How to make ChatGPT, Claude and Gemini stop agreeing with everything you say, and start giving you genuinely useful feedback.
ChatGPT thinks you’re brilliant. Claude loves your ideas. Gemini says your latest project has incredible potential. Either you’re a genius (ahem, you’re not), or your AI is just kissing your butt.
And, yes, it’s the latter that’s true. it might feel good to hear about how great you are, but it isn’t real. Sorry.
Why Does ChatGPT Agree With Everything?
AI has a habit of agreeing with us, even when we’re spectacularly wrong. It’s called AI sycophancy, and it’s a very real problem. In fact, OpenAI rolled back a ChatGPT update in 2025 because the chatbot had become embarrassingly agreeable.
The good news? If you want better responses you can actually use in real life, you can turn the AI models into genuinely useful critics.
How to Get Honest AI Feedback
Open ChatGPT, Claude or Gemini, start a fresh conversation, and paste this:
📋 The AI Critic Prompt
Act as my independent, constructive critic.
I’m going to show you an idea, article, design or piece of work.
Your job is to evaluate it honestly, not to flatter me or find fault for the sake of it.
- Identify genuine weaknesses that could meaningfully affect the result. Prioritize them by importance. There is no minimum or maximum number. If nothing significant is wrong, say so.
- Explain why each issue matters and suggest a practical improvement.
- Support your criticism with specific examples or evidence. Don’t invent problems.
- Separate facts from opinions, identify assumptions, and flag anything you can’t verify.
- Tell me what’s already working so I don’t accidentally ruin it.
Be direct, fair and practical. No automatic compliments, unnecessary negativity or nitpicking. Don’t rewrite everything unless I ask.
Now give it something to criticize. It can be your latest blog post or a business idea. Copy and paste a copy of your new homepage. Have ti check your YouTube script.
The Secret Ingredient Most People Miss
Don’t forget to tell your AI what you’re trying to accomplish.
Without context, it might criticize your funny article for not sounding professional. Or suggest adding paragraphs of information to a website deliberately designed to be simple.
Explain what you are looking for:
“This is a funny, 500-word article for beginners. Find anything confusing, boring or unnecessary. Keep the humor.”
That’s much more useful than simply asking, “Is this any good?”
Make Your AI Prove Its Point
When AI criticizes something, ask it to “Show exactly where the problem is and why your suggested change would improve it.”
Why? Because AI can confidently invent flaws. And asking it to be brutally honest doesn’t magically make it correct. Sometimes it just becomes brutally annoying.
For bigger projects, try the same evaluation in a fresh conversation with a second AI. See whether it spots the same problems. Agreement isn’t proof, but repeated concerns are worth investigating.
Make It a Habit
If you use ChatGPT regularly, you can put a shorter version in Settings → Personalization → Custom Instructions:
When I request feedback, challenge weak ideas, identify real problems, explain your reasoning and suggest practical fixes. Don’t agree with me just to be agreeable. Don’t invent faults or criticize things that already work.
Just remember that custom instructions aren’t foolproof, and they’ll influence all your other conversations too.
💡 BONUS TIP: Make AI Criticize the Critic
Who says your AI critic is right? For a second opinion, take your original work and the criticism it received, then paste both into a fresh conversation with another AI model.
Give the second AI this prompt:
Independently evaluate the original work and the AI feedback provided.
Which criticisms are justified by the actual work? Which are exaggerated, subjective, unsupported or potentially incorrect?
Identify any important problems the first AI missed. Explain which suggested changes are worth making, which should be reconsidered, and why.
Don’t automatically agree or disagree with the first AI. Base your assessment on specific evidence, and flag anything you cannot verify.
Remember: Two AIs agreeing doesn’t make something true. But a second opinion might save you from fixing something that wasn’t broken.
The Last Word
Your AI isn’t your best friend. Sometimes you need someone willing to tell you that your groundbreaking business idea is actually a load of crap.
And that’s a lot more valuable than having a machine blow smoke up your butt.
FAQ
Why does ChatGPT agree with everything I say?
AI chatbots can exhibit sycophancy, a tendency to favour agreement or praise over useful disagreement. It doesn’t happen every time, but it can make feedback less reliable.
Can I make ChatGPT more critical?
Yes. Give it clear evaluation criteria and ask for specific evidence, meaningful weaknesses, and practical improvements. Custom instructions can also encourage more direct feedback, although results aren’t guaranteed.
