ChatGPT prompt engineering: a practical guide

Why structure beats length

Most ChatGPT failures are not model failures — they are instruction failures. A prompt that says "write a blog post about AI" forces the model to guess audience, tone, length, and structure. A prompt that opens with You are a B2B SaaS editor, defines 600 words, three H2 sections, and banned phrases removes guesswork.

My Copyprompt templates follow this pattern because it survives model updates better than clever one-liners.

The four-block template

Use four blocks in order:

1. Role + objective — "You are a senior technical editor. Your job is to…"
2. Context — paste facts, URLs, or bullet inputs the model must use
3. Rules — 4–7 dos and don'ts (cite sources, no hallucinated stats, markdown output)
4. Output format — headings, table, JSON schema, or word count

Example skeleton:

You are a customer success lead. Draft a renewal email for a B2B analytics account.

Context:
- Customer: {COMPANY}
- Usage trend: {METRIC}
- Renewal date: {DATE}

Rules:
- Lead with value delivered, not feature list
- One clear CTA for a 15-minute call
- Under 180 words
- No exclamation marks

Output: subject line + body in markdown

Placeholders

Curly-brace tokens like {COMPANY} tell humans what to swap without confusing the model. Replace every placeholder before sending; leaving {COMPANY} in place is the most common copy-paste mistake.

Temperature and follow-ups

For factual or coding tasks, ask for step-by-step reasoning then final answer only. For creative drafts, request two variants (concise vs narrative) in one shot to avoid re-rolling.

Common pitfalls

• Stacking conflicting styles ("be brief" + "be exhaustive")
• Hidden constraints buried in paragraph 3 — put rules in a list
• No failure mode — add "If data is missing, ask one clarifying question"

Next steps

Browse ChatGPT prompts on My Copyprompt for ready-made templates, or submit your own after you validate it twice in production.