Page 60 - CyberSecurityEssentialsEnglish

What to notice: Output quality does not come from prompt length alone. It comes from a
cycle of diagnosis, measurement, and refinement.
Source: Prepared by the author based on published prompting and iterative testing guidance

[2–6].

2.1 What Is a Prompt, and Why Does Context Change the Result?

◆ Key idea: a prompt is a temporary work specification.
A prompt is everything you give the system so that it can understand the task: a question,
instructions, context, source material, an image, examples, and constraints. Prompt

engineering is the design of those inputs, their trial on realistic cases, the evaluation of
the outputs, and then their improvement or the replacement of part of the solution.

A prompt is the input that begins an interaction with an AI model or system. It may be a
short piece of text, a multi-turn conversation, a file with instructions, an image with a

question, or a system message inside an application. Reducing it to the sentence we type
into a chat box therefore misses an important part of the picture. The effective prompt may
also include the preceding conversation, standing instructions, the tools available to the

system, and the data the application retrieves.

Prompt engineering is the systematic process of designing, testing, and improving those
inputs until the output performs a defined function more consistently. The word engineering

does not mean that every user needs to write code. It means that there is an objective,
criteria, trials, and adjustments. If the task is a short, low-risk email, the process may be

simple. If the task is extracting fields that will enter another system, we will need structure,
testing, and tighter control.

Context is the information that makes the task clear in its real setting. The phrase

“summarize the report” specifies an action, but it does not specify why we are summarizing,
who will read it, what decision follows, whether the summary must stay within the report,
how long it should be, or how risks should be presented. An executive summary for a board

is not the same as a note for operations staff, and an explanation for a student is not the
same as a comparison for a researcher, even when the source is identical.

Good context is not the largest possible quantity of information. It is the smallest quantity

that changes the decision, the language, or the accuracy. Add what changes the output and
remove detail that does not affect it. Each piece of information can be tested with one
question: if I removed it, could the system still produce the same result? If the answer is yes,

it may be noise. If removing it forces the system to guess, it is part of the context that
matters.

59