Prompting that actually works

A language model has no idea who you are, what you are working on, or what a good answer would look like to you. Everything it knows about your situation arrives in the prompt. That is why two people can ask about the same topic and one gets something useful while the other gets vague filler. The difference is rarely the model. It is the ask.

So why does the wording matter so much? Because a model works by predicting what text should come next, and your prompt is the only pattern it has to continue. A vague prompt matches millions of possible situations, so the model produces something like the average answer across all of them, which is exactly what generic filler is. A specific prompt narrows the pattern down to your situation, and the prediction sharpens with it. You are not persuading the model to try harder. You are giving it a tighter pattern to complete.

Give it something to work with

Be specific about the task, and give the context a helpful stranger would need. Who is this for? What have you already done? What are the constraints? A model told you are in year 11, writing a persuasive speech for a class audience, with a three minute limit, will produce something far closer to useful than one told nothing.

Think about what happens when you ask a relief teacher for help with an assignment they have never seen. Before they can say anything useful they need the task sheet, your draft and the marking criteria, and a good one will ask for exactly those things. A model will not ask. It will answer with whatever it has, so missing context never slows it down; it just quietly lowers the quality of what comes back. The habit worth building is handing over the task sheet before anyone has to ask for it.

Say what you want back

State the format: a numbered list, a table, three options with reasons, two hundred words. If you can, show an example of the kind of output you want, because models are excellent at matching a pattern once they have seen one. Vague requests get the model's default shape, which is usually a long, polite essay you did not ask for.

One job at a time

There is one more upgrade worth knowing. Big requests produce shallow answers, because the model spreads its effort across everything you asked for at once. Ask for a full essay plan, feedback on your draft and a better title in one go and you get a thin version of all three. Ask for them one at a time, feeding each answer into the next request, and every step gets the model's full attention. Working in steps also keeps you in control of the direction, which matters more than it sounds, because each step is a chance to correct course before the next one builds on a mistake.

Weak versus strong, same task

• Weak: Make my speech better.
• Strong: I am in year 11. Below is my draft of a three minute speech arguing that school should start later. Identify the two weakest arguments, explain why each is weak, and suggest one stronger replacement for each. Keep my tone, and reply as a numbered list.

Notice what the strong version does. It gives context, defines the task precisely, sets the format, and protects what should not change. Same student, same speech, same model. Wildly different result.

There are no magic words

A common belief is that good prompting is a bag of secret tricks: special phrases that switch the model into genius mode. There is no such switch. The occasional trick does circulate, and some even work for a while, but the reliable gains all come from the same place, which is saying clearly what a helpful stranger would need to hear. Prompting well is the same skill as briefing a person well. If you can write a brief that gets a good result from a machine that cannot read your mind, you can write one that gets a good result from a busy human, and that skill turns up in every job you will ever have.

Iterate and interrogate

The first answer is a first draft of the conversation, not the end of it. Push back: ask it to go deeper on one point, cut the waffle, or take the opposite view. Ask it to explain its reasoning, and to name sources you can actually check. A prompt is not a wish you make once. It is the start of a working session.

Some follow up moves are worth memorising because they work on almost any answer. Ask what it assumed about your situation, and correct the assumptions that are wrong. Ask what someone who disagrees would say. Ask for the same answer at half the length, because the shorter version is usually tighter than anything you would get first go. And when an answer misses, resist the urge to start a fresh chat and reword the prompt from scratch. Tell the model what was wrong with the answer, in one plain sentence, and let it repair the attempt. That is usually faster, and it teaches you what your original prompt left out.

Where this shows up in your week

The same four upgrades work well beyond schoolwork. They apply when you want help planning training for your team, drafting a message to your manager about swapping a Saturday shift, or understanding a concept your teacher raced through. Compare help me study for my biology test with something like this: I have a year 10 biology test on cells on Friday, here are the topics from the revision sheet, quiz me one question at a time and tell me what to revise based on my wrong answers. The second version costs you thirty seconds of typing and changes the whole session.

Prompting is the mechanical half of using these tools well. The other half is deciding what to hand over in the first place, and that decision matters more than any wording trick, so the next lesson is about exactly that.