Prompt Engineering Basics, Explained From Zero

“Prompt engineering” sounds like a specialised technical skill, but at its heart it's something far more ordinary: communicating clearly with a system that can't read your mind. This guide starts from nothing — you only need to have once typed a question into an AI — and builds up the small set of habits that separate a frustrating answer from a genuinely useful one.

1 · A prompt is just what you ask

A prompt is simply the text you give an AI model: your question, your request, your instruction. There's nothing mystical about it. But here's what trips everyone up at first — the model cannot see what's in your head. It doesn't know what you already know, what you really meant, or what a good answer looks like to you. All it has is the exact words you typed. So the entire quality of the answer is decided by the quality of that text. Prompt engineering is just the skill of writing that text well, and it's far more about clarity than cleverness.

2 · Vague in, vague out

The most common mistake is a prompt that's too vague, and it fails predictably. Ask a model to “write about dogs” and it has to guess a hundred things you didn't say: for whom, how long, what tone, a poem or a paragraph or a fact sheet. Faced with all those unknowns, it does the only safe thing — it produces something generic that technically matches your words but almost certainly isn't what you wanted. This is the rule to internalise first: the model fills every gap you leave, and it usually fills it with something average. A weak answer is very often not a weak model — it's a prompt that left too much unsaid.

3 · Be specific

The single highest-value habit is specificity. Instead of “write about dogs,” say exactly what you want: a 100-word, friendly introduction to dog vaccines, for first-time owners. Notice what that packs in — the length, the tone, the precise topic, and the audience. Each was a gap the model would otherwise have guessed at. You're not being bossy; you're being clear, and models reward clarity enormously. A useful test: read your prompt back and ask, “could this be answered in a way I'd hate?” Every time the answer is yes, you've found a gap to close. Specific prompts feel like more work, but they replace three rounds of “no, not like that” with one good answer.

4 · Give it context

The second habit is giving context — the background the model needs but couldn't know. A model has broad general knowledge, but it knows nothing about your situation: your project, your customer, the thing you're actually working on. If a good answer depends on any of that, put it in the prompt. Writing to an unhappy customer? Include what they're unhappy about. Debugging code? Paste the code and the error. Think of the model as a brilliant new colleague on their first day — enormously capable, but with zero knowledge of your world until you brief them. The context you provide is that briefing.

5 · Show the format you want

The third habit is describing the shape of the answer, not just its content. Need a bulleted list? Say so. Want a table, a short paragraph, a numbered guide, a specific number of items? Say that too. And if your code will read the answer, ask for a precise structure. Without this, the model picks a format for you — often not the one you needed — and you reformat by hand or ask again. Stating the format up front is nearly free and saves the whole round-trip.

6 · Set the role

A fourth habit gives a lot of control for little effort: tell the model who to be. Open with “you are a patient tutor explaining to a beginner,” or “you are a senior lawyer reviewing a contract,” or “you are a terse assistant that answers in one sentence.” That single line quietly sets the tone, depth, vocabulary, and assumptions of everything that follows. The same question answered “as a tutor for a ten-year-old” versus “as an expert briefing other experts” produces two completely different, both-useful answers. Setting a role is one of the cheapest, most powerful moves in prompting — and it hints at a deeper idea: models are steered by layered instructions, of which the role is the most immediate.

7 · Iterate: the first try is a draft

The last habit is a mindset: treat the first answer as a draft, not a verdict. Even a great prompt rarely nails it first time, and that's normal — prompting is a conversation. Too long? “Make it half the length.” Too formal? “Warmer, please.” Missed the point? Tell it what it missed. Each correction builds on the last, because the model sees the whole exchange. This is far faster than agonising over a perfect prompt up front, and it's how experienced people actually work — they steer in a few quick turns. Don't judge a model by its first reply; judge it by where you can get in three or four rounds of gentle, specific correction.

8 · A simple test you can run this week

9 · Glossary — every term, spelled out

Prompt

The text you give an AI model — the entire instruction it works from.

Prompt engineering

The skill of writing prompts that reliably get the answer you want; mostly clear communication, not tricks.

Specificity

Stating length, tone, audience, and format so the model doesn't have to guess.

Context

The background the model needs but couldn't know — your situation, data, and constraints.

Role

Telling the model who to be (“a patient tutor”), which sets tone, depth, and vocabulary in one line.

Iteration

Refining an answer through a few quick corrections rather than restarting from scratch.