Why uploading more content into AI makes it worse, not better

What we're learning from building AI clones in our first cohort.

This is a 🏴‍☠️ Founding Members–Only 🏴‍☠️ post. Founding Members get access to the Pirate Eddie Bot to ask category design questions, weekly actionable insights, the full library with 30+ audiobooks 2000+ mini-books, and more. See the Founders Deck here.

---

Dear Friend, Subscriber, and Category Pirate,

We just kicked off our first-ever AI Clone Cohort—a small group of pirates building AI versions of themselves using the frameworks we’ve been developing with our own Pirate Eddie and Pirate Christopher bots.

We’re learning a ton. And we want to share what we’re figuring out in real time.

And the first thing we’ve learned?

Almost everyone is using AI wrong.

Not “wrong” as in they’re bad at technology. Wrong as in they’re obsessing over the wrong lever—and it’s costing them the one thing category designers can’t afford to waste.

Their differentiation.

Here’s what we mean:

We know a guy—smart guy, successful—who spent six months uploading documents into his ChatGPT. Hundreds of files. Training data. Transcripts. Frameworks. He was convinced this was why his AI was producing great results.

“You have to train it,” he told us. “Upload everything. The more context you give it, the better it gets.”

We think he may have been wrong.

AI wasn’t getting better because of his uploads. The underlying model was getting better the entire time he was “training” it. He attributed the improvement to his effort—when really, OpenAI was just shipping better software every few weeks.

This is the most common mistake people make with AI. They obsess over the context window and ignore the prompt. The context window is your content. The prompt is your thinking.

If you dump all your content into an AI with a generic prompt, you get generic output that sounds like everyone else.

You’ve just commoditized your own Intellectual Capital.

A category designer doesn’t use AI the way everyone else does—because category designers don’t think the way everyone else does.

The prompt is where your Point Of View lives.

It’s where your frameworks, your language, your different lens gets encoded into the machine.

Generic prompt = generic output = commodity.

Category-designed prompt = differentiated output = transformation.

After months of testing across Claude, ChatGPT, Gemini, Grok, and our own Pirate Eddie Bot—building AI clones, writing mini-books, designing categories, running a business—we’ve landed on something we haven’t seen anyone else say clearly.

There are four things that determine the quality of your AI output.

And almost everyone has the priority order backwards.

1. Model Quality—Which LLM are you using?

This matters more than anything else. Not all models are built the same—and they’re getting more specialized, not less.

ChatGPT has gone consumer. It’s optimized for the masses. Claude writes circles around it. Grok is built for real-time social data. Gemini is wired into Google’s ecosystem.

This is category design in action.

Each LLM is designing for a different use case—a different category of user. The question isn’t “which AI is best?” The question is “which AI is best for what I’m trying to do?“

You wouldn’t hire a tax accountant to design your brand—so why would you use a consumer-optimized chatbot to do deep strategic work?

2. Prompting—How you talk to AI.

This is the skill that holds its value. Models will change. Pricing will shift. Context windows will expand.

But your ability to give clear, structured, step-by-step instructions? That stays valuable no matter what else moves underneath you.

Here’s the difference:

A bad prompter barfs everything at AI in one giant paragraph and walks away. “Here’s everything I know. Figure it out.”

A good prompter does the same brain dump—gets everything out of their head first—but then adds the layer most people skip.

Direction.

“Here’s everything I know. Now here’s what I want you to do with it. Step by step. For this audience. In this format. In this tone. Toward this outcome.”

The brain dump is the raw material. The prompt is the blueprint. You need both. Most people only do the first part and wonder why the output feels like a B-minus book report.

Prompting is a skill. The people who are intentional about it get wildly better output than the people who aren’t.

Your prompt is how you teach AI to think like you.

Not like the internet. Not like a generic business consultant. Like the person who has spent years building Intellectual Capital in a specific category.

If you can’t articulate your thinking clearly enough to prompt an AI, you can’t articulate it clearly enough to sell it to a client, either. The prompt is a forcing function for clarity.

3. What You Pay—Free, $20, or $200?

This one’s a parabola, not a straight line.

Free is limiting. Twenty bucks a month unlocks a massive leap. But jumping to $200 doesn’t give you 10x more. It depends on your use case.

Pay enough to unlock the model you need. Don’t assume expensive means useful.

4. Context Window—What you upload.

Here’s where everyone gets it wrong.

The context window—how much information you feed AI—is also a parabola. Upload nothing, AI has no foundation. Upload everything? AI gets confused. It can’t prioritize. It doesn’t know what matters.

We learned this the hard way. We overloaded one of our bots with too much training data. The result? It got worse. It couldn’t find the signal in the noise.

Enough context is powerful. Too much is poison.

Think of it through the lens of the Four Creator Capitals. Your Intellectual Capital is your best content—your sharpest frameworks, your clearest POVs, your most battle-tested ideas. The context window should get your best Intellectual Capital, not all of it.

Curate, don’t dump.

So what does the right priority look like?

Pick the right model for the job. Learn to prompt well. Pay for what you need. And be surgical—not sloppy—with what you upload.

Now go use this.

Take the prompting principles from this post and go try them right now.

If you’re a Founding Member, you already have access to the Pirate Eddie Bot—an AI trained on 200+ Category Pirates mini-books, every framework, every POV, every piece of languaging we’ve built over the last three years.

Here’s a challenge:

Open up the Pirate Eddie Bot and prompt him using all five principles at once. Give him a role. Tell him who you are. Walk him through your problem step by step. Paste in an example of what “good” looks like for your business. Then compare that output to what you were getting before.

The difference will be obvious.

Some prompts worth trying today:

You are my category design strategist. I run [your business]. Help me identify whether I’m competing in an existing category or designing a new one. Here’s how I currently describe what I do: [paste your positioning]. Pressure test it.

I need to find my super geo. Here’s what I know about my best customers: [details]. Help me figure out where more of them are concentrated.

Here’s a piece of content I wrote. Help me write something similiar about [insert new topic]. Here’s what I currently think about it [insert your current POV]. 

The better you get at prompting, the more valuable the Pirate Eddie Bot (or any other LLM) becomes.

(Not a Founding Member yet? Founding includes unlimited access to the Pirate Eddie Bot, digital copies of all 6 of our big books, the conversation thread below, the entire Founder’s Deck, and more. Join as a Founding Member here. →)

Here’s How To Actually Become A Better Prompter

Most people treat prompting like Google search. Type a question. Hit enter. Hope for the best.

That’s not prompting. That’s wishing.

The people getting the best output from AI—the ones building businesses, writing books, designing categories with it—they treat prompting like a conversation with a very smart, very literal colleague who has zero context on your life.

Here are the principles that actually matter: