⚡ Activation

💭 in your head · nothing to type

Picture a mage reciting an incantation — but swapping one word. Not the meaning, just a single word in the right slot. And suddenly: the wrong potion brews, the door stays shut, the dragon doesn't sleep.

Run it in your head: which AI answer frustrates you most — too generic, confident-but-wrong, correct-but-unexplained, or just watery? Hold it. Now picture it as a misfired incantation — where exactly in the prompt is the wrong word?

👁 Reflection

💭 in your head · nothing to type

Now shift the frame. Picture a piano tuner with a tuning fork, listening to a string that rings just a hair off the true pitch.

Hold that sound. At the collision of "incantation" and "instrument tuning" — what do you hear differently? Which note is AI missing — and what in the prompt needs to be turned until it rings true?

💡 Concept

Phrases that often help

Problem Phrase Effect
Generic answer Think step by step For models without built-in reasoning; modern reasoning models do this themselves, and the phrase can get in the way
Too abstract Give a concrete example Validates the abstraction against a real case
Hallucinations Only answer if you're sure Can cut down on inventing facts, but doesn't rule it out — still check the facts
Unclear logic Explain your reasoning Shows the thinking — easy to spot mistakes
Unstructured output Use the format: [table/list/JSON] Sets the response structure
Vague task Be specific Reduces generalizations

Example: with and without magic

Without phrases:

Prompt: "How should I plan a weekend trip?"
Result: Generic answer, 50% useful info

With phrases:

Prompt: "How should I plan a weekend trip?
Think step by step, explain your reasoning.
Give a concrete example for an e-commerce platform.
Only answer if you're confident in the analysis."
Result: Structured answer with logic, 95% useful info

The percentages are notional: they show the difference, they weren't measured.

What does NOT work