The same AI tool gives one person mediocre answers and another excellent ones. The difference is rarely in the model — it is in how we ask. That has a name: prompt engineering, the art of writing instructions the AI understands and executes well.
Why the request matters so much
An LLM responds to what it reads. A vague request ("write about marketing") yields a vague answer. A precise request, with context and goal, yields something useful. You do not need to be a programmer — you need to be clear, as if delegating to a competent colleague who cannot read your mind.
The ingredients of a good prompt
- Context: who you are, who it is for, the situation ("I manage an industrial SME...").
- Clear task: what you want exactly ("summarize in 5 points", "write a reply email").
- Format: list, table, formal or informal tone, word count.
- Examples: showing one or two examples of what you expect improves the result enormously.
Give the AI a role
Starting with "act as an experienced financial analyst" or "you are a copywriter specialized in healthcare" steers the model toward the right tone and knowledge. Defining a role is one of the simplest and most effective ways to raise answer quality.
Iterate instead of giving up
The first prompt is rarely the best. If the answer does not fit, do not start over — refine: "shorter", "with a concrete example", "in a more direct tone". The conversation is a tool; each adjustment gets you closer to what you need.
The most common mistake
Asking too much at once. "Make me a complete marketing plan" gives a generic result. Break it down: first the target audience, then the channels, then the messages. Small, specific steps always beat a giant, vague request.
In practice
Write prompts the way you delegate work: with context, goal, format and examples, and iterating. It is a skill you learn quickly and it improves everything you do with AI. When did AI last disappoint you — and did your request have enough context?
