Prompt engineering is writing instructions so a model reliably does what you want. It is closer to writing a clear brief for a new colleague than to a magic spell.
Be clear about the task
State what you want, who it is for and what a good result looks like. Compare:
- Weak: "Improve my resume."
- Better: "Rewrite the work-experience bullets below for a customer support role. Keep every fact true, start each bullet with an action verb, keep each under 20 words."
Zero-shot, one-shot and few-shot
- Zero-shot: just the instruction, no examples.
- One-shot: one example of the input and the output you want.
- Few-shot: several examples. This is the most reliable way to fix a format or tone.
Task: Rewrite the bullet in a stronger, factual style.
Example input: Was responsible for answering calls
Example output: Resolved 40+ customer calls per shift with a focus on first-call resolution.
Input: Helped the team with reports
Output:Role, context and constraints
- Role: "You are an experienced HR recruiter reviewing resumes for warehouse roles."
- Context: who the reader is, what the goal is, what the input is.
- Positive constraints: say what to do ("use plain English").
- Negative constraints: say what to avoid ("do not invent employers or dates").
Templates and variables
In an application you rarely write each prompt by hand. You write a prompt template with variables such as {job_title} and {resume_text}, then fill them in. Keep templates in version control, as you would code.