What a prompt actually is

Nearly every prompt that works has the same three parts. Learn to spot them and you have most of what people call prompt engineering.

The three parts

Context is what the model needs and cannot guess: the audience, the codebase, the constraints, the source text. Instruction is the task. Output shape is what should come back, meaning length, format and sections.

Disappointing prompts usually lack the third. Summarise this has context and an instruction but no shape, so the model picks one, and it picks a competent, generic summary aimed at nobody.

Order changes the answer

Put the input last. Models give more weight to instructions near the end of a prompt, and a marked block at the bottom, after a line like --- text ---, is far less likely to be read as an instruction. It is also your first defence against prompt-injection, where text inside your input tries to act like a command.

The working order: context, instruction, output shape, input.

Vague against specific

Compare these two.

Make this text better please.

Rewrite this product description for a Slovak audience. Keep it under 60 words, keep the two technical specs exactly as they are, and drop the marketing adjectives.

The second is not polite or detailed for its own sake. Each clause takes one decision away from the model: who it is for, how long, what must stay, what to cut.

The rule that saves the most time

Let the model answer nothing. Ask for three problems and you get three problems, real or not, because it completes the pattern you gave it. Add a line such as: if there are none, say so in one line. The invented findings stop.

What to do next

Pick a prompt you already use and write down which of the three parts it lacks. Then open the few-shot page in the glossary. One worked example is usually quicker than describing the shape in words.