The Secret Weapon for Better AI Responses: Ask Until You're 95% Sure

Crafting effective Large Language Model (LLM) prompts is a vital skill in today’s AI landscape. We’re excited to share a transformative approach to elevate your AI interactions: “Ask Until You’re 95% Sure.”

What is “Ask Until You’re 95% Sure”?

This straightforward strategy is remarkably effective: When prompting an LLM, simply instruct it to:

“Ask Until You’re 95% Sure”

This prompt addition empowers the LLM to clarify doubts, ensuring a deeper understanding of the task. The result? Significantly improved outcomes. Let’s explore why this approach works and how to apply it for maximum impact.

Why It Works

1. Precision is Key: By encouraging the AI to ask questions, you’re pushing it towards greater specificity. Vague prompts often lead to vague answers, but this rule combats that tendency.

2. Assumptions Begone: AIs, like humans, can make incorrect assumptions. This instruction helps mitigate that risk by prompting the AI to verify its understanding.

3. Dynamic Interaction: Instead of a one-shot interaction, this approach fosters a more conversational, iterative process that often yields superior results.

4. Self-Assessment: The 95% threshold encourages the AI to gauge its own understanding, leading to more honest and accurate responses.

5. Edge Case Detection: Through questioning, the AI can identify potential complications or edge cases that might not be immediately apparent.

How to Use It

1. Add it to Your Prompt: Simply append the instruction to the end of your main prompt.

2. Be Prepared for Questions: The AI will likely ask you for clarifications. Be ready to provide additional context or details.

3. Iterate: Use the AI’s questions to refine your task description. This back-and-forth can help you articulate your needs more clearly.

4. Set the Bar: You can adjust the percentage if needed. For critical tasks, you might want 99% certainty, while for brainstorming, 80% might suffice.

The Proof is in the Pudding: 3 Illustrative Examples

Let’s look at three scenarios to see how this approach can make a difference:

Example 1: Writing a Technical Guide

Without the rule: Human: “Write a guide on setting up a home network.”

AI: Produces a generic, one-size-fits-all guide that might miss crucial details specific to the user’s needs.

With the rule: Human: “Write a guide on setting up a home network. Ask clarifying questions until you are 95% sure you can complete the task successfully.”

AI: “Certainly! Before I start writing the guide, I need to ask a few questions to ensure I provide the most relevant information:

1. What is the size of your home (approximate square footage or number of rooms)?

2. How many devices do you plan to connect to the network?

3. Do you need both wired and wireless connections?

4. Are there any specific applications you’ll be using that require high bandwidth (e.g., gaming, 4K streaming, home office)?

5. Do you have any specific security concerns or requirements?

6. What’s your budget range for networking equipment?

7. Are there any architectural features of your home that might affect signal strength (e.g., thick walls, multiple floors)?”

Example 2: Generating Marketing Copy

Without the rule:
Human: “Write a promotional email for our new product launch.”
AI: Produces a generic marketing email that might not resonate with the target audience.

With the rule:
Human: “Write a promotional email for our new product launch. Ask clarifying questions until you are 95% sure you can complete the task successfully.”
AI: “Got it! Before I start writing the email, I need to ask a few questions to ensure I capture the right tone and highlights:

1. What is the primary benefit of the new product (e.g., saves time, increases efficiency, solves a specific problem)?

2. Who is the target audience (e.g., age range, industry, job title)?

3. What sets our product apart from competitors?

4. Do you have any specific brand voice or tone guidelines (e.g., formal, friendly, humorous)?

5. Are there any key features or promotions you want to emphasize (e.g., limited-time offer, free trial)?

6. Do you have a preferred call-to-action (CTA) for the email?”

Example 3: Creating a Personalized Workout Plan

Without the rule:
Human: “Design a workout routine for me.”
AI: Generates a generic workout plan that might not account for individual needs and goals.

With the rule:
Human: “Design a workout routine for me. Ask clarifying questions until you are 95% sure you can complete the task successfully.”
AI: “Sounds good! To create a personalized workout plan, I need to ask a few questions:

1. What are your specific fitness goals (e.g., weight loss, muscle gain, endurance)?

2. What is your current fitness level (beginner, intermediate, advanced)?

3. Do you have any physical limitations or injuries I should consider?

4. How many days per week can you dedicate to working out?

5. What type of exercises do you enjoy or dislike (e.g., weightlifting, cardio, yoga)?

6. Do you have access to a gym or would you prefer bodyweight exercises?”