Prompt Technique

What is Prompting and Why It Matters Today

We're in the early stages of an AI revolution. At this point, the real question isn't "Should we use AI or not?" but rather "How can we effectively leverage AI's capabilities?"

I began seriously exploring how to properly use AI a few months ago. As a non-professional prompt engineer like myself, I've discovered that there are just a few techniques you'll use daily—techniques we can quickly cover so you can reap most of the benefits.

Prompt engineering is the process of refining AI prompts to improve their accuracy and effectiveness. This blog, referencing Anthropic's​ (Claude) ​prompt engineering guide​, ​will cover​ basic techniques that will quickly ​help you achieve​ 80% efficiency in prompting, while building a foundation for exploring more advanced techniques later to gain that additional 20% improvement over time. I prioritize examples over raw theory, as I believe this helps you quickly grasp the idea behind each skill.

Note: Throughout this blog, terms like "AI" and "chatbot" refer to generative intelligence models. I use these terms because they're more common in everyday conversation. Also, I won't include lengthy example outputs as they might make the blog intimidatingly long at first glance, when our goal is to quickly cover useful techniques. You can find example outputs for each technique in the Anthropic documentation (most examples come from here) and the Prompt Engineering Guide.

General Tips and Techniques

Specificity & Avoiding Vagueness

This is the foundation of all prompting techniques. When interacting with AI, think of it as a brilliant but very new employee. Imagine showing your prompt to a colleague with minimal context about the task and asking them to follow your instructions. If they might be confused, your AI chatbot likely will be too.

To be more specific about your task:

• Provide contextual information: What will the result be used for? Who is the audience?
• Clearly state what you want: If you need the AI to perform a categorized task and return only the label without explanation, say so explicitly.
• Break instructions into sequential steps: Use numbered lists or bullet points to ensure the AI completes the task exactly as you want

Example:

To Do or Not to Do?

A key tip when designing prompts is to focus on what the AI should do rather than what it shouldn't do. This encourages specificity and directs the AI toward behaviors that lead to better responses. Example:

Example:

Use Examples

Examples are your secret weapon for getting AI to generate exactly what you need. By providing a few well-crafted examples in your prompt, you can dramatically improve the accuracy, consistency, and quality of outputs. You can wrap your examples in <example> tags (or nested within <examples> tags for multiple examples).

Example:

Chain of Thought (CoT)

One of the most powerful prompting techniques is encouraging the AI to think through problems step by step, which dramatically improves performance. Chain of Thought was first introduced in Wei et al. (2022) by outlining specific steps for AI to follow in its thinking process, you increase the accuracy and coherence of results. Later, zero-shot CoT was introduced, simply adding "Let's think step by step" to prompts and letting the AI build its own thinking process.

Example:

Note: XML tags (<thinking>) work well with Claude, but ChatGPT and some other models perform better when guided to think with markdown formatting like ###Think###.

Use XML Tags

When your prompts involve multiple components like context, instructions, and examples, XML tags can be a game-changer. They help the AI parse your prompts more accurately, leading to higher-quality outputs.

For hierarchical content, use nested tags: <outer><inner></inner></outer>.

Example:

Giving AI a Role

In complex scenarios like legal analysis or financial modeling, role prompting can dramatically improve performance.

Example:

Chain Complex Prompts

When working with complex tasks, AI can sometimes struggle if you try to handle everything in a single prompt. Breaking down complex tasks into smaller, manageable subtasks improves accuracy, clarity, and traceability (making it easier to identify which part of the prompt caused problems). This approach also guides the AI to double-check, give feedback, and improve the result.

Example:

Conclusion

After exploring these techniques, I hope you'll apply them to your daily interactions with generative AI models. Personally, I find Chain of Thought (CoT) and XML tags most useful for my daily tasks. Always remember to be crystal clear about both the context and the specific task you want the AI to perform. This is why these models are particularly powerful for senior-level users who can provide specific guidance on what they need.

If you'd like to discuss any of these techniques further or share additional tips you use daily with generative AI, please comment below so we can continue learning from each other.