Reliable prompt patterns include giving the AI a persona, asking for step-by-step reasoning, showing examples (few-shot), asking it to critique its answer, letting it interview you first, using templates, comparing options, setting constraints, using checklists and grounding answers in provided sources. Combining two or three patterns usually works best.
Patterns beat tricks
Prompt "hacks" go out of date as models change. Patterns last, because they are really just clear communication. Here are ten that professionals use daily.
1. The persona pattern
Give the AI a relevant role and audience.
"You are an experienced IELTS examiner. Review my essay as you would in a real exam, for a student targeting band 7."
Why it works: it sets the expertise, standards and vocabulary.
2. The step-by-step pattern
Ask for reasoning before the answer on complex problems.
"Work through this problem step by step, explaining each step, before giving the final answer."
Why it works: it reduces careless reasoning errors and shows you where mistakes happen.
3. The few-shot pattern
Show examples of the output you want.
"Here are two example flashcards in my format: [examples]. Create 15 more on cell biology in exactly this format."
Why it works: examples communicate style and format better than descriptions.
4. The critique pattern
Ask the AI to review its own output.
"Now critique your answer. What is missing, what might be wrong, and what would an expert disagree with?"
Why it works: models often catch errors when asked to look for them.
5. The interview pattern
Let the AI ask you questions first.
"I want a revision plan. Before you write it, ask me up to five questions about my exams, time and weak areas."
Why it works: it gathers context you forgot to provide.
6. The template pattern
Give a fixed structure to fill.
"Summarise this article using: Main claim / Evidence / Weaknesses / One question to ask the author."
Why it works: consistent, scannable output you can compare across many items.
7. The comparison pattern
Ask for options and trade-offs, not one answer.
"Give me three ways to structure this presentation, with the pros and cons of each, and recommend one for a non-technical audience."
Why it works: you make the decision with better information.
8. The constraints pattern
Set clear limits.
"Explain in under 120 words, with no jargon, using one everyday analogy."
Why it works: constraints force focus and prevent generic padding.
9. The checklist pattern
Ask the AI to check work against criteria.
"Check my lab report against this checklist: aim, method, results table, analysis, evaluation. For each, say present or missing and how to improve."
Why it works: structured feedback is more actionable than "this is good".
10. The grounding pattern
Restrict the AI to your sources.
"Answer only using the text below. Quote the sentence that supports each point. If the answer is not in the text, say so."
Why it works: it dramatically reduces hallucinations. See AI hallucinations explained.
Combining patterns
The most effective prompts combine patterns. For example, persona + grounding + template:
"You are a history tutor. Using only the source below, fill in: Provenance / Content / Purpose / Reliability. Quote evidence for each."
| Situation | Best patterns |
|---|---|
| Learning a topic | Persona, constraints, critique |
| Revising | Interview, checklist, few-shot |
| Research | Grounding, template, comparison |
| Writing feedback | Persona, checklist, critique |
| Problem solving | Step-by-step, critique |
Practise with feedback
Patterns become natural with practice. In our prompt engineering course, you build a tested prompt library for your own study or work, using premium AI tools.
Frequently asked questions
Do these patterns work in every AI assistant?
Yes, the principles work across major assistants, although results vary by model. Test important prompts in your chosen tool.
Should I always use several patterns?
Only when the task needs it. Simple questions need simple prompts.
Can I save prompts to reuse?
Yes. Keep a prompt library in a document, or use custom instructions and saved assistants in your AI tool.
