A few higher-leverage prompting techniques

In this guide, you will learn three higher-leverage prompting techniques that lift your results once the basics feel natural: thinking step by step, giving the AI a role, and breaking a big task into pieces.

Ask it to think step by step

When a question needs a bit of reasoning, a problem with several stages, a decision with trade-offs, an AI often does better if you ask it to work through the steps out loud rather than leaping to an answer. Adding a line like “think it through step by step before you answer” nudges it to lay out its reasoning first.

This trick has a name: it is called chain-of-thought prompting, because you are asking the AI to show its chain of reasoning rather than only the conclusion. The benefit is twofold. The answer tends to be more reliable, and you get to see the working, so you can spot where you disagree.

It is most useful for maths, logic, planning, and any “which should I choose and why” question. For a quick factual answer it is overkill, so save it for the times that genuinely need thought.

Give it a role to play

The AI can adjust how it answers depending on who you ask it to be. “You are a patient primary school teacher” produces a very different explanation from “you are a plain-speaking financial adviser,” even for the same question. Naming a role sets the tone, the vocabulary, and the level of detail all at once.

This is a friendly, everyday version of something the tools do behind the scenes. Many AI products come with a hidden instruction that shapes their default personality and rules, called a system prompt. You cannot always edit that, but giving a role in your own request is a simple way to steer the same dials yourself.

A role also pairs beautifully with examples. If you have read our guide on showing the AI what you want, you will recognise this as another form of aiming: “act as a copy editor, and here are two sentences in the style I like” combines a role with few-shot prompting for a sharp, specific result.

Break a big task into pieces

When a task is large, asking for the whole thing at once often gives you a shallow version of everything and a great version of nothing. The fix is to split it up. Ask for an outline first, agree on it, then ask the AI to write each section in turn.

This mirrors how you would tackle the job yourself. You would not write a report in one unbroken sitting without a plan; you would sketch the shape, then fill it in. Letting the AI do the same gives you control at each stage and a much stronger finished result.

Try it yourself

Let’s combine a role and step-by-step thinking on a real decision. Paste something like this:

You are a level-headed adviser who is good with money.

I'm deciding between two second-hand cars. One is cheaper to buy but older
with higher mileage; the other costs more but has a service history and
lower mileage.

Think it through step by step, weighing running costs, reliability, and
resale, then give me a clear recommendation at the end.

Notice how the role sets the tone and the “step by step” line makes the reasoning visible, so you can follow the logic rather than taking the verdict on trust.

A common first instinct

A common first instinct, once you hear that techniques like these exist, is to cram all of them into every prompt at once. You will get cleaner results if you reach for the one that fits the task: step-by-step for reasoning, a role for tone, splitting up for big jobs. Choosing the right tool for the moment is the real skill behind what people call prompt engineering, and it grows naturally with practice.

Next steps

You now have a few reliable techniques to draw on beyond the basics. To round out your prompt craft, it helps to know the pitfalls too, so next we will look at the common prompting mistakes and how to sidestep them.