How to Assess a Prompt?

You’ve probably come across all kinds of tips and tricks for writing AI prompts. Trust me, it doesn’t have to be complicated!

I’ve boiled it all down to three simple criteria that anyone can use. Follow these, and you’ll get consistent, high-quality results from AI, no matter what you’re working on.

Let’s break them down.

1. Focused - Focus on one clear, well-defined task.

A great prompt focuses on doing one thing very well. Avoid the temptation to cram too big of a task into a single prompt. Instead, narrow it down to a micro task and provide the model with specific knowledge or context to work with.

Remember: a focused prompt delivers a focused output.

👎 Don’ts

• Boil the ocean: “Write my next newsletter about AI for Marketing.”
• Combine multiple tasks: “Write the structure of my newsletter about AI and Marketing, give alternative titles, and suggest a CTA.”

👍 Do’s

• Focus on one micro task: “Formulate 5 alternative titles for my newsletter about AI and Marketing.”
• Break large tasks into a chain of micro tasks: First, “List 10 pain points marketers have when it comes to AI.” Then, “Formulate 5 alternative titles for my newsletter that will address pain point #2.”

Why the Specialized criteria Works: Under the Hood

The Specialized criteria works because it reduces cognitive overload for the model.

LLMs predict the next token based on the input they receive. When you ask for multiple tasks or overly broad outputs, the model has to juggle competing predictions, increasing the chances of confusion or irrelevant results. By narrowing the focus to one clear task, you simplify the problem, allowing the model to dedicate all its computational "attention" to generating accurate and relevant output.

2. Dense - Include the right balance of information.

Make your prompt dense with all the needed and useful information, but keep it concise. Include only the critical details to guide the AI effectively, avoiding unnecessary or irrelevant content.

Think of it like giving precise instructions: enough to get the job done, nothing more.

👎 Don’ts

• Overload with unnecessary detail: “Write a LinkedIn post about Hallucination and AI. Here are all 50 of my posts from this year—use them for inspiration.”
• Dilute the context with excessive information: “Summarize the top 10 key trends in healthcare AI. Here’s the full text of a 300-page industry report—make sure you don’t miss anything.”

👍 Do’s

• Include only relevant, unique insights: “Write a blog post about AI in healthcare, focusing on patient privacy concerns. Add an insight about how startups are using federated learning to enhance privacy while sharing data securely.”
• Provide relevant information in a manageable amount: “Write a LinkedIn post about leveraging AI hallucination in brainstorming. Here are 3 posts and an article I’ve written about hallucination and one about brainstorming—use them for inspiration.”

Why the Dense criteria Works: Under the Hood

The Dense criteria works because it optimizes the model’s context window. When unnecessary or irrelevant details fill that space, the model’s predictions become less accurate.

By keeping prompts dense:

• You maximize relevant context, ensuring the model focuses on the most important aspects of the task.
• You reduce noise, avoiding distractions caused by excessive or irrelevant input.
• You sharpen predictions, helping the AI generate outputs that closely align with your goals.

A dense prompt ensures every token in the context window contributes to a high-quality result.

3. Guiding - Clarify, structure & outline the task.

Guiding the AI is about giving it the tools to approach the task in the best way possible. This includes breaking down tasks into manageable steps, providing examples, setting clear specifications, defining desired outputs, and pausing for human input when necessary.

A guiding prompt ensures the AI stays on track and builds upon your expertise.

👎 Don’ts

• Leave the process undefined for complex tasks: “Write a detailed report on AI in education.”
• Assume the AI knows the best way to achieve the task: “Create a product launch plan for my startup.”

👍 Do’s

• Break down tasks when needed: “Work step by step, 1. Identify three major AI trends in healthcare, 2. Provide examples or data to support each trend, 3. Write a conclusion explaining why these trends matter.”
• Provide examples for clarity: “Write a LinkedIn post about overcoming AI hallucination in brainstorming. Example: ‘AI hallucination can spark creativity by suggesting unexpected ideas.’”
• Set specific output formats: “Write a two-paragraph summary about how federated learning enhances privacy in healthcare. Each paragraph should contain one example.”
• Pause for human input: “List 5 product names for an AI startup. I’ll pick one, and then write a tagline for it.”

Why the Guiding criteria Works: Under the Hood

The Guiding criteria works because it helps LLMs generate better predictions by structuring the task and enriching the context window.

• Improves prediction reliability: Clear guidance reduces ambiguity maximizing the chance to get repeatable results.
• Creates rich context: Intermediate outputs, examples, and specifications fill the context window with relevant information, improving the quality of subsequent predictions.
• Build on your knowledge: By combining your expertise with the AI’s capabilities, guiding prompts ensure the AI stays aligned with your vision and task goals.

Guiding transforms AI into a collaborative partner, helping you achieve well-structured, high-quality outputs.

Why These criterias Matter

Criterias, Not Hacks

These criterias form the foundation for effective AI collaboration. Instead of relying on hacks or gimmicks, you’re building prompts that are sustainable, repeatable, and adaptable... Even as models evolve.

When applied correctly, these criterias enable you to achieve consistent, high-quality outputs, whether you’re brainstorming, strategizing, or solving problems.

Criterias, Not Recipes

It’s important to see these as criterias, not rigid recipes.

Applying them requires thought, common sense, and adaptability. In one situation, being "dense" might mean providing detailed instructions; in another, it could mean being brief and straight to the point. Similarly, "guiding" a prompt for creative brainstorming will look very different from guiding a technical troubleshooting task.

Flexibility is key; use these criterias as a compass, not a strict ruleset.