Key Takeaways
- Magic words and secret phrases matter far less than clear context.
- Longer prompts are not automatically better prompts.
- You rarely need to be polite or threatening for good results.
- Iterating on a reply beats crafting one perfect prompt.
Myth 1: There are secret magic words
People collect “power phrases” like incantations. In reality, models respond to clarity, not spells. “Act as a world-class expert” adds little; a specific description of the task, the audience and the format adds a lot. Context beats keywords every time.
Myth 2: Longer prompts are better
A wall of text often makes results worse, not better, because the important instruction gets buried. Say what you need, give the relevant background, specify the output, and stop. A tight three-sentence prompt frequently beats a rambling paragraph.
Myth 3: You must be polite (or aggressive)
Please and thank you do not meaningfully change output, and neither do threats or all-caps demands. Tone is for you, not the model. Spend that energy on being specific about what “good” looks like for this task.
Myth 4: One perfect prompt is the goal
Chasing the flawless first prompt wastes time. The fastest path to a great result is a decent prompt followed by a quick correction: “shorter”, “more concrete”, “drop the intro”. Treat it as a conversation, not a one-shot.
Myth 5: The same prompt works everywhere
Models differ. A prompt tuned for one assistant may underperform on another. Do not over-invest in portable “prompt libraries”; invest in the habit of describing tasks clearly, which transfers across every model.
The bottom line
Good prompting is not a secret language. It is clear thinking made explicit: state the task, the context, and the output you want, then refine. Drop the myths and you will get better results with less effort.