My Prompting Principles

What I Learned about Prompting

When I meet people at events and conferences, I often ask them to share their favorite prompts. Most people light up talking about their work with prompts and this often leads to an interactive discussion about best practices, what works with which model and how their prompting skill has evolved over time.

Once in a while I also run into people who are still very early to prompting, not fully comfortable and for a number of reasons, they are hesitant to use language models. They discuss what they are trying to achieve and ask for advice. Often, the issue boils down to prompting.

So in this short post, I am skipping detailed commentary and examples and going straight to sharing my key principles about prompts. If you are new to prompting, you are welcome to read my earlier post on prompting.

Illustration of prompt refinement. Source: Napkin.ai

Here’s my list:

1. Collaborate with AI: Treat AI as a helpful assistant and a thoughtful companion with an excellent knowledge base, which can compose specific responses for your request. It can reason and it can be creative in its own ways. Don’t think of it as just another point-and-click tool or as a glorified search engine.
2. Use conversations, back-and-forth dialogue to progressively get more value. There’s a reason its called chatbot. Question or critique the AI’s replies, ask for corrections, clarifications and refinements. Iterate for deeper insights. Ask the model if it understands your instructions.
3. Be clear, specific and detailed. State your expertise level, your goals and context. Explain what you’re trying to accomplish and how you’ll use the information. Include as much context as possible.
4. Break complex tasks into smaller steps. Ask for one thing at a time rather than overwhelming with multiple requests. Ask the AI to “think step by step” or use reasoning models for complex tasks.
5. Provide examples and specify your preferred format and length. Explicitly state how you want information presented (bullet points, charts, JSON format.)
6. Set creative boundaries. Define style, tone, scope to guide the AI’s imagination.
7. Request alternative perspectives. Ask for different approaches or viewpoints on a topic.
8. Ask for the prompt itself — If unsure how to ask something, request suggestions for ‘what should I ask‘ (aka interview style). You could even ask AI to write the most to effective prompt for you and then edit that prompt for your specific needs — e.g. ask ChatGPT to write a prompt for Midjourney.
9. Be aware of knowledge limitations — Understand that language models are static of frozen based on their knowledge cutoff date (the last date of model pre-training). Don’t assume that the model is connected to real-time information sources (although newer models have started using search as a tool but always verify.)
10. Verify information produced by the model independently. Remember that AI can make mistakes or generate plausible-sounding but incorrect information.

I am certain that this is not a comprehensive list — I purposely kept it at 10 so that I could keep it concise. I look forward to hearing how you use language and other AI models.