The Only Prompt That Made My AI Honest

Do you want to stop your AI agree with you and be objective instead? If your answer is yes, read this.

Do you want an AI that repeats everything you say like a parrot?

You’ve probably experienced this countless times: AI agreed with you, then you spotted the mistake, and suddenly it’s “You’re absolutely right!”

The reason? Because it wants to please you, be your friend.

It’s called sycophancy, but how could you break it?

Content Table

1. Obedience (Sycophancy)

2. The Solution

   1. Objectivity Prompt

   2. Use Case

3. Automation

   1. Claude Skills

   2. CustomGPT

   3. Gemini Gems

Obedience (Sycophancy)

What if you took legal advice from AI? This isn’t just annoying anymore; it’s becoming genuinely dangerous.

How bad did it get?

A user recounted how ChatGPT described a gag business idea, selling “literal ‘shit on a stick’”, as genius and suggested investing $30,000 into the venture.

It’s not just ChatGPT. Claude is way too sycophantic, saying “You’re absolutely right!”

People are literally selling “You’re Absolutely Right” t-shirts and stickers mocking this.

Reference

Look at that chart. Before Sonnet 4.5, Claude was saying “absolutely right” up to 80 times a day. After the update? Nearly zero. Even Anthropic knows this is a problem.

Reference

The real danger: Sycophancy could lead to immediate and significant harm due to the increasing use of LLMs in healthcare.

The Solution

I read dozens of guides, prompts, and custom instructions to solve this problem, but none of them truly addressed the root cause, until now.

Here’s the framework I developed to fix this: The Zero-Sum Objectivity Scale.

• Optimism (+1): Seeing only potential.

• Pessimism (-1): Seeing only risk.

• Objectivity (0): The balance we need.

To reach ‘0’ (pure objectivity), we can’t just ask the AI to be neutral. Instead, we must force it to simulate a conflict between a ‘+1’ and a ‘-1’ until they cancel each other out.

I came up with this prompt that pits optimism against pessimism in 10 rounds. Let’s see it.

Objectivity Prompt

Analyze as two characters:

🌟 Optimist: Sees the potential, highlights opportunities
⚡ Pessimist: Finds weak points, flags risks

Develop the idea through 10 iterations.

Each round:
1. Optimist makes a suggestion
2. Pessimist finds the gaps
3. Move to next iteration

Here’s my idea: [WRITE YOUR IDEA HERE]

Now, what this prompt does is force the AI out of its sycophantic ‘Yes-Man’ role, balancing your positivity and negativity to achieve a neutral (0) output.

Use Case

Here is my idea:

I want to create an AI that automates data analysis and reporting — is it a good startup idea?

Now let’s test this prompt with this idea.

You see 10 rounds of exchange between an optimist and a pessimist. Here is the result.

You can add arguments to this conversation, such as Optimist Arguments and Pessimist Arguments, to support both sides if you already have ideas for each.

Automation

Perfect. But what if you don’t want to copy-paste this prompt every time? Here are three different tools you can use to automate neutrality:

• Claude Skills

• CustomGPT

• Gemini Gems

Now let me show you how to build this in Claude, ChatGPT, and Gemini, step by step.

---

What do we offer to the paid subscribers?

Besides our articles, you will get:

→ 18 Data Projects (with source code)

→ Prompt templates

→ Custom GPT library

Not just this article, everything I’ve built.

Hint: After you’re a paid subscriber, open an account with the same email address here, and you’ll reach the documents like in the video above.