DEV Community

Cover image for You’re Using ChatGPT Wrong. Here’s How to Prompt Like a Pro
Mashraf Aiman
Mashraf Aiman

Posted on

You’re Using ChatGPT Wrong. Here’s How to Prompt Like a Pro

Many people treat ChatGPT like a magic search box and then wonder why the output is vague, off-target, or too generic. The difference between a mediocre answer and a professional-grade result is almost always the prompt. Prompting is a skill you can learn quickly. Below is a concise, practical guide that teaches how to ask for exactly what you need and get repeatable, high-quality results.

Why prompts fail

  • Lack of context. The model doesn’t know your constraints, audience, or what you already tried.
  • Vague instructions. “Help me with X” is a start, not a specification.
  • No expected format. Without a format requirement, the output is free-form and inconsistent.
  • No iteration plan. Many problems are solved best with a short loop: ask → refine → repeat.

Core principles of pro-level prompting
1) Give minimal but sufficient context

Tell the model what matters: audience, purpose, restrictions, and any prior attempts. Example: “I’m writing a technical blog for mid-level backend engineers about connection pooling in Node.js.”

2) Define the role and tone

If you want a tutor, reviewer, or reporter, say so. Example: “Act as an experienced software architect and explain...” Add tone: “concise, direct, and professional.”

3) Specify exact output format

JSON, markdown, bullet list, a one-paragraph summary, or a 10-step checklist — be explicit. Example: “Return only a Markdown article with H2 headings and a 150–200 word introduction.”

4) State constraints and edge cases

Include limits like word counts, banned words, or regulatory constraints. Example: “Do not use the words ‘always’ or ‘never’. Keep the summary under 180 words.”

5) Provide examples and few-shot demonstrations

Show what good looks like. If you want a rewrite, paste a short sample and ask for an improved version in the same voice.

6) Ask for step-by-step or multi-stage outputs when needed

For complex tasks, request a staged response: plan → draft → revise. Example: “First give a 3-step plan, then produce the draft article.”

7) Use iterative refinement: evaluate, then refine

Treat the model like a coauthor: generate, critique, and ask for revisions with specific change requests.

8) Use meta-prompts sparingly and precisely

You can instruct the model how to self-evaluate — for example, “Also output three test prompts you would use to validate this content.” But avoid asking for hidden chain-of-thought.

Practical prompt templates (copy and adapt)

A. Blog post (technical)

You are an experienced backend engineer and technical writer. Produce a Markdown article titled "Connection Pooling Best Practices". Audience: intermediate Node.js developers. Length: 900–1100 words. Include: brief intro, 4 H2 sections (why pooling matters, common mistakes, recommended patterns, production checklist), one short code snippet (Node.js with pg-pool), and a 3-bullet TL;DR. No emojis. Use a professional tone.
Enter fullscreen mode Exit fullscreen mode

B. Debugging help (code + context)

I have this failing test in Jest. Here is the minimal code and error.
<paste code and error>
Explain the cause in one paragraph, list 3 possible fixes ranked by safety, and show the safest fix as a patch formatted in a unified diff. Also include a one-sentence test to validate the fix.
Enter fullscreen mode Exit fullscreen mode

C. Brainstorming (ideas)

Act as a product strategist. I need 10 feature ideas for an education app that increase retention for new users. For each idea, give: 1-sentence summary, expected implementation risk (low/medium/high), and one KPI to measure success. Keep it to 10 items.
Enter fullscreen mode Exit fullscreen mode

D. Resume / LinkedIn copy (concise)

You are a professional resume editor. Rewrite this bullet into a 1-line achievement that includes: metrics, technology used, and outcome. Original: "Built a reporting system for sales." Limit to 20–25 words.
Enter fullscreen mode Exit fullscreen mode

Advanced tips and patterns

  • Anchor with constraints first: start prompt with “Constraints:” then list them. This prevents later ambiguity.
  • Use role + context + task structure: “Role: X. Context: Y. Task: Z.” This reduces back-and-forth.
  • For code, request tests: “Provide unit tests using Jest that validate the main edge cases.” Tests make the output immediately actionable.
  • Avoid vague adjectives: replace “good” or “better” with “shorter than 150 words” or “avoid passive voice.”
  • If you need creativity, ask for many ideas but narrow the output: “Give 20 ideas, then pick the top 3 with reasons.”
  • When collaborating, ask for explainable edits: “Show the original line, the edited line, and a one-line rationale for the change.”

Examples of bad → good prompts

Bad:

Help me write an intro paragraph for my blog.
Enter fullscreen mode Exit fullscreen mode

Good:

Write a 120–150 word introduction for a blog aimed at mid-level backend engineers about avoiding common SQL anti-patterns. Tone: practical, slightly conversational. Include a one-sentence example of an anti-pattern.
Enter fullscreen mode Exit fullscreen mode

Bad:

How do I fix this bug?
Enter fullscreen mode Exit fullscreen mode

Good:

I’m getting this error: <paste exact stack trace>. Environment: Node.js 18, Postgres 14, sequelize v6. Show the most likely root cause and three prioritized fixes. For the top fix, give a 6-line code patch.
Enter fullscreen mode Exit fullscreen mode

When to use system messages or persona framing
If your interface supports role/system messages, use them to set persistent context: “You are an engineering mentor who explains concepts with diagrams and concise code examples.” Reserve the user message for the specific task.

Guardrails and ethics

  • Don’t ask the model to produce proprietary or personal data it cannot access.
  • For legal, medical, or safety-critical advice, request a cautious, non-definitive summary and follow up with professional review.
  • Ask for sources when factual claims are made, and confirm in external systems if necessary.

Workflow for high-impact tasks (pattern you can reuse)

  1. Prep: give context, constraints, desired format.
  2. Draft: ask for a complete first draft.
  3. Critique: request enumerated weaknesses and improvements.
  4. Revise: ask for a new version incorporating exact changes.
  5. Validate: request tests, checklists, or follow-up prompts.

Quick checklist before you hit enter

  • Is the desired output format explicit?
  • Did you give the minimal necessary context?
  • Did you set a role and tone?
  • Did you include constraints and edge cases?
  • Is there a follow-up or iteration plan?

Closing note
Great prompting is a small time investment that multiplies the usefulness of the model. Be explicit. Treat the model like a specialist you’re hiring for one task. Give context, define deliverables, and iterate. You’ll get better answers, faster.

If you want, I can:

  • Turn one of the templates into an automation-ready prompt for your team, or
  • Create a short checklist card you can paste into a ticketing tool.

— Thanks,
Mashraf Aiman
AGS NIRAPAD Alliance,
Co-founder, CTO, ENNOVAT

Top comments (0)