Prompt engineering in practice

Prompting as clear communication gave you the briefing template. This page is the workshop: patterns you’ll reuse, with concrete examples. No magic spells — just clearer asks.

Pattern 1 — Role + job + deliverable

Tell the model who it’s helping, what to do, and what to hand back.

You are helping a career switcher write a plain-language cover letter.
Job: Turn my bullet notes into a 3-paragraph letter.
Deliverable: Draft only — no commentary. Keep under 250 words.
Notes: …

Pattern 2 — Paste the source; constrain the truth

When summarizing or extracting, put the source in the prompt and ban freelancing.

Using ONLY the text between <doc> and </doc>, list three risks mentioned.
If a risk isn’t in the doc, omit it. Quote a short phrase for each.

<doc>
…paste article…
</doc>

Pattern 3 — Show the shape (few-shot)

One tiny example beats a paragraph of abstract style notes.

Rewrite each sentence to be shorter and more concrete.

Example:
In: We will endeavor to optimize the customer journey.
Out: We’ll make checkout simpler.

Now rewrite:
In: …

Pattern 4 — Checklists and self-review

Ask for a draft then a pass against your rules.

1) Draft a friendly email postponing a meeting to next Tuesday afternoon.
2) Then list any unclear details I still need to fill in (time zone, link, agenda).

Pattern 5 — Iterate on purpose

Keep a short loop: try → name what’s wrong → tighten one constraint.

• “Cut by 40%; keep the deadline.”
• “Same facts; less enthusiasm.”
• “Table with columns: claim | source quote | confidence (high/med/low).”

Patterns to skip (for now)

• Huge “mega-prompts” you don’t understand.
• Prompts that demand secrets or other people’s private data.
• Pretending the model is a licensed professional for high-stakes advice.

See also: Glossary — Prompt, Prompt engineering.

Takeaway

Prompt engineering here means deliberate briefing and revision — not a secret dialect. Practice on real tasks; keep the personal AI workflow for verify and cite.