The AI Handbook

The AI Handbook

I Wasted 2 Hours a Week Training AI Until I found This Fix

Your AI forget everything because you're missing these 5 key ingredients.

Ryan Stax's avatar
Ryan Stax
Dec 02, 2025
∙ Paid
A spy operating an old computer in a large server room

I opened Claude for the 47th time that week.

Started typing my question.

Then stopped.

Because I realized I was about to explain my business. Again.

Who I serve. What I sell. How I write. My entire business model.

For the 47th time.

That’s when it hit me.

I wasn’t using AI. I was fighting it.


The Five Levels of Context

Every AI you use has read trillions of documents.

It has the equivalent of multiple business degrees.

The problem? It knows nothing about YOU.

Your business. Your customers. Your voice. Your processes.

So when you ask it for a marketing strategy, it guesses.

And the guess is always generic garbage.

Here’s what most people miss: there are five levels of context, and most people only use one.

Level 1: Training Data What the AI learned from billions of documents (you can’t control this)

Level 2: System Prompt Hidden instructions from the AI company (you can’t edit this, but you can exploit it)

Level 3: User Preferences How you want AI to communicate with you (set once, applies everywhere)

Level 4: Project Knowledge Your actual business context loaded into specific workspaces (this is the game-changer)

Level 5: Your Prompt What you type in the chat box (10x more powerful when layers 1-4 are set up)

Let me show you what each level does.


Level 1: Training Data (The Baseline)

This is everything the AI learned during training.

Trillions of documents. Billions of conversations. Thousands of textbooks.

It knows business concepts, marketing frameworks, writing styles.

But it knows NOTHING about your specific business.

Watch what happens when you only use Level 1:

Prompt: “Give me a marketing strategy for our flagship product.”

Result: Generic nonsense.

It guesses your audience (and gets it wrong). It suggests SEO (which you don’t use). It recommends partnerships (which you’re not doing). It makes up everything because it has zero context.

This is what 90% of people are getting from AI.

Expensive generic advice they could find on any blog.


Level 2: System Prompt (The Hidden Layer)

Every AI has a 120-page document controlling its responses.

You didn’t write it. The AI company did.

You can’t edit it.

But you CAN exploit it.

Claude has seven trigger words:

If you use these words in your prompts, Claude does deeper analysis automatically:

  • Comprehensive

  • In-depth

  • Thorough

  • Detailed

  • Extensive

  • Complete

  • Strategic

Watch the difference:

Same prompt as before, but add two words:

“Give me an in-depth comprehensive marketing strategy for our flagship product.”

Suddenly:

  • It analyzes market size

  • It evaluates competitive landscape

  • It thinks about positioning strategy

  • It goes deeper on implications

Same chat. Same AI. Zero new context.

Just exploited the system prompt.

The result:

10x better output just by using trigger words that force deeper thinking.

This is Level 2.

Most people never learn this exists.


Level 3: User Preferences (How AI Talks to You)

This is the first level YOU control.

Tell the AI once how you want it to communicate.

It remembers forever across every conversation.

Where to set this:

Claude: Click your initials → Personal Preferences ChatGPT: Settings → Custom Instructions

What to tell it:

How you want information formatted Level of detail you prefer What you never want to see How conservative or aggressive to be

Example User Preferences:

Answer in bullet points and sub-bullet points.
Be conservative in your thinking and estimations.
Tell me your level of certainty for each claim.

Save that once.

Now every chat automatically:

  • Uses bullet format (no more walls of text)

  • Gives conservative estimates (no more pie-in-the-sky BS)

  • Shows confidence levels (so you know what’s solid vs. speculative)

You never ask for these things again.

Set it once. Get it everywhere.


Level 4: Project Knowledge (The Game-Changer)

This is the most underappreciated, underutilized feature in all of AI.

And it’s the most transformative.

What it does:

You upload your actual business documents into a Project.

Every conversation in that Project automatically has access to everything.

Your products. Your pricing. Your customers. Your processes. Your metrics.

The AI becomes a fully onboarded team member instead of an intern on day one.

Example:

Same prompt: “Give me a marketing strategy for our flagship product.”

But this time, I’m in a Project with these documents uploaded:

  • Product details and positioning

  • Customer testimonials and applications

  • Previous cohort data and feedback

  • Marketing strategy doc with messaging pillars

  • Content calendar and distribution channels

Now watch what happens:

The AI knows:

  • The exact name of my product (Second Brain Enterprise)

  • We’re in Cohort 2

  • We serve businesses with 10-50 employees

  • We work with digital agencies and professional services

  • Previous participant names and their testimonials

  • We publish via email newsletter and LinkedIn

  • We use Luma for event signups

  • We have a direct sales team

  • Our specific competitors and positioning

I didn’t type any of that in my prompt.

It already knows because it’s in the Project Knowledge.

How to set this up:

Click “Projects” in your AI (Claude, ChatGPT, etc.)

Create a new project for a specific work area.

Upload relevant documents:

  • Strategy docs

  • Customer data

  • Product information

  • Past performance reports

  • Team processes

  • Brand guidelines

Every conversation in this Project has instant access to everything.

No more explaining context. No more pasting background info. No more generic outputs.

The AI knows your business.


Level 5: Your Actual Prompt (Maximum Power)

Here’s where everything comes together.

When you have Levels 1-4 set up, your prompts become 10x more powerful.

Watch this:

I ask Claude to help me create a better prompt:

“Help me create a new prompt that will generate a comprehensive in-depth marketing strategy for cohort 3 of Second Brain Enterprise which will take place in October 2025. Do an analysis of all the complaints as well as the wins and benefits that past participants have reported. Include assignments and OKRs for every single member of the team and propose at least 10 ideas of YouTube videos that we could create with a direct connection to our persona and the benefits we offer.”

Claude generates the perfect prompt because it has all the context.

Then I use that prompt in a new chat.

The result:

A complete marketing strategy that would take me 2-3 WEEKS to create manually.

Done in minutes.

With:

  • Specific program details

  • Exact differentiators

  • Real testimonials as proof points

  • YouTube video ideas that actually make sense

  • Case study templates

  • Email sequences

  • Sales enablement materials

  • Discovery questions

  • ROI framework

  • Objection handlers

  • Team assignments and OKRs

This is the power of context engineering.

Not prompt engineering.

Context engineering.


The Payoff

Without context (most people):

Every conversation starts from zero. You explain your business repeatedly. You get generic outputs. You spend hours editing AI responses. You treat AI like an expensive search engine.

With context (operators):

AI knows your business completely. You ask simple questions and get specific answers. Outputs match your business reality. One-sentence prompts outperform essays. You treat AI like your most valuable team member.

The difference isn’t the AI model.

It’s not the prompt.

It’s the context.


What Happens Next

Most people will read this and do nothing.

They’ll bookmark it. They’ll think “I should set this up.” They never will.

Because knowing what to do isn’t the same as doing it.

The difference between people who use AI and people who BUILD with AI?

They have a system to set this up.

They don’t figure it out alone. They follow a roadmap.

That’s what paid subscribers get below.

The exact implementation system. Step-by-step instructions for each level. Copy-paste prompts that walk you through setup. Testing protocols to verify it works.

Everything you need to stop explaining your business to AI ever again.

Free subscribers get the concept. Paid subscribers get the implementation roadmap.

$15/month gets you this plus every operator system I build.

Founding rate locked forever.

[Upgrade to Paid →]

Already a paid subscriber? Keep reading.


User's avatar

Continue reading this post for free, courtesy of Ryan Stax.

Or purchase a paid subscription.
© 2026 Ryan Stax · Privacy ∙ Terms ∙ Collection notice
Start your SubstackGet the app
Substack is the home for great culture