21 Rules for Prompting Claude (2024 Edition)

Prompt Tip 1. Never end questions with "right?" - Not even with "maybe?" It's worse. - Do not ask yes/no questions. - "Compare X vs Y for this [context]" is far better. - Otherwise, AI is very good at following the signals. - It won't argue against, to please you. Prompt Tip 2. "Think step by step" doesn't work. - No "Think step-by-step." - No "get information, and then work on a plan." - Instead, add a role in just a few words ("…as a reliability engineer"). Prompt Tip 3. Never trust AI's cited sources. - AI is very good at sounding confident. - Open every single link. - Mentioned links may not exist. Prompt Tip 4. Only give 3 rules in a prompt. - If you have more rules, give them in the follow-up prompts. - 2nd follow-up prompt: "Check your draft against these ONE AT A TIME, then revise." - It doesn't read beyond 3 rules. Prompt Tip 5. Do not give examples anymore. - NO "Example 1: [solved case]." - Give a goal instead. - Paste any necessary real numbers & context. - Mention what you ruled out ("I didn't change format, send time, or subject"). - End the prompt with: "Ask me for the data you need before you answer." Prompt Tip 6. No prompting rule is ever green. - Test hacks that were tested on 2023 models. - When you have any new model, re-test your favorite tricks before trusting them.