⚠️ The trap of trusting the first output
When you start using AI, you review everything: you read every word, check every claim. But the results get good, you relax—and fall into the trap. There’s a joke that sums up the risk: one person uses AI to turn a bullet point into a long, structured email; on the other end, the team uses AI to turn that email back into a bullet point. Everyone transforming, no one reading.
🧠 The central idea
The better the AI, the more dangerous it is to let your guard down. The standard is what keeps “good enough” from becoming “no one really read it.”
- •At first, you review everything; over time, you trust too much.
- •The first version is rarely the best—just the fastest.
- •Using AI to write is great; accepting its output without reading it is the mistake.
— The "em dash" signal
Sometimes the sign of “this is AI” is small—like the em dash. AI fills everything with em dashes because it was trained on formal texts. If a colleague who knows you gets a text from you with five em dashes, they think: "So-and-so didn't write this; it's AI". And the moment you think this, it changes how you interpret the entire message — "did it read it? Is it all true?".
Human judgment sits between the draft and the signature—illustrative diagram.
🏆 Study the best in your field
Taste is built through reference. If you work in sales, study great sales emails. If you work in marketing, study excellent landing pages. You can only recognize "great" in AI output if you know what great looks like in your field.
✓ What to DO
- ✓Collect examples you consider excellent.
- ✓Ask “why is this good?”—clarity? confidence?
- ✓Compare the AI output against this benchmark.
✗ What NOT to do
- ✗Copy and paste what you liked without understanding why.
- ✗Thinking "any text" from AI is ready to use.
- ✗Confusing formal with good.
📁 Build your reference library
Save what’s good and what sounds like you. A library of examples becomes both your benchmark for taste and material you give AI as context (Track 3). When something is great, don’t just save it—write down why it’s good and what you like about it.
Create a “references” folder
By type of deliverable: emails, proposals, posts, reports.
Write down the "why"
What makes this clear, trustworthy, and in your voice.
Use as context
Feed the AI good and bad examples so it gets your tone right.
🔄 Feed the Corrections Back In
The feedback loop is for the good e for the bad. When the AI writes something and you change five things, say: "I changed these 5 points, and here’s why". Update your instructions so it’s closer next time. That’s how you train the system to understand your taste.
💡 Practical tip
Keep a “style rules” file (in the project/CLAUDE.md/GPT instructions). Turn every recurring correction into a rule. Within weeks, AI will deliver in your tone by default.
📝 Example of structured feedback
✍️ Your name is on the work
In the end, AI generates the work — your taste decides what deserves your name. If you produce something with AI, your signature is on it. If it’s great, you get the credit. If it’s bad, you get the blame. It doesn’t matter who wrote it: your name is on it.
🖋️ The subscription rule
Before sending anything made with AI, ask: "Would I feel comfortable putting my name on this?" If you hesitate, the criteria haven’t finished doing their job.
Self-recovery (optional): what role does “taste” play in the age of AI?
✅ Module Summary
Next Module:
2.4 — Protection & Multiple Income Streams