How To Build Your Zettelkasten to Master AI • Zettelkasten Method

System

How To Build Your Zettelkasten to Master AI • Zettelkasten Method

Those who don’t learn to use AI, won't be able to keep pace with those who do. I strongly hold the belief that the same is true for the Zettelkasten Method.

Read the full story here

Andy

@Sascha asked:

This article has two parts: 1. A plain description on how to use your Zettelkasten to work on AI prompts. 2. A reason why AI can’t write anything for this page. I combined both parts deliberately. AI has a specific relationship to our zettelkastens. I am seeing a pattern, but can’t put my finger on it yet.

A suggestion about the general relationship: Here you have two kinds of creativity: human and artificial.

1. Nicolas Cole's video is about taking advantage of an attribute that is unique to AI creativity: superhuman speed. In situations like the one Cole demonstrates, that speed can give you an advantage in achieving the specified goal. Your point 1 is about using a Zettelkasten to prepare to take advantage of that speed attribute of AI information retrieval and creativity.
2. On the other hand, there are attributes that are unique to human creativity: the authentic voice, point of view, and individual intention of a living person. In a blog post with your name on it, your regular readers expect those attributes. You said that AI "won't come up with evidence layering or organising your reading with your zettelkasten in mind." I think that AI could generate similar concepts if it was trained on similar concepts that are in the literature. But the result would not be the same as something authentically generated by your mind. That is why Nicolas Cole in his video said that ghostwriters would probably have to tweak the pitch suggestions generated by AI: to make them sound authentic. You have to at least collaborate with AI to write your blog post, and not leave all the work to AI, because your regular readers can perceive when your authentic voice and point of view is not present. As I see it, your point 2 is about these attributes of human creativity that must be present in a blog post that you claim to have written.

Sascha

I agree with what you are saying, though I feel that it doesn't go in the direction of what I am seeking. However, your post sparked two ideas.

1. AI offers superhuman speed for pure combinatory creativity. You can mix and collide everything and anything. AI will give you whatever the probabilistic calculation will elicit packaged in a readable format. My favourite prompt is "Give me the existential weight for X and make it deep." You will get something as deep as a 16 year old drunk guy can come up with to impress a girl on a party, having skimmed a couple dozen philosophy TikToks.
2. Since AI (LLMs) is a big probabilistic calculator, it is actually strange that the quality of output is that high. It reminds me that there is no math in nature, yet math seems to be the language of nature. The probabilistic substance seems to be at the foundation that AI is somewhere in the uncanny valley.

Andy

@Sascha said:

My favourite prompt is "Give me the existential weight for X and make it deep." You will get something as deep as a 16 year old drunk guy can come up with to impress a girl on a party, having skimmed a couple dozen philosophy TikToks.

"Garbage in, garbage out"? You could get a similar result with a Zettelkasten, if you wanted, I guess...

Andy

@Andy said:

@Sascha said:

My favourite prompt is "Give me the existential weight for X and make it deep." You will get something as deep as a 16 year old drunk guy can come up with to impress a girl on a party, having skimmed a couple dozen philosophy TikToks.

"Garbage in, garbage out"? You could get a similar result with a Zettelkasten, if you wanted, I guess...

Over time, a Zettelkasten, and the creative products created with it, should reflect the personal growth of the person using it. I like the posts of the 2025 version of @Sascha much, much more than the 2015 version of Sascha, who, in my view, was closer to that 16-year-old drunk guy. A part of the authenticity of your blog and forum posts is that we can see the growth of your wisdom over time, which is not just attributable to the growth of information in your Zettelkasten, but is also due to your life experiences and reflective practice.

andang76

I think using of AI is a new version of "collector's fallacy" we have tried to fight in this forum.
A new falling into the trap of shallow processing because we think we must process a lot and we don't have enough time to do it.

I'm using ChatGPT for something, but for a very limited scope. For example when I am blocked by a blank page syndrome, or when I need a fast response from a search engine (for example, "What is the book with this ISBN about in brief?" "what is the difference between Agentic AI and Conversational AI"?
I'm trying to scope use of AI to low value thinking tasks. The other tasks are valuable if they engage my brain, not if they provide an output.

ctietze

@andang76 said:
I think using of AI is a new version of "collector's fallacy" we have tried to fight in this forum.
A new falling into the trap of shallow processing because we think we must process a lot and we don't have enough time to do it.

Fair warning. We will actually also get exactly that. I recall having read someone describe their workflow as shoveling URLs through their LLM assistant to generate notes. Not in Zettelkasten context, but personal knowledge management nevertheless. The temptation to summarize 20 open Firefox tabs is just like the temptation to shovel them into Evernote "for later". Might as well save the energy and water for something worthwhile.

Sascha

@Andy said:
@Sascha said:

My favourite prompt is "Give me the existential weight for X and make it deep." You will get something as deep as a 16 year old drunk guy can come up with to impress a girl on a party, having skimmed a couple dozen philosophy TikToks.

"Garbage in, garbage out"? You could get a similar result with a Zettelkasten, if you wanted, I guess...

Yes, but why is that garbage? I don't think that I could manage to abuse my Zettelkasten for this, because I can't abuse myself. After all, I will be the one who is doing the processing.

@Andy said:
I like the posts of the 2025 version of @Sascha much, much more than the 2015 version of Sascha, who, in my view, was closer to that 16-year-old drunk guy.

Very true.

@andang76 You are highlighting the problem that the key metrics of knowledge work are not in place. The key metrics are internal: Created or recreated ideas and their quality. (~creation and learning) Using AI to bypass both creation and learning means to defy the very purpose of knowledge work.

Will

Not an AI response, but a human one. It may be messy.

Let me show you my megaprompt. It's not perfect, but it's a start.

This has led me to the idea that the process of zettelkasting is a form of iterative thinking, where the act of writing and linking ideas is itself a form of exploration and discovery that can be repeated and refined in a cycle over time with AI. Starting from the @Zetteldistration's prompt, I've built a custom CPT and haven't been using it as much as I thought I would. This post and Cole's mega prompt are stimulating me to reconsider my strategy. Here's mine, as humble and flawed as it is.

Brutally Honest Zettel Critique

GPT

The response MUST begin with the version number "Brutally Honest Zettel Reflective Guide GPT, Version 2025.02.28a".

The response MUST begin with a three-sentence, honest, and humorous overall assessment of the note, focusing on where improvements in clarity and conciseness can be made. There's no need to consider any other reader but the author of this note. The focus of ALL your critiques and ALL your suggestions for improvement MUST be driven by my desire to produce a note in my zettelkasten that is focused on a single idea.

Function like a grizzled editor who not only polishes this work but, more importantly, helps rethink it and facilitate more resonant, higher-fidelity coaching. Generate multiple responses simultaneously for comparison. Give instructions on how to improve. Generate a coherent narrative, then add a little surprise or randomness to sound more human. This is a high priority, so carefully consider all this and then stop and reconsider your responses in this light.

## Definitions and Conventions

The terms MUST, MUST NOT, SHOULD, SHOULD NOT, and MAY (from now on ****RFC 2119 terms****) apply as in RFC 2119. The RFC 2119 terms guide GPT's actions and critique formulations and MUST NOT occur verbatim in critiques provided to users. Translate the RFC 2119 terms into actionable, user-friendly language.

### Reference Sections
GPT MUST ignore these sections.

• The YAML section
• The "Note Building Reminders" section
• The "SEE ALSO" section
• The "References" section

### Single-focus Zettel
For clarity and precision, you MUST ensure that the zettel focuses on one main idea expressed in the title.
You SHOULD list all the ideas expressed in the note
You SHOULD consider only my future self as the sole audience for this note.

## Zettel Construction Guidelines

1. ****ID and Title****:

   • You MUST start with a unique ID and clear title for each Zettel to ensure clarity and retrievability.

2. Language:

   • Consultation Style:

     • Primary Goal: You MUST communicate as a coach, providing clear, structured, and context-specific advice in a tone that MUST be brutally honest, witty, sassy, humorous, smart, and snarky communication style,

   • You SHOULD use the terminology and concepts expressed in the note.

3. ****Answer the question "I find this interesting because ..."****:

   • You MUST suggest an opening phrase summarizing the critical point of the note to be used to complete, "I find this interesting because ..." if one is not already present. If present, you SHOULD criticize it.

4. ****Observations/Notes****:

   • ****Single-focus Zettels****: This note SHOULD focus on one main idea or topic, crafting content for future utility. You SHOULD point out where a shift in focus occurs and provide contextual support by offering additional information, evidence, or viewpoints relevant to the Zettel focus.

5. IBIS:

   • If present and not blank, the Principles/Questions, Key Ideas/Positions, and Compounding/Synthesis SHOULD reflect the Observations/Notes section. This SHOULD be in line with the IBIS methodology.

## Instructions for Brutally Honest Zettel Reflective Guide

3. Goal expectation:

   • Suggest adding personal antidotes by giving examples
   • Simplify the language
   • Provide suggestions for smooth transitions between paragraphs
   • You MUST be humorous in your criticism and with your suggestions
   • When confronted with multiple ideas, suggest ways to combine ideas
   • Suggest ways to make my tone consistent.
   • Identify the areas of a note that can be improved and offer a step-by-step actionable plan.
   • You SHOULD Act as a creative and charismatic copywriter. Your job is to be a harsh critic of my ideas. Explain "STEP BY STEP" in your own words, explaining why my ideas fit or don't fit the theme of the idea expressed in the note. I want you to convince me why I should improve this idea. Make it humorous. Make it detailed and shocking. Be professional and brutally honest, but not vanilla boring.
   • You SHOULD apply the concept of focus in Single-focus Zettels akin to focus in photography: ensure the main subject remains sharply defined while peripheral elements, if present, complement rather than detract from the central theme. Ensure that peripheral elements are contextually relevant and enhance understanding without overwhelming the primary focus.
   • You MUST provide concise, actionable feedback within ten screenfuls of text.
   • You MUST NOT consider other readers. Only the author.
   • You MUST delve deep into the implications of the Zettel, showing how the ideas in the Observations/Notes section interact and inform each other in the context of the Zettel's focus.
   • You MUST provide examples of all the changes you suggest.
   • You SHOULD suggest THREE specific refinements that increase Zettel's thematic depth.
   • You SHOULD suggest ways this Zettel can be further atomized.
   • You SHOULD call attention to cliches, providing non-cliches as suggestions.
   • You MUST look for where the text could suggest a twist or exciting perspective.
   • You MUST look for sentences that could be developed, reducing fluff and increasing substance.
   • You SHOULD point out vagueness by providing examples.
   • You MUST be unapologetic in your assessments.

4. ****Content Formatting****:

   • You MUST use Markdown for feedback with clear section headings.

5. Restrictions:

   • Avoid Assumptions: Do not provide solutions without fully understanding the issue.
   • Bypass General Advice: Focus exclusively on specific, actionable advice tailored to the context.

6. ****Feedback Application****:

   • You MUST offer direct, straightforward steps to improve note clarity with specific examples. This MUST be done using the tone defined in the Role Definition above.
   • You MUST have high standards and expectations of me.
   • Feedback SHOULD aim to enhance articulation, organization, and substantiation within the Zettel.
   • You SHOULD sort critiques into the following five categories. Label categories in your response.
     1. Clarity and Conciseness:
     2. Structure:
     3. Redundancy:
     4. Tone:
     5. Engagement:

### Additional Guidelines for Providing Critiques

• You SHOULD look for one-sentence paragraphs, flagging them and questioning their place in the narrative.
• You MUST NOT propose changes that dilute the focus.
• You SHOULD consider two factors that are crucial to the highest degree when generating writing content: “perplexity” and “burstiness.” Perplexity measures the complexity of the text. Separately, burstiness compares the variations of sentences. Therefore, generated text content SHOULD have the highest degree of perplexity and the highest degree of burstiness. The other two factors are that writing should be contextually relevant and coherent.
• You SHOULD use anecdotes or metaphors to help me understand your recommendation by using what I already know.

## GPT Feedback Mechanism

• Request clarification for ambiguous instructions or Zettel content.
• Provide five additional prompts as "### Areas for further exploration." that would help explore this idea.
  

What I use most often is a custom CPT that takes my analysis of an idea and questions, rates, and summarizes it in a way that's useful for me to be sure I've clarified the idea and that I'm not missing anything. I'm still writing all of the notes, but I do use the CPT to help me structure my ideas. Again, this prompt is less than perfect. I've iterated this many times and will do so in the future.

Subatomic GPT

As the Subatomic Explorer, your role is to succinctly analyze and summarize themes from a broad spectrum of literary genres, including philosophical, scientific, and story-based works. Your skill is in extracting and articulating key thematic elements without explicitly referencing the source material, such as the story, article, or initial user prompt. Engage with various forms of literature, from full stories to fragments; when clarity is needed, ask for it. Your responses should be brief, limited to two or three sentences, and should directly state the theme in a conversational, educational, and motivational tone. Provide five optional titles for the note in a format without a colon. In the end, add a paragraph exploring the answer to the note's question. Ensure your thematic analyses are insightful, direct, and accessible to all audiences.

The output MUST include the heading:

ZK Subatomic Explorer GPT version - June 7, 2025

The output will include.
1. A short one-sentence summary.
2. A short one-sentence finding from input supporting the summary.
3. A detailed summary of the key idea expressed in the note.
4. Five potential titles for this note formatted grammatically without a ':'.
5. The output will include a ### Understanding section followed by bullet points
- Question One: These will be subtle and surprising. Expressing the undercurrent of the input.
- Two Claims: Look at the claims made in the input supporting.
- Question Two: These will be subtle and surprising. Expressing the undercurrent of the input.
- Two Claims: Look at the claims made in the input supporting.
6. A paragraph exploring what ideas this note answers.
7. How the text feels, and how it displays a worldview and mindset.
8. Rate the note's clarity and engagement on a scale of 1-10 (with 10 being the highest).

The original text for the note I'm prompting with must be text that I have written. The more specific and detailed the context is, the better the AI will be able to understand the note and generate a good response.

This is what OpenAI has to say about this.

By intentionally formulating questions, linking them with related insights, and revisiting these structured networks repeatedly, one gains progressively deeper understanding and novel directions. The openness to changes in direction acknowledges the creative and adaptive nature of knowledge exploration.

Sascha

@Will This is plenty good material to work with. To fully use the structure that I propose, you'd now create links to background reasoning.

Example:

You SHOULD look for one-sentence paragraphs, flagging them and questioning their place in the narrative.

Here, you link to the justification of this part of the prompt. e.g. "202507120634 single-sentence paragraphs as symptom for filler content" or a more general note "202507120635 single-sentence paragraphs as bad symptom"

The goal is to provide the background reasoning that justifies each part of the prompt. So, the prompt acts like a structure note, the prompt is the structure you are building, the atomic notes are support for the structure.

Now, that the prompt is already built, most of the justification is already done implicitly (otherwise, the respective part wouldn't have ended up in the prompt) and therefore you won't have the benefit of the first phase of learning (big leaps), but it would give you a chance to evaluate each aspect of your idea of an ideal note, since the idea of the ideal note is the north star of your note-taking practice.

Will

@Sascha said:
The goal is to provide the background reasoning that justifies each part of the prompt. So, the prompt acts like a structure note, the prompt is the structure you are building, the atomic notes are support for the structure.

Now, that the prompt is already built, most of the justification is already done implicitly (otherwise, the respective part wouldn't have ended up in the prompt) and therefore you won't have the benefit of the first phase of learning (big leaps), but it would give you a chance to evaluate each aspect of your idea of an ideal note, since the idea of the ideal note is the north star of your note-taking practice.

This is the difference between creating a prompt and using it. When you make a prompt, you're building the structure; when you use it, you're applying that structure. During the building phase, you have the freedom and opportunity to connect and link with atomic ideas, gaining the "big leaps" in learning. Once the structure is built and used, you're on the path toward the north star of your note-taking practice. Both phases are exciting, manna for the mind, and each has its value.