Learn prompt engineering from scratch with practical templates you can use immediately. This guide gives you a repeatable system for better prompts: goals, context, roles, output formats, testing and iteration.
Prompt engineering is designing instructions, context and output formats so AI models deliver consistent results. Instead of vague requests, you provide goals, constraints and examples—turning the model into a reliable assistant rather than a guess machine.
A strong prompt follows a simple structure: goal, role, context, constraints, and output format. Start simple, test with real inputs, then refine only what improves quality (structure, fewer hallucinations, more consistent tone).
CRISP is a memorable framework you can apply to almost any prompt—so your instructions become clearer, testable, and reusable.
Context: [Who is the audience? What inputs do you have?]
Role: You are a [role] with [focus].
Instructions: 1) [step] 2) [step] 3) [step]. Avoid: [pitfalls].
Style: [tone], format as [Markdown/table/JSON].
Parameters: language=[...], length=[...], include=[...], quality criteria=[...].A good prompt is not longer by default—it’s more specific. Here’s a concrete example you can reuse.
Write a blog article about SEO.You are an SEO expert specialized in SaaS. Write a 1,200-word blog article for marketing managers. Use H2 headings, include practical examples, and end with a checklist. Keep it actionable and avoid generic advice.For complex tasks, break the problem into sub-steps (analyze → plan → produce). In practice, this often beats one huge prompt—especially for long-form writing, strategies, debugging, or analysis.
Provide 1–3 examples (input → output) so the model matches style, format and quality. This is extremely effective for recurring tasks like social posts, product copy, emails, code reviews and docs.
Define a role explicitly (e.g., SEO consultant, Symfony developer, support agent). A good role sets standards and priorities—reducing vague answers and improving consistency.
Instead of asking for hidden reasoning, use techniques that improve reliability and make outputs easier to validate.
Most “bad results” come from missing context, unclear success criteria, or doing too much at once.
Use these as starting points, then run them through the Prompt Optimizer and save your best versions as Skills.
You are a B2B SaaS marketer. Create a 14-day LinkedIn content plan for [product], targeting [ICP]. Output as a table with Hook, Angle, CTA, and KPI.You are a senior developer. Review this code for bugs and performance issues. Provide: 1) critical issues, 2) suggested fixes, 3) improved version, 4) tests. Format in Markdown.You are a teacher. Explain [topic] to a 14-year-old with a simple analogy, then give 5 practice questions with answers.You are a support agent. Draft a reply to this ticket: [ticket text]. Be empathetic, ask 1–2 clarifying questions, and provide step-by-step instructions.You are an SDR. Write a 120-word cold email for [persona] about [offer]. Include a personalized opener, one benefit, one proof point, and a low-friction CTA.You are a data analyst. Analyze this dataset summary: [summary]. Provide 3 insights, 2 hypotheses, and recommend next steps. Output as a structured bullet list.Pick a real task and test three variants: (1) more context, (2) stronger constraints, (3) add one example. Compare outputs for quality and consistency, then keep the best pieces as reusable prompt building blocks.