Two people can type into the exact same chatbot and walk away with completely different quality answers. Not because one has a “better” AI — because how you phrase a request genuinely changes what you get back. That gap has a name now: prompt engineering.
Remember, it’s a guessing game
From part 1: an LLM predicts the most likely next words based on everything before them. That means your prompt isn’t just a question — it’s the entire scene the model is guessing from. A vague scene gets a vague, generic guess. A specific, well-set scene gets a specific, useful one.
VAGUE “Write something about dogs.” SPECIFIC “Write a 3-sentence, upbeat Instagram caption about a rescue dog’s first day home, no hashtags.”
Same model, same second. But the second prompt narrows down the “next word” guessing so much that there’s barely room left to go generic.
The four things a good prompt sets
- Task — What exactly do you want made, fixed, or explained?
- Context — Who’s it for? What do they already know?
- Tone / format — Formal or casual? A list, a table, three sentences?
- Boundaries — Length limits, things to avoid, a strict word cap
You don’t need all four every time — but the more of the “scene” you set, the less guessing the model has to fill in on its own.
two habits worth stealing
Show, don't just tell
Give one or two examples of the exact style you want before your real request. The model matches the pattern you demonstrated.
"Here's a caption I liked: [example]. Now write one for this photo."
Ask it to think step by step
For anything with logic or math, add "explain your reasoning step by step" — it noticeably improves accuracy on harder questions.
Turns a rushed guess into a shown-your-work answer.
the bigger picture
Not a magic spellbook — just clear communication
You’ll see long lists of “prompt hacks” floating around, and a few genuinely help. But nearly all of them boil down to something you already know how to do: communicate clearly, the same way you would delegating a task to a coworker who’s brilliant but has never met you before, and can’t read your mind. That framing beats memorizing tricks every time.
where you've already used it
- Chat assistants — Better prompts = fewer follow-up corrections
- Image tools — Style, lighting, and composition words all steer the output
- Coding assistants — Specifying the language, framework, and constraints upfront