What is Prompting?
Prompting is how you talk to an AI. It's the text you type into ChatGPT, Claude, or any LLM. But prompting isn't just asking questions - it's giving instructions, providing context, setting expectations, and telling the model how to behave. Good prompting is the difference between a vague answer and exactly what you need.
Same request, three different prompt qualities. The output quality tracks the input quality.
Every detail you add removes one ambiguity the model has to guess about.
A prompt is everything you send to an LLM
When you type into ChatGPT, you're only seeing part of the picture. A prompt has three layers, and the model reads all of them before predicting what comes next.
Instructions that set the model's behavior, sent once at the start. "You are a helpful coding assistant. Always explain your reasoning." The personality layer the model follows all conversation long.
What you actually type. The question, the request, the document you want summarized. This changes every time you hit send.
Previous messages in the chat. They get re-sent with every new message so the model remembers what you already talked about.
The model reads all three layers, combines them, and predicts what word should come next. Every word of every layer matters.
The model knows nothing about you unless you tell it
An LLM has no context about you, your company, or your goals unless you provide it. A vague prompt gets a generic answer because the model plays it safe. A specific prompt gets a tailored answer because you removed the ambiguity.
Be specific
"Write a 500-word email" not "Write something." Length, format, audience - every constraint narrows the possible outputs.
Provide examples
"Use a tone like this: [example]." One sample is worth a paragraph of description. The model learns the pattern instantly.
Set constraints
"Do not mention competitors. Keep it under 200 words." Negative constraints are as useful as positive ones.
System prompts are the most underused tool in prompting
System prompts are the "personality layer." They're instructions set once at the start of a conversation that the model follows throughout. User prompts are what you type each time.
"You are a patient math tutor. Explain step by step. Never give just the answer."
Set once. Follows every interaction. Prevents you from having to repeat instructions in every single message.
"How do I solve 3x + 7 = 22?"
Changes every time. The system prompt handles the standing instructions so each user message can stay focused on the question.
Good system prompts pay for themselves in saved typing. Put your standing instructions there and stop repeating yourself.
Five things beginners get wrong, and how to fix them
Multiple questions in one prompt
The model answers the first or the easiest one and ignores the rest. Split them into separate messages.
Assuming the model knows
It doesn't know your company, your goals, or your preferences. If you don't say it, the model has to guess.
Skipping examples
One good example is worth a paragraph of description. Show the model what you want, don't just describe it.
Over-prompting
A three-page system prompt full of contradictions confuses the model. Keep it focused on what actually matters.
Expecting clarifying questions
Most models default to answering, not asking. If you want follow-up questions, tell the model explicitly.
Foundation techniques you can start using today
These techniques are the foundation of prompt engineering. They don't require special tools or APIs - just smarter phrasing in the prompt you're already writing.
Give two or three examples of what you want, then ask for a new one. The model picks up the pattern from your examples.
Add "Think step by step" to get better reasoning on math, logic, and multi-step problems. Simple but surprisingly effective.
"You are a senior software engineer reviewing this code for security vulnerabilities." The role shapes the model's focus and depth.
"Return your answer as JSON with fields: summary, sentiment, key_points." Tell the model the format and it follows.
"A prompt is a work order. The quality of the output starts with the quality of the instructions."
Be specific. The model can't read your mind. Every detail you leave out is a detail it has to guess.
Use system prompts for standing instructions. Stop repeating yourself in every message.