The Craft of Prompt Engineering: Getting the Best from AI

A prompt is an instruction

A prompt is simply the words you give an AI to tell it what you want. The AI does not read your mind. It reads your text, then predicts a helpful continuation. So the quality of what comes back depends almost entirely on the quality of what you put in.

Think of it like talking to a very fast, very well-read new coworker who has never met you, knows nothing about your project, and will take you completely literally. If your request is vague, you get a vague answer. If your request is clear and specific, you usually get something useful on the first try.

Vague in, vague out

Compare these two requests. The first is what most people type. The second is what an experienced user types.

The strong prompts win because they say who it is for, how long it should be, what tone to use, and what to avoid. None of that is technical. It is just being specific.

The core habit

Before you hit send, ask yourself: have I said what I want, who it is for, how long it should be, and what 'good' looks like? If you can answer those, you are already prompting well. Everything else in this course is just sharpening that one habit.

Give it the background

The AI only knows what you tell it in the conversation. Context is the background it needs to do the job: who you are, who the audience is, the goal, and any facts it cannot guess. Pasting the relevant document, the rules, or the prior draft directly into the prompt almost always beats describing it from memory.

A good pattern is: state the situation, paste the material, then ask the question. For example: 'I run a small bakery. Here is our current menu (pasted below). Suggest three new seasonal items that fit our style.'

Show, don't just tell: few-shot examples

One of the most powerful tricks is giving examples of what you want. Showing the AI two or three sample inputs and the matching outputs you would have written is called few-shot prompting. The AI copies the pattern. This is far more reliable than describing the pattern in words.

For instance, if you want product names rewritten in a punchy style, show it: 'Plain: Blue cotton t-shirt → Punchy: The Everyday Blue Tee.' After two or three of those, it will match your style on the rest.

Set a role

You can also tell the AI who to be. A role or system prompt frames its perspective: 'You are a patient math tutor for a 10-year-old' produces very different output than 'You are a strict academic editor.' The role shapes vocabulary, tone, and what it pays attention to.

Combine all three and the AI has what it needs to behave like a true expert collaborator rather than a generic search box.

Ask for the thinking

For anything involving logic, math, or multiple steps, tell the AI to work step by step before giving its final answer. This is called chain-of-thought prompting. When the model writes out its reasoning, it makes fewer mistakes, and you can spot where it went wrong.

A simple phrase like 'Think through this step by step, then give your final answer' often turns a wrong answer into a right one. For word problems, planning, or comparisons, it is one of the highest-value additions you can make.

Give it boundaries

Constraints tell the AI what it must and must not do. Length limits, banned words, required facts, tone rules, reading level: all of these narrow the output toward what you actually want. Positive constraints ('use simple words a beginner understands') tend to work better than negative ones, but both help.

Specify the format

If you need the answer in a particular shape, say so exactly. Ask for a numbered list, a table, JSON, a single sentence, or a markdown heading. AI is good at following format instructions, but only if you give them. 'Return your answer as a table with columns Name, Pros, Cons' will get you a clean table every time.

Reasoning, constraints, and format work together: the reasoning improves accuracy, the constraints keep it on task, and the format makes the result easy to use.

Treat it as a conversation

You rarely get the perfect result on the first try, and that is fine. The real skill is iterating: read what came back, tell the AI what to change, and let it revise. 'Good, but make it shorter and warmer' is a perfectly valid next message. Each round narrows in on what you want.

Save prompts that work well. Once you find wording that reliably produces good output, reuse it as a template and just swap in the new details.

Common failure modes

AI fails in predictable ways. Knowing the patterns lets you fix them quickly.

Always verify

The most important rule of all: check the output. AI is a fast, fluent assistant, not an oracle. For facts, names, numbers, dates, and anything that matters, confirm against a trustworthy source before you rely on it. Used this way, iteration plus verification gives you both speed and accuracy.

The tool shapes the prompt

The same core skills apply everywhere, but each kind of AI tool rewards a slightly different style of prompt. Knowing the differences saves you a lot of frustration.

Chat assistants

For everyday chat (writing, summarizing, explaining, brainstorming), lead with context and a clear goal, then iterate in the conversation. Paste the material you are working with. Ask follow-up questions freely; the assistant remembers the thread.

Coding tools

For coding assistants, be precise about the language, the framework, the inputs and outputs, and any constraints. Paste the actual error message and the relevant code. Ask it to explain its plan before writing, and to handle edge cases. Then run the code and feed back what happened; debugging is a loop.

Image generators

For image tools, describe the subject, the style, the composition, the lighting, and the mood, roughly in that order. 'A watercolor painting of a quiet mountain village at dawn, soft pastel light, wide landscape view' beats 'a village.' Generate, then refine: change one thing at a time so you can see what each word does.

One caution that crosses all tools: when working with cultural, historical, or sacred subjects, AI can confidently get details wrong. Verify before you share anything that represents a community or tradition.

A simple recipe

You now have all the pieces. Here is a reliable order to assemble them into one strong prompt. You will not need every part every time, but this checklist covers the vast majority of tasks.

A worked example

'You are a friendly community newsletter editor. Our gurdwara is hosting a free langar and blood drive on Saturday. Write a short announcement for our members. Keep it warm and welcoming, under 120 words, no jargon. End with the date, time, and a one-line call to action.' Notice how it quietly uses role, context, task, constraints, and format all at once.

Keep practicing

Prompting is a craft you improve by doing. Start with a clear request, read the result honestly, refine, and verify the facts. Save the prompts that work. Over a few weeks these moves become automatic, and you will get noticeably better results than someone typing one-line questions into the same tools.

The goal is not clever tricks. It is clear thinking, written down. If you can explain what you want to a smart human, you can prompt an AI well.