Editing Prompt Engineering Basics (section)

== The core techniques ==

=== Be specific about what you want ===

This is the most impactful improvement most people can make. Compare:

| Vague prompt || Specific prompt

| "Summarize this document" || "Summarize this document in 3 bullet points, focusing on budget implications for our team"
| "Write a response to this email" || "Write a professional but warm reply declining the meeting request, suggesting an async alternative"
| "Help me with my presentation" || "Give me 5 compelling opening lines for a presentation about remote work to an audience of skeptical middle managers"

You're not writing code. You're writing a brief — the same way you'd brief a colleague who's smart but doesn't know your context.

=== Give AI a role (system prompts) ===

Telling AI ''who it is'' changes the quality of its output dramatically. This is the foundation of the [[Your First AI Team Meeting|Your First AI Team Meeting]] exercise.

``<code>
You are a senior financial analyst reviewing a startup's pitch deck.
Focus on: revenue model assumptions, burn rate, and market size claims.
Be skeptical but constructive. Flag anything that seems unrealistic.
</code>`<code>

Why this works: it constrains the model's vast knowledge to a specific perspective and expertise level. Without a role, AI defaults to generic helpfulness. With a role, it brings a point of view.

=== Show examples (few-shot prompting) ===

If you want AI to produce output in a specific format or style, show it what good looks like:

</code>`<code>
Classify each customer comment as POSITIVE, NEGATIVE, or NEUTRAL.

Comment: "Love the new dashboard, it's so much faster"
Classification: POSITIVE

Comment: "The update broke my saved filters"
Classification: NEGATIVE

Comment: "I noticed you changed the sidebar layout"
Classification: NEUTRAL

Comment: "Can't believe you removed the export feature, this is useless now"
Classification:
</code>`<code>

Two or three examples are usually enough. The AI picks up the pattern — format, tone, judgment criteria — from your examples rather than having to guess what you mean.

=== Ask for step-by-step reasoning ===

For complex questions, adding "think step by step" or "show your reasoning" dramatically improves accuracy. This is called chain-of-thought prompting:

</code>`<code>
A company has 150 employees. They're cutting 12% of staff across
three departments: Engineering (80 people), Sales (45 people),
and Operations (25 people). The cuts should be proportional
to department size.

Think step by step: how many people are cut from each department?
</code>`<code>

Without the step-by-step instruction, AI often jumps to a wrong answer on multi-step problems. With it, the model works through the math visibly, and you can check each step.

=== Structure your prompt clearly ===

When your prompt has multiple parts — context, instructions, content to process — use clear structure to separate them:

</code>`<code>