OpenAI Cookbook

1. Use Role + Goal + Guardrails.
↳ Tell the model who it is, what to achieve, and what to avoid.

2. Layer your context.
↳ Give background first, rules second, and the task last.

3. Put key instructions last.
↳ The model heavily weighs the final line.

4. Use a Chain of Verification.
↳ Think step-by-step, then check each step for errors before moving on.

5. Try Dual-Pass answers.
↳ Draft first, then improve with a self-check rubric.

6. Force “I don’t know” honesty.
↳ Prevents confident but wrong answers.

7. Switch perspectives.
↳ Solve from two angles, then merge the best parts.

8. Control with delimiters.
↳ Wrap rules or data in ``` or XML tags for clarity.

9. Prime with examples.
↳ Show 2 good and 1 bad example before asking for output.

https://cookbook.openai.com/ 

https://cookbook.openai.com/examples/gpt-5/gpt-5_prompting_guide 

https://ruben.substack.com/p/chatgpt-5 

Tips by Israel Ayliffe

• Think harder
• Break this down, identify issues, then solve step by step.
• Show your thinking before giving the answer.
• Write "continue" if it gets stuck.
• Bonus: Add this to your ChatGPT custom instructions:
  "Always use deep thinking: Break every task into smaller parts. For each part, consider different angles and challenge your own ideas. Question your assumptions. Double-check your logic. Point out what you're uncertain about and mention other possible approaches."

https://www.linkedin.com/posts/israel-ayliffe_upyourpromptinggame-midjourneyai-chatgpt-activity-7363245824958423041-FDeT?utm_source=share&utm_medium=member_desktop&rcm=ACoAAEzpisQBkPff_sWgageV6FNY-ufwF1-zQ0s