Fundamentals

Hands-on guide to optimizing prompts for AI Actions

Unlocking the full potential of AI Actions through effective prompt engineering

AI Actions is designed to extract valuable insights from articles quickly and efficiently. Crafting effective prompts is the key to getting the most out of this feature.

In this guide, we'll go through a real-world example of how to optimize a prompt used by one of our users. We'll break down the process step-by-step, showing how to transform a good prompt into a great one that delivers more accurate and useful results.

By the end of this guide, you'll have practical strategies to:

• Structure your prompts for clarity and effectiveness
• Avoid common pitfalls that can confuse AI models
• Enhance the quality and relevance of your reports

Note: For purposes of illustration, we focused these examples on cyber threat intelligence use cases. The general guidance can apply to other non-cybersecurity use cases as well.

The power of working backward

Before diving into the prompt analysis and optimization, let's look at what the user wants to achieve (Figure 1). This approach of starting with the desired output and working backward is a powerful strategy in prompt engineering: by visualizing the desired output, we can more effectively craft a prompt to achieve it.

Figure 1: The desired output, including the 'Overall Score' column with detailed breakdown

This output includes:

• Categorization of vulnerabilities by severity
• Detailed information for each vulnerability
• An Overall Score column with a breakdown of the score components

Now that we know what we're aiming for, we’ll examine the original prompt that was crafted to achieve this result.

Original prompt analysis

Here's the original prompt used to generate the vulnerability brief:

Figure 2: The original prompt used for generating a vulnerability brief

Strengths of the original prompt

• Clearly defines the expected output (vulnerabilities brief with categorized severity levels)
• Provides a structured format for the output (table with specific columns)
• Includes instructions for calculating a custom "Temporal Score"

Areas for improvement

• Terminology: The use of "Temporal Score" might cause confusion with existing cybersecurity metrics.
• Instruction order: The flow of instructions could be more logical and easier to follow.
• Text structure: Large blocks of text make the prompt harder to parse quickly.
• Calculation clarity: The method for calculating the overall score isn't explicitly defined.

Optimization process

By refining the prompt we will address these issues:

Clarify terminology

• "Temporal Score" is an established cybersecurity term. Replace it with a more generic term like "Overall Score" to avoid confusion.
• Explanation: This change ensures that our custom metric is clearly distinguished from standard industry terminology.

Restructure instructions

• Move the calculation of the Overall Score to the beginning of the prompt.
• Explanation: This logical ordering helps the AI understand the full context before creating the final output.

Improve readability

• Break down long paragraphs into bullet points and numbered lists.
• Use clear headings and subheadings to organize information.
• Explanation: This structure makes the prompt easier to read and understand, both for the AI and for human users reviewing the prompt.

Define calculations explicitly

• Clearly state the formula for the Overall Score: Sum of Recency, Exploitation Status, Patch Availability, and Impact Severity scores.
• Explanation: Explicit instructions reduce the chance of misinterpretation and ensure consistent results.

Add step labels

• Introduce clear labels for each major step in the process (e.g. "Step 1: Calculate Overall Score").
• Explanation: This helps break down the task into manageable chunks and makes the prompt's structure more apparent.

Optimized prompt breakdown

Here's our optimized prompt with explanations for each section:

Figure 3: The optimized prompt with improved structure and clarity

The optimized prompt addresses the issues identified in the original version:

• Clear structure: The prompt is now organized into distinct steps, making it easier to follow.
• Explicit calculations: The overall score calculation is clearly defined at the beginning.
• Consistent formatting: Scoring guides are presented in a uniform, easy-to-read format.
• Logical flow: Instructions progress logically from data gathering to output generation.
• Precise output instructions: The final step clearly specifies how to present the results.

These improvements enhance the prompt's clarity, making it more effective for both the AI model and human users reviewing the prompt.

A note on results presentation

While our optimized prompt aims for the output in Figure 1, consider this alternative presentation:

Figure 4: Revised results of the AI prompt including Overall Score and detailed calculations outside the table

The advantages of this approach are:

• Enhanced readability: Presenting the detailed breakdown outside the table gives the AI more room to express its reasoning clearly.
• Improved explainability: This format allows for more detailed explanations of each score component, which can be crucial for understanding the rationale behind the overall score.
• Easier verification: With the breakdown separated from the table, it's easier for analysts to verify the calculations and spot any potential inconsistencies.
• Flexibility: This approach provides flexibility in how much detail to include without compromising the table's readability.

While this presentation differs from the original request, it aims to explain how giving the AI more freedom in expressing its output can lead to more comprehensive and useful results. When implementing this in your own prompts, consider the trade-offs between a compact presentation and a detailed explanation based on your specific needs.

Key Takeaways

• Use clear, unambiguous terminology to avoid confusion
• Structure your prompt logically, with calculations and data gathering before output generation
• Break down complex tasks into clearly labeled steps
• Use bullet points, numbered lists, and tables to improve readability
• Provide explicit formulas and criteria for any calculations
• Include clear instructions for the desired output format
• Test and iterate on your prompts to continually improve their effectiveness

Conclusion

Optimizing the prompts for AI Actions is an iterative process that combines clear communication, logical structure, and an understanding of how AI models process information. By applying the techniques demonstrated in this guide, you can significantly enhance the quality and relevance of the insights you gather through AI Actions.

Try AI Actions

Synthesize, translate, extract data, create reports, summarize, and more.Start free trial