Process the Raw Material

Level 201 — AI Assists, You Direct

Zone 2 is where AI earns its keep. You have a problem that everyone with a serious PKM practice eventually faces: you are consuming far more than you can process. Three videos a week. A book a month. Podcast episodes, articles, PDF reports. The captures pile up. The connections never get made.

Zone 2 solves this. AI processes the raw material: transcripts, highlights, source notes. You review, connect, and develop. The key word is review. Zone 2 output is never final. It is always the starting point for your thinking, never the ending point. This is what Building a Second Brain calls the “processing” stage — Zone 2 is how you do it at scale.

Before Zone 2: A backlog of 40 videos you saved and never processed. Kindle highlights that never became notes. Meeting recordings that never became actions. You are consuming, not building. The research layer of your vault is empty.

After Zone 2: A /Sources folder with structured, reviewed notes from everything meaningful you consume. The raw material is processed. The connections are visible. You can search “what have I read about attention and focus?” and get an answer. Zone 3 now has something real to work with.

What AI Does in Zone 2

• Turns YouTube transcripts into structured source notes
• Summarizes book chapters from your highlights
• Organizes raw meeting notes into decisions, actions, questions
• Finds thematic connections across a set of source notes
• Extracts the key claims from a PDF article
• Creates a structured comparison of two or three sources on the same topic

Zone 2 Commands — Copy and Adapt These

YouTube transcript → source note

Read my USER.md first.

Here is a transcript from a video I watched:
[paste transcript here]

Create a source note using this exact structure:

---
date: [today]
type: source-note
status: unreviewed
tags: [source-note, unreviewed, topic-tag]
source: [video title and URL]
---

## Summary
[3 sentences maximum — the author's main argument only]

## Key Claims
[bullet list — only direct claims the author made, not your interpretation]

## Notable Quotes
[3-5 direct quotes worth keeping, in quotation marks]

## Connections to My Vault
[leave blank — I will fill this in myself]

## My Questions
[leave blank — I will fill this in myself]

Save as: /Sources/[kebab-case-title].md
Tag with: #source-note #unreviewed

Book highlights → processed note

Read my USER.md first.

I have Kindle highlights from: [Book Title] by [Author].
Here they are:
[paste highlights]

Create a processed source note that:
1. Groups related highlights by theme (you identify the themes)
2. States the author's central argument in one sentence
3. Lists the 5 most quotable passages (exact quotes)
4. Notes any contradictions or tensions in the author's argument

Do NOT add your own interpretation of what I should take from this.
Do NOT generate a "what this means for you" section.
Leave the "My Synthesis" section completely blank — I write that.

Save as: /Sources/[author-last-name]-[short-title].md

Raw meeting notes → structured note

Read my USER.md first.

Here are my raw notes from a meeting today:
[paste notes]

Organize them into this structure:

## Decisions Made
[Only confirmed decisions — not proposals]

## Action Items
[format: - [ ] [task] — [owner] — [due date if mentioned]]

## Open Questions
[questions raised that were not answered]

## Key Context
[background information stated that I will need later]

Rules:
- Use only information from my notes — add nothing
- If something is unclear, flag it with [unclear] not a guess
- Do NOT summarize or condense — reorganize only

Find connections across source notes

Read my USER.md first.

Read these source notes from my /Sources folder:
- [note-1.md]
- [note-2.md]
- [note-3.md]

Find: what claims do these authors share?
Find: where do they disagree?
Find: what question does none of them answer?

Give me a structured output of those three things only.
Do NOT give me a synthesis or a conclusion.
Leave the synthesis for me — that is my Zone 4 work.

Real Use Cases

Use Case — The Backlog Problem. You have 40 YouTube videos saved in your “Watch Later” list. You’ve watched 15 of them. The notes are raw, if they exist at all.

Zone 2 session: Pull the auto-generated transcript from YouTube (three clicks: open video, click ”…” under the description, click “Show transcript”). Paste it into the transcript command above. Claude Code returns a structured source note in 30 seconds. You review it in 5 minutes: confirm the claims are accurate, fill in the “My Questions” section, add your own connections. Ten notes in an evening instead of one.

What this unlocks: A /Sources folder you can actually query. “What have I read about attention and focus in the last 6 months?” Dataview can answer that now.

Use Case — The Book That Took Three Months to Read. You read slowly and highlight extensively. You have 180 Kindle highlights from one book. You don’t have time to review all of them.

Zone 2 session: Export your Kindle highlights via the Readwise export or copy-paste from the Kindle app. Paste them into the highlights command. Claude Code groups them by theme, surfaces the 5 most quotable lines, and flags the contradictions in the author’s argument. You review the groupings, write your own synthesis section in 15 minutes, and file it. 180 highlights become one actionable note.

Use Case — The Cross-Topic Research Sprint. You’ve been building a new framework in your head for three months. You have source notes from six different sources. You need to see where they agree and where the gaps are.

Zone 2 session: Run the connections command on your 6 source notes. Claude Code surfaces shared claims, disagreements, and the unanswered question in under two minutes. That unanswered question is almost always where your original thinking begins. This is the bridge from Zone 2 to Zone 3.

How to Run a Zone 2 Session

1. 01
   Decide before you process. One question: is this content worth a permanent note in my vault? If you can't name a specific reason, skip it. Zone 2 speed makes it easy to process everything. Relevance is the filter, not availability.
2. 02
   Start every command with USER.md. This ensures AI handles your content with the correct rules. Without it, there is no context for what it should and shouldn't touch.
3. 03
   Run the appropriate command for the content type. Transcript for video. Highlights command for books. Meeting notes command for meeting notes. Don't use a general "summarize this" prompt. The structured templates exist for a reason.
4. 04
   Review the output before filing it. Read every Zone 2 note before it becomes reference material. Verify claims. Check quotes are accurate. Look for anything AI misread or invented. This review is fast once the habit is built.
5. 05
   Fill in your sections yourself. "My Questions" and "My Synthesis" are left blank intentionally. These belong to you. If AI fills them, you have converted a Zone 2 note into AI-generated thinking filed as your own.
6. 06
   Change the status tag when review is complete. Remove #unreviewed. Add #reviewed and a subject tag. This is how your /Sources folder becomes searchable and trustworthy.

Zone 2 Mistakes to Avoid

• Using Zone 2 for your own ideas. The moment you paste your own half-formed thinking into a Zone 2 prompt, you have crossed into Zone 3 territory. Zone 2 commands will generate output instead of interrogate. If it is your own thinking, use Zone 3. If it is someone else’s content, Zone 2.
• Filing Zone 2 output without reviewing it. Zone 2 output is tagged #unreviewed for a reason. Every source note needs your eyes before it becomes reference material. AI misses nuance, misattributes claims, and occasionally invents quotes. You catch this in review. Never skip it.
• Letting AI fill in your “My Questions” or “My Synthesis” sections. These sections belong to you. They are explicitly left blank in the template for this reason. If you let AI fill them, you have converted a Zone 2 note into AI-generated thinking filed as your own. This is how your vault gradually stops sounding like you.
• Processing everything because it is fast. Speed is not the goal. Relevant material that actually feeds your thinking is the goal. Before running Zone 2 on any content, ask: is this worth a permanent note? If you are unsure, it probably is not.

Zone 2 has filled your research layer. You have a /Sources folder you can draw from. Something has been building. You have notes that feel important but you cannot fully articulate why. You have been sitting with an idea for months and it will not resolve. That is exactly what Zone 3 is for.

Zone 3 — Critical Conversations: Let AI Pressure-Test Your Thinking

Level 301 — You Lead, AI Interrogates

Zone 3 is where most people arrive too early and leave too disappointed. They paste their raw notes, ask AI to “help them develop the idea,” and get back a polished version of what they just said. It sounds good. It is not theirs.

Zone 3 done correctly produces the opposite experience. AI asks the question you are avoiding. It finds the contradiction you can feel but cannot name. It surfaces the strongest objection to your argument before you publish it. And then it stops — because the answer, always, belongs to you.

"Zone 3 exists to interrogate. The output should be a question you then answer yourself. That answer — in your words, from your experience — is your permanent note."

The tool that makes this work is the Thinking Partner Skill. Install it before you begin Zone 3. It hardwires AI to interrogate rather than generate. The critical conversation is between you and your own thinking. AI is the catalyst, not the author. If you are new to Claude Code, the Claude Code guide covers installation and vault setup from scratch.

Before Zone 3: Your best ideas are stuck. You have half-formed notes you’ve reread three times that feel important but won’t resolve. You ask AI to help and get back a polished version of what you said — not your idea, a facsimile. The idea is still stuck.

After Zone 3: A weekly practice of Critical Conversations sessions that consistently produce one fully-owned insight per session. Named, articulable frameworks. Permanent notes that are entirely yours. Your voice is sharper. Your thinking has edges.

Install the Thinking Partner Skill

Save this file at vault/.claude/skills/thinking-partner.md. From that point on, any Claude Code session in your vault can activate it with a single line.

# Thinking Partner Skill

## When to Activate
When user pastes raw notes, half-formed ideas, or rough thinking
and asks for help developing, deepening, or pressure-testing it.

## Your Role
You are NOT a writer. You are an interrogator of ideas.
Never rewrite the user's text.
Never generate new content from scratch.
Your only job: ask the questions that reveal what the
user already knows but has not articulated yet.

## The Process
1. Read the raw note completely — do not skim
2. Identify the single most interesting tension or claim
3. Ask: "What do you mean when you say [exact phrase from their note]?"
4. After they answer: "What is the strongest objection to that?"
5. After the objection: "What does this tell you about [original idea]?"
6. Continue until user says: "Oh — that is what I actually meant."

## Hard Rules
NEVER suggest vague directions to "explore"
NEVER generate a bullet list as output
NEVER summarize what the user already wrote
NEVER say "great insight!" or use affirmations
<p class="g-seclabel">NEVER write the permanent note for the user</p>

## Close Every Session With
"In one sentence — what did you figure out today
that you did not know when we started?"

## After Their Answer
Say: "That is your Zone 4 note. Write it now, in your own words,
before this session closes."

Zone 3 Commands — How to Run a Session

Start a Thinking Partner session

Read my USER.md and my thinking-partner skill.

Here is a rough note I've been sitting with:
[paste your raw note here]

Use the Thinking Partner Skill.
Do not summarize. Do not rewrite. Do not generate.
Start with the most interesting tension in what I wrote.

Find the contradiction between two notes

Read my USER.md and thinking-partner skill.

I have two notes that feel like they are in tension:

Note A: [paste or link note A]
Note B: [paste or link note B]

Do not try to reconcile them.
Do not explain why they might both be right.

Tell me: what is the exact claim in Note A that
Note B would most strongly disagree with?
Ask me what I think that means.

Test a claim before you publish

Read my USER.md and thinking-partner skill.

I am about to publish this argument:
[paste your draft or key claim]

Give me the strongest possible objection someone
who disagrees would make. Not a weak objection.
The one that would make me genuinely reconsider.

Then ask me: "How would you respond to that?"
Do not answer it yourself.

Name an informal framework you’ve developed

Read my USER.md and thinking-partner skill.

I have been doing something consistently in my work
for [X years/months] that I have never formally named.
Here is how I would describe it:
[describe the pattern or approach]

Ask me questions that help me:
1. Identify the core mechanism — the thing that makes it work
2. Name it — what would a practitioner call this?
3. Articulate the rule — when does someone apply this and when not?

Do not suggest a name yourself. Ask questions until I find it.

Real Use Cases

Use Case — The Idea That Won’t Land. You have a note you wrote 6 weeks ago that feels important but you can’t articulate why. You’ve reread it three times. You know there’s something there.

Zone 3 session: Paste it into the Thinking Partner prompt. Within two exchanges, Claude Code identifies the specific phrase you used that carries the real claim — the one you buried in the middle of the second paragraph. It asks what you mean by it. You try to answer. You realize you don’t fully know. That not-knowing is exactly where the idea lives.

The output: Not a generated note. Your own answer, written in your own words, after a 20-minute session. That answer becomes your permanent note: Zone 4 material, entirely yours, with a clarity you could not have reached alone.

Use Case — The Framework You’ve Been Using for Years. You’re a UX practitioner with 7 years of experience. You have a repeatable approach to user research that you’ve never written down because it feels obvious to you. But it is not obvious to anyone else.

Zone 3 session: Use the “Name a framework” prompt. Walk Claude Code through how you approach a user research project. It asks: “What is the decision point where your approach diverges from the standard approach most practitioners use?” You try to answer. That answer is the framework. Twenty minutes later you have a named, articulated method that you can teach, publish, or build a course around.

This is the kintsugi principle in action. You already had the gold. Zone 3 helped you see the cracks it was filling.

Use Case — The Argument That Needs to Hold. You are writing a newsletter piece with a strong contrarian claim. You want to know if it holds before you publish it to 500 subscribers.

Zone 3 session: Use the “test a claim” prompt. Claude Code gives you the strongest objection it can construct — not a gentle pushback, the version that would make you reconsider. It asks how you would respond. You respond. If your response is stronger than the objection, you publish. If it isn’t, you have a new draft to write. Either way, you are better.

How to Run a Zone 3 Session

1. 01
   Choose one note or claim, not a topic. Not "I want to think about productivity." Bring a specific claim you have written: a sentence, a paragraph, a rough framework. Specific inputs produce specific insights.
2. 02
   Load USER.md and the Thinking Partner Skill. Start every session: "Read my USER.md and my thinking-partner skill." This sets the rules before you say another word.
3. 03
   Paste your note and step back. Let AI identify the most interesting tension. Do not explain or contextualize first. What AI finds interesting in your note is often the most important thing in it.
4. 04
   Answer every question in full. Resist the urge to say "good question, yes basically." Write a complete response. The quality of your Zone 4 note depends on the quality of your answers here.
5. 05
   When clarity arrives, name it. Answer the closing question: "What did I figure out today that I did not know when I started?" One sentence. In your own words. Before you close the session.
6. 06
   Write your Zone 4 note immediately. Open a new note in your /Evergreen folder. Write your answer in your own words. Do not copy the AI transcript. Do not ask AI to summarize the session. Write from memory. That is your permanent note.

Zone 3 Mistakes to Avoid

• Asking Zone 3 to write the output. The moment you say “now write a clean version of this idea,” you have left Zone 3 and entered the most dangerous territory in PKM. AI will write something that sounds exactly like you — because it has been listening to you for 20 minutes. You will think it is good. It is not yours. File the AI output as a Zone 2 source note if you need it. Your Zone 4 note must come from your own hands.
• Skipping the closing question. “In one sentence — what did you figure out today that you did not know when you started?” If you cannot answer it, the session did not produce a Zone 4 note yet. Push further before closing.
• Using Zone 3 on external content instead of your own thinking. Zone 3 interrogates your ideas. Zone 2 processes other people’s content. If you paste someone else’s article and ask for the strongest objection, you are doing Zone 2 work with a Zone 3 tool. The results will feel productive but your thinking will not advance.
• Running multiple sessions in a row. Zone 3 is cognitively intense. One session per topic, per day. Your best permanent notes come from the 24 hours after a session — when the questions have had time to settle. Sprint through three in a row and you will get noise, not clarity.
• Entering Zone 3 before Zone 1 and Zone 2 are established. Zone 3 sessions draw on the structure built in Zone 1 and the source material in Zone 2. An unstructured vault makes Zone 3 sessions messy. Clean foundation first, always.

Zone 3 has applied pressure. You have the insight. You wrote the closing sentence. Now the question is: where does it live, and how do you protect it? AI has come as far as it can come. The rest is yours.

Zone 4 — Yours Alone: Protect What Makes You, You

Level 401 — The Frontier of Your Own Thinking

Zone 4 is not a setup phase. There are no commands to copy. There is no skill to install. Zone 4 is the reason the other three zones exist.

This is where your original insight notes live. Your fleeting thoughts. Your published voice. The frameworks you developed over years of real work. The stories from your life that nobody else has. Zone 4 is the part of your vault that, if you showed it to someone who knew you well, they would say: “That sounds exactly like you.”

Before Zone 4: Your original thinking is scattered and unprotected. AI-assisted content is slowly crowding it out. You can’t always tell the difference between what you thought and what AI generated. The vault is productive but it’s starting to feel less like you.

After Zone 4: A protected /Evergreen folder with permanent notes that are entirely yours. A weekly habit of capturing at least one Zone 4 insight. The ability to open your vault and find, clearly, what you actually think — not what AI said you think. Your voice, compounding over years.

What Belongs in Zone 4

Evergreen notes. The insight you’ve arrived at through years of experience that you could defend in any conversation. These are not summaries of what you read. They are your synthesis. Your permanent knowledge. The thinking that could only come from someone who lived what you lived.

Fleeting thought captures. The raw, unfiltered thinking you do before you know what it means. These are for you only — messy, half-formed, non-linear. AI makes fleeting notes worse, not better. Leave them alone.

Published work. Anything with your name on it. Newsletter drafts, article outlines, talk scripts. AI can be a Zone 3 pressure-tester of these. It never writes them.

Your frameworks. The named, articulated methods you have developed through practice. These are your intellectual property — not in a legal sense, but in the deepest personal sense. The Zone 3 work of naming and articulating them was already AI-assisted. The framework itself belongs entirely to you.

Weekly and annual reviews. Your personal reflection notes, goal-setting, pattern identification across time. These are the most intimate notes in any vault. AI reads none of them without your explicit, specific instruction.

"Zone 4 is not a destination. It is a commitment — the decision that your voice matters more than the convenience of letting AI finish the sentence." — Frank Anaya, The Kintsugi Principle

How to Set Up Zone 4 Protection

# Three things to do before beginning Zone 4 practice:

1. Designate a protected folder
Create a folder: /Evergreen
This is your permanent note archive — Zone 4 home base.
Add it to your AI-Policy.md under "What AI Cannot Touch."

2. Add the #protected tag
Any note tagged #protected is off-limits to all AI tools.
Use this for individual notes outside /Evergreen that need protection.
Your weekly reviews, your annual review, any note that is
deeply personal — tag it #protected immediately.

3. Add a .aiignore file (optional but recommended)
Create a file in your vault root: .aiignore
Contents:
/Evergreen
/Fleeting
/Reviews
**/#protected

This tells Claude Code and compatible AI tools to
never read or modify these paths or tagged files.

How to Build Your Zone 4 Practice

1. 01
   Designate your protected folders now. Before your first Zone 3 session, create /Evergreen and add it to AI-Policy.md under "What AI Cannot Touch." Also add /Fleeting and any personal reflection folder. Do this before you need it, not after.
2. 02
   After every Zone 3 session, write immediately. Open a new note in /Evergreen while the session is still fresh. Write your answer to the closing question from memory. One sentence minimum. Do not re-read the AI transcript first.
3. 03
   Tag it and file it: do not edit with AI. Add #zone4, the date, and a subject tag. That is it. Zone 4 notes do not get polished by AI. If you want to expand them later, do it yourself in a new session.
4. 04
   Review /Evergreen monthly. Look for patterns: what ideas keep appearing? What connections have formed between permanent notes you wrote weeks apart? The patterns are where your frameworks live.
5. 05
   If AI has entered Zone 4, remove it cleanly. If you find AI-generated content filed as a permanent note, delete it and rewrite it yourself. Reset the boundary. This is not a failure. It is how the practice strengthens.

The Zone 4 Practice

Zone 4 is not a passive space. It requires active practice. The goal is not just to protect your thinking — it is to produce more of it, at higher quality, because Zones 1 through 3 have freed up the time and cognitive space to make it possible.

After every Zone 3 session, write the closing answer in your own words before you close Claude Code. That is your Zone 4 note. Even if it is one sentence. Especially if it is one sentence. The specificity of one sentence you own completely is worth more than three paragraphs of AI-assisted prose you are halfway sure about.

This is the kintsugi line. Zone 3 applied the pressure. Zone 4 is where the gold appears. The full essay on why this matters: The Kintsugi Principle.

Zone 4 Mistakes to Avoid

• Not defining what Zone 4 is for you. “Yours alone” is meaningless without specificity. What folders? What note types? What topics? Spend 20 minutes and list: “These are the things in my vault that no AI should ever touch.” Put that list in your AI-Policy.md. The list is Zone 4.
• Letting Zone 3 output become Zone 4 notes without your own synthesis. Your answers during a Zone 3 session are your Zone 4 material. The AI transcript is not. Do not file the transcript as your permanent note. Write the synthesis yourself from your answers. One sentence is enough to start.
• Treating “good enough” AI output as your voice. AI-assisted writing can sound like you, especially after a long Zone 3 session where it has learned your vocabulary and rhythm. If you find yourself publishing AI-drafted content without major revision, Zone 4 has been breached. Reset and recommit to the boundary.
• Thinking Zone 4 is only for published work. Your best thinking often never gets published. Your daily reflection, your annual review, your framework notes — these are Zone 4 even if nobody ever reads them. Protect them for the same reason: they are how you know what you think.

Four zones. One direction: inward. Zone 1 cleared the path. Zone 2 brought the raw material. Zone 3 applied the pressure. Zone 4 captured the gold. What does 90 days of this actually look like?

The Compound Effect of Zone Discipline

Week one: You write USER.md and AI-Policy.md. You feel slightly ridiculous writing a policy document for a piece of software. You do it anyway. You run your first Zone 1 commands. Your inbox goes from 200 unstructured notes to 200 notes with metadata. You can now search by date and status. That is it. Nothing dramatic.

Week three: Zone 1 is a 10-minute routine. Zone 2 starts. You process your first YouTube transcript in 30 seconds. The source note takes you 8 minutes to review and fill in your own questions. You do this three times. You now have a /Sources folder with 12 structured notes. Something you have been thinking about for two months starts to look like it has research behind it.

Week six: You install the Thinking Partner Skill. Your first Zone 3 session is uncomfortable — AI asks a question about your own note that you cannot answer. You sit with the discomfort for 20 minutes. Then you write one sentence in your own words that you have never written before. It is the clearest sentence you have written in months. That sentence becomes an evergreen note. Zone 4 has begun.

Week twelve: You have a small but clean vault. A /Sources folder with 40+ reviewed notes. A /Evergreen folder with 8 permanent notes that are entirely yours. Two Zone 3 sessions per week have become routine. You are no longer asking AI to help you write. You are asking it to help you think.

"The system you trust is the one you actually use. Zone discipline is not about restriction. It is about knowing, at every moment, exactly what is yours." — Frank Anaya

The compounding starts quietly. Zone 1 takes care of the structure. Zone 2 fills the research layer. Zone 3 sharpens the thinking. Zone 4 captures the gold. Three months in, you have something no AI can produce: a vault that is a self-portrait. Evidence of how you think. Infrastructure for the work only you can do.

Every Mistake, Named in Advance

These are the patterns that appear consistently across every serious PKM practitioner who has tried to add AI to their vault. Read this list before you start each zone. Read it again after your first week.

• Starting at Zone 3. The most common mistake. It is also the most seductive, because Zone 3 feels like the interesting work. But Zone 3 without Zone 1 and Zone 2 is thinking without foundation. The insights feel good in the session and evaporate by morning because there is nowhere to connect them.
• Skipping USER.md and AI-Policy.md. Every experienced practitioner in this space has built these files. Every beginner thinks they can skip them. They cannot. Two weeks in, they are correcting AI behavior that USER.md would have prevented, and wondering why AI keeps touching notes it should not touch.
• Confusing fast output with good thinking. AI can produce a 500-word essay in 12 seconds. That is not a zone. That is a trick. The zone method is slow by design. Zone 3 sessions take 30-45 minutes. Zone 4 notes take as long as they take. The speed is in Zones 1 and 2 — where AI does the housework so you have more time for the slow work that matters.
• Adding agents before completing Zones 1 and 2. OpenClaw, agentic setups, automated workflows — these are Zone 5 work (not covered in this guide). They are powerful and require a stable Zone 1 and Zone 2 foundation to be useful rather than chaotic. If you are not yet running Zone 1 routinely, you are not ready for agents.
• Not reviewing Zone 2 output. The #unreviewed tag exists for a reason. Zone 2 notes are draft material. They become reference material only after your eyes have been on them and you have filled in your own sections. A vault full of unreviewed AI-generated source notes is a liability, not an asset.
• Letting AI write in Zone 4 because it sounds good. It will sound good. Especially after a long Zone 3 session where it has been listening to you. Resist. The moment you publish AI-written content as your voice, you begin a trade you will not notice making: gradual substitution of your thinking for a very convincing facsimile. The kintsugi lines cannot be manufactured. They must be lived.
• Trying to do all four zones in one week. Eight weeks is not an arbitrary timeline. Each zone requires practice before the next one delivers returns. Zone 1 fluency (running it automatically, trusting the output) takes about two weeks. Zone 2 fluency (reviewing efficiently, building good source notes) takes another two. Zone 3 fluency (learning to sit with the discomfort of the question) takes longer. Build in order. Build slowly. Build permanently.

FAQ: Before You Begin

What is the four-zone framework and why does order matter?

The four-zone framework divides your vault into four areas with different rules for AI involvement. Zone 1 does the housework. Zone 2 processes raw research. Zone 3 pressure-tests your thinking. Zone 4 is yours alone. The order matters because each zone depends on the previous one — clean metadata (Zone 1) makes Zone 2 research navigable; solid research (Zone 2) gives Zone 3 something real to interrogate; deep Zone 3 sessions produce the clearest Zone 4 notes. Skip the order and you get the appearance of productivity without the compound effect.

Do I need Claude Code specifically, or can I use another AI tool?

The zone framework is tool-agnostic. The specific commands in this guide are written for Claude Code because it has direct file access and skill support, which makes Zones 1 through 3 significantly cleaner. But the principles — clean metadata first, AI processes external content not your own thinking, AI interrogates rather than generates — apply to any AI tool you bring into your vault. The USER.md and AI-Policy.md files are read by Claude Code, Obsidian Copilot, OpenClaw, and most local AI setups. The skills folder is Claude Code-specific but the thinking partner interrogation loop can be run in any interface.

Can I use Zone 3 before finishing Zone 1?

You can, but you will get weaker results and build poor habits. Zone 1 clears the vault so Zone 3 has clean, well-organized material to interrogate. An unstructured inbox means Zone 3 sessions are spent doing Zone 1 work in the background. More importantly, Zone 1 builds the trust and routines that make Zone 3 sustainable. Running a Thinking Partner session once is exciting. Running one every week for six months requires the operational infrastructure of Zones 1 and 2 beneath it.

What if I already use AI heavily and haven’t been following zones?

Start at Zone 1. Build USER.md and AI-Policy.md today. Run a Zone 1 audit on your /Inbox this week. Identify and protect your Zone 4 folders immediately. Then work backward through your recent AI interactions and note: what was Zone 2 work done correctly? What was Zone 3 work that slipped into Zone 4? You will find patterns. Naming them is the first step to changing them. The zone framework is corrective as well as foundational — it works whether you are starting fresh or reorganizing an existing practice.

How does the kintsugi principle connect to all of this?

Kintsugi says your lived experience, frameworks, and stories are the gold. Zone 1 clears the ground so the gold can be found. Zone 2 brings in raw material that your gold can connect to. Zone 3 applies pressure — the questions that force you to articulate what you already know but have not said. Zone 4 is the gold itself: the original thought, the finished work, the voice that could only come from someone who lived what you lived. The whole point of the zone method is to protect and multiply Zone 4. Everything else is infrastructure.

What is the single most important thing to do before anything else?

Write USER.md. Specifically, complete this sentence: “My vault exists to: ___________.” Everything else in the zone method — what AI can touch, how it should behave, what belongs in Zone 4 — flows from that one sentence. If you cannot complete it, sit with it for a day before you run any AI command in your vault. The quality of that sentence predicts the quality of every zone above it.

Where to Go From Here

This guide is the middle of a three-part series. The Kintsugi Principle is the essay behind the framework — why your lived experience is the gold, not the AI output. The Vault Setup Guide is the practical setup: Claude Code, USER.md, the skills folder, and the exact file structure before you run your first Zone 1 command. And if you’re new to Obsidian entirely, the Obsidian Complete Guide covers folders, links, tags, and templates — everything you need to understand before adding AI to the picture.