What is a prompt, and why does it matter how you write one?

A prompt is everything you give a model before it answers: the question, the context, the examples, the tone you set and the role you assign it. It is not a search query. A search engine finds documents that already exist; a prompt shapes something that does not exist yet, which is why two people asking the same question in different words get meaningfully different answers.

Why the wording changes the output

The model generates a likely continuation of the text it has been given. Change the text and you change what is likely. Ask "explain photosynthesis" and you get the encyclopedia continuation, because that phrasing usually precedes encyclopedia-shaped text. Ask "explain photosynthesis to someone who thinks plants eat soil" and you get something else entirely, because that phrasing usually precedes a correction.

You are not giving instructions to an assistant who understands you. You are setting up the conditions under which certain kinds of text become probable. Teenagers find this idea genuinely interesting once it lands, and it explains every prompting technique without them having to memorise a list.

The four moves that actually matter

Say who it is for. "For a fourteen-year-old who has already read the chapter" produces a different and better answer than nothing at all.

Give it something to work from. Pasting your own notes, your own draft, your own data narrows the space it is generating into. The single biggest quality improvement available is supplying material rather than asking from nothing.

Say what the output should look like. Length, form, whether you want the reasoning or just the answer. Ambiguity gets filled with the most common shape, which is rarely the shape you wanted.

Ask for the disagreement. "What is the strongest objection to this?" or "What would someone who disagrees say?" This is the move that turns the tool from an agreeing machine into something useful, and it is the one students almost never make on their own.

Why prompt tricks age badly

Lists of magic phrases circulate constantly, and most of them are artifacts of one model version. "Take a deep breath" and "you are a world-class expert" had measurable effects on some systems at some points and much less on others. A student who has memorised twenty incantations has learned something that expires; a student who understands that a prompt sets up a context has learned something that does not.

This is why we teach prompting as one modest skill inside AI literacy rather than as the subject itself. Prompt engineering as a job title is already thinning out. Knowing why a model responds to context will still be useful in a decade.

The part that is a safety lesson

Whatever goes into a prompt goes to whoever operates the tool, and in a consumer product it may be retained and may be used to improve the system. That means a prompt is not a private space. No names of other people, no grades, no medical detail, no passwords, nothing you would not paste into an email to a stranger.

Teach the rule with the reason, and it survives contact with adulthood. "Do not put anything in the box you would mind a stranger reading" is a sentence a fourteen-year-old will still be using at thirty.

A good five-minute exercise

Take one question. Write three prompts for it: bare, with an audience, and with material supplied. Run all three. Put the answers side by side and ask which is best and why — and then ask the harder question, which is whether the best one is actually more accurate or merely better fitted to what you asked for. Those are different things, and the difference is where the real lesson lives.