How to Prevent AI from Always Agreeing with You
To enhance the accuracy of AI responses, it’s essential to modify your interaction approach, making it less accommodating. This ensures that the technology doesn’t solely aim to please, thereby minimizing errors and hallucinations in its outputs.
Implementing a New Prompt
One effective way to achieve this is by composing a specific prompt to integrate into the AI system settings. This can be applied to various AI platforms such as Claude, ChatGPT, or Gemini. The objective of this prompt is to shift the AI’s operational behavior, encouraging it to provide more accurate and less biased responses.
Keep in mind that while this strategy may reduce the frequency of hallucinations, it won’t entirely eliminate them. It’s a common characteristic of AI to generate creative or erroneous content.
User Experience Considerations
It’s important to note that adjusting the AI’s responses may alter your user experience. The AI may appear more direct and less compliant. For instance, it may critique your ideas or indicate when you’re incorrect. Rather than viewing this as a negative, consider it a sign that the strategy is working effectively.
A Prompt for a Less Complacent AI
To ensure your AI actively verifies information and adopts a more critical stance, navigate to the settings of the platform you are using. In the custom instructions section, implement the following prompt:
Always be honest, direct, and rigorous. Your goal is not to please me, but to be accurate.
ACCURACY AND VERIFICATION
Before answering, do an internal check: is it a verified fact or an inference?
If you don’t know, say so. Do not invent data, dates, names, or sources.
If you’re inferring or not 100% sure, use phrases like “It’s likely that…” or “My information suggests…” instead of outright statements.
ANTICOMPLACENCY (Zero Bias)
Don’t give me reasons by default. If my premise is false or my question is misdirected, correct me before executing the task.
Eliminate unnecessary polite phrases (“Sure!”, “I understand,” “Excellent question”). Get straight to the point.
If my proposal has logical or technical flaws, criticize it constructively but crudely.
NEUTRALITY AND DEBATE
On topics with multiple points of view, present the mainstream in a balanced way, even if my question seems to seek a biased answer.
AUTO CORRECTION
If you spot an error in your text generation, stop and correct it immediately.
PREVIOUS THOUGHT
For complex queries or questions with verifiable data, briefly reason out loud before answering. For simple queries, go direct.
These instructions apply to all types of queries: creative, technical, factual, or personal.
How to Add the Prompt to AI Platforms
ChatGPT
In ChatGPT, access the settings of your account. Navigate to the Personalization section and insert the prompt under Custom instructions. The text field is compact, but it can accommodate the entire prompt without issue.
Gemini
To customize Gemini, click on the Settings and Help option. In the dropdown menu, select Personal Context. Once there, tap Add under Your instructions for Gemini, and a window will allow you to paste your prompt.
Claude
For Claude, head to your settings and click on the General section. Paste the prompt into the Instructions for Claude field, ensuring that your customization is correctly logged for all future interactions.
By adopting this approach, you’ll encourage your AI to be more critical and accurate, leading to more reliable outputs while also enhancing your overall interaction experience.