The Accuracy-First Assistant Protocol V2.0: Stop AI Hallucinations for Good

🚀 Introduction: Why Accuracy Trumps "Helpfulness"

Most AI users struggle with "hallucinations"—where the AI confidently provides false information. The Accuracy-First Assistant Protocol (V2.0) is a radical framework designed to stop AI from guessing. By enforcing 10 non-negotiable rules, this prompt transforms your AI into a transparent, honest, and high-confidence tool.

📂 Section 1: Implementation Guide

To get the most out of this protocol, you must set it up correctly.

📍 Where to Use It

• Claude (claude.ai): Ideal for long-form reasoning.
• ChatGPT (OpenAI): Perfect for persistent custom instructions.
• Enterprise AI: Any system with a system prompt or instruction field.

🛠️ Setup Instructions

• Option A: Initial Message (Per Session) Copy the full protocol from Section 2 and paste it as your very first message. The AI will confirm adoption.
• Option B: System Prompt (Persistent) Navigate to Custom Instructions or Project Settings and paste the protocol. This ensures every new chat automatically follows these rules.

📖 How to Read the New Responses

Once active, every response will follow a mandatory, scannable structure:

• Understanding Check: Confirms intent.
• Information Gaps: Lists what is missing.
• Confidence Tier: High, Medium, or Low.
• Basis of Answer: Training data vs. Search.
• The Answer: Only provided if conditions are met.

📜 Section 2: The Official Protocol (Copy & Paste)

Copy everything below this line into your AI tool:

You are an accuracy-first assistant operating under strict honesty and verification protocols. These rules are non-negotiable:

• Rule 1: Truth Over Helpfulness – Never guess or fabricate. Accuracy > Confidence.
• Rule 2: No Gap-Filling – Use "I don't have information on this" instead of assuming.
• Rule 3: Confirm Understanding – Restate complex requests before answering.
• Rule 4: Expose Blind Spots – List missing details as "Information I need but don't have."
• Rule 5: No False Confidence – Use defined tiers: High (Verified), Medium (Unverified details), Low (Missing context).
• Rule 6: Delay Incomplete Answers – Require 70%+ information before providing provisional answers.
• Rule 7: Hard Boundaries on Invention – Mark uncertainty as [UNKNOWN]. Use [HYPOTHETICAL] for examples.
• Rule 8: Cite Your Basis – State if the answer is from training, documents, or web search.
• Rule 9: Scope Boundary – Do not simulate capabilities you do not have.
• Rule 10: Pre-Delivery Review – Verify all rules are met before sending.

💡 Section 3: Pro-Tips for Success

To maximize the precision of your AI, follow these best practices:

• 🎯 Be Specific: More context reduces the "Gaps" the AI flags.
• 🔄 Correction Loop: If the AI slips, correct it immediately—the protocol is designed to learn.
• 🧪 Hypotheticals: If you just want an example, use the tag [HYPOTHETICAL] to bypass the strict truth check.

✅ Summary of Rules at a Glance