The Principles of Effective Prompting

The same question, formulated in two different ways, can produce radically different results in a language model. Not because the model is capricious, but because the input text — the prompt — is literally all the context the model has to produce a response. What is not in the prompt, the model infers. And its inferences may not match what you wanted.

Learning to write good prompts is not an expert’s trick — it is the fundamental skill for working with AI efficiently.

Why the prompt matters so much

A language model has no knowledge of your specific situation, your goals, your constraints or your audience beyond what you tell it. When you receive a generic, unhelpful or wrongly-toned response, the most likely cause is not that the model is incapable: it is that it did not have enough information to do better.

Prompting is, in essence, precise communication. The same skills that make a brief effective, or that make an instruction to a colleague clear, make a prompt effective.

The four elements of an effective prompt

A prompt that works well usually contains four elements. Not all are mandatory in every case, but when something fails, the cause is usually the absence of one of them.

1. Role

Tell the model from which perspective or with what expertise it should respond. This is not fiction: it actively changes which training patterns the model activates.

Weak: “Explain inflation.” Better: “You are an economist explaining inflation to someone with no economics background who has just received their first paycheque.”

2. Context

Provide the background information the model needs to understand your specific situation. Without context, the model responds to the most generic version of the question possible.

Weak: “Can you improve this email?” Better: “I am writing to a client who has not replied to my proposal for three weeks. I want to follow up without seeming pushy. Our relationship is professional but warm. The current email is: [email].”

3. Task

Define precisely what you want the model to do. Vague instructions produce vague responses. Specific instructions produce actionable results.

Weak: “Help me with my presentation.” Better: “Identify the three weakest arguments in this presentation and propose how to strengthen each one with more specific data or concrete examples.”

4. Format

Specify how you want the response structured. If you do not, the model will choose the format that seems most natural for the context — which may not be the most useful for you.

Weak: (no format specification) Better: “Respond as three brief bullet points, each with a concrete action on the first line and an example on the second.”

Precision over brevity

A common mistake is assuming prompts must be short. They do not. The optimal length is whatever contains all the necessary information without adding noise.

A 50-word prompt that precisely defines the role, context, task and format produces better results than a 10-word one that leaves too much to the model’s inference.

That said, repetition and redundancy do not help. Length should come from precision, not filler.

Common mistakes

Asking for what you want rather than specifying the task. “Give me good ideas for my business” does not describe a task; it describes a wish. A task is: “Generate ten product ideas for an online florist targeting corporate offices in mid-sized cities, ordered from lowest to highest initial investment.”

Assuming the model knows the context. The model does not know who you are, what company you have, what you have already tried, or what your audience is unless you say so in the prompt.

Not specifying length or format. The model will tend towards medium-length prose responses. If you need something different, ask for it.

Asking for several things in one prompt without structuring them. If you have three different tasks, a prompt that mixes them produces a response that mixes all three. It is better to treat them separately or ask the model to address them sequentially.

The base template

This structure works for most situations:

ROLE: [Who you are or from what perspective you should respond]

CONTEXT: [Background information you need to know]

TASK: [What exactly I want you to do]

CONSTRAINTS: [What the response should not include, limits]

FORMAT: [How I want the response structured]

You do not need to use these labels literally. You can incorporate the same elements in natural prose. What matters is that they are all present.

Prompting improves with practice. The first version of a prompt is rarely optimal — the real process is to iterate: see the response, identify what information was missing, add it, repeat. In the next chapter we will look at more advanced techniques for more complex situations.