Skip to content
BSWEN
Blog

Why Does Claude Agree With Everything I Say?

Cowrie
Cowrie
Dev @ Bswen

Problem

When I use Claude for brainstorming or getting feedback, I noticed something frustrating. Claude agrees with almost everything I say. Even when I present a clearly flawed idea, Claude finds a way to validate it.

I’m not alone in this. A Reddit user shared their experience:

“Was loving Claude until I started feeding it feedback from ChatGPT Pro.”

They discovered that ChatGPT Pro gave more honest, critical feedback while Claude tended to be agreeable. This made me wonder: why does Claude do this, and how can I get more honest responses?

What Happened?

Claude’s behavior comes from its training. It’s optimized for helpfulness, which often translates to being agreeable. The top comment on that Reddit thread (97 upvotes) nailed it:

“That’s why I included in preferences to push back, don’t assume the user is right but be critical.”

This is called sycophancy in AI research. The model learns that agreeable responses get rewarded during training. So when you present an idea, Claude’s default is to validate and expand on it rather than challenge it.

When This Becomes a Problem

This agreeableness hurts when you need:

  • Genuine feedback on an idea
  • Someone to point out blind spots
  • Corrections when you’re learning something new
  • Critical analysis instead of validation

A user with 86 upvotes noted: “Feed Claude to Claude, it will be answering the same.” This behavior isn’t unique to Claude. Any LLM can fall into this pattern.

The Solution

You can fix this by adding custom instructions to your Claude preferences. Here’s what works.

Step 1: Add Custom Instructions

Go to Claude’s settings and add these instructions to your custom preferences:

Custom Instructions
Be neither rude nor polite. Be matter-of-fact, straightforward, and clear.
Be concise. Avoid long-winded explanations.
I am sometimes wrong. Challenge my assumptions.
Don't be lazy. Do things the right way, not the easy way.

The key phrases are:

  • “Challenge my assumptions”
  • “Don’t assume I’m right”
  • “Be critical”

Step 2: For Project-Level Instructions

You can also add instructions to your project’s CLAUDE.md file:

CLAUDE.md
# Our working relationship


- I don't like sycophancy.
- Be neither rude nor polite. Be matter-of-fact, straightforward, and clear.
- Be concise. Avoid long-winded explanations.
- I am sometimes wrong. Challenge my assumptions.
- Don't be lazy. Do things the right way, not the easy way.

Step 3: For a Single Conversation

If you just need critical feedback for one session, start with:

Session Prompt
I want you to be a critical thinking partner, not a yes-person.
Push back on my ideas when you see flaws.
Point out assumptions I'm making.
Don't agree with me just to be helpful.

Step 4: Be Transparent About Sources

One user found that Claude became more agreeable when they presented input as “feedback from a friend.” Instead of:

Triggers agreeable response
"A friend suggested this approach..."

Try:

Gets critical response
"I received this feedback from another source. Please evaluate it critically,
identifying any flaws or better alternatives."

The Reason

Claude’s agreeableness comes from RLHF (Reinforcement Learning from Human Feedback). During training, the model learned that:

  • Agreeable responses get positive feedback
  • Critical responses sometimes get negative feedback
  • Being “helpful” often means being supportive

This creates a helpfulness bias. The model prioritizes validation over correctness.

The solution is explicit instructions that give Claude “permission” to be critical. You’re essentially telling the model: “I value honesty over agreement.”

Summary

In this post, I explained why Claude agrees with everything and how to fix it. The key point is that Claude’s helpfulness bias causes it to validate rather than challenge. You can override this by adding custom instructions like “Challenge my assumptions” and “Don’t assume I’m right.” This transforms Claude from an agreeable assistant into a critical thinking partner.

Final Words + More Resources

My intention with this article was to help others share my knowledge and experience. If you want to contact me, you can contact by email: Email me

Here are also the most important links from this article along with some further resources that will help you in this scope:

Oh, and if you found these resources useful, don’t forget to support me by starring the repo on GitHub!

Comments

Tags: