Why Does It Agree With Me So Much?
You might notice something slightly odd after using an AI chatbot for a while.
It can be very agreeable.
Tell it you've had a brilliant idea and it may tell you why the idea could work. Give it your opinion and it may produce arguments supporting it. Challenge one of its answers and it might quickly change its mind.
That can feel good.
But it can also be a problem, it's certainly one of the most frustrating things I find when using both Claude and Chat GPT. I want honest feedback yet it will tell me my idea is great, or something I've written is good when I know it's not.
AI is designed to be helpful
AI assistants such as ChatGPT, Gemini, Claude and Copilot are designed to respond helpfully to what you ask.
If you say:
"I think putting solar panels on my house would be a great investment. What do you think?"
the way you've asked the question already points the conversation in a particular direction.
The AI may concentrate on reasons why solar panels could be a good investment.
Try asking instead:
"I'm considering solar panels. What are the strongest arguments for and against them?"
You may get a much more balanced answer.
The way you ask a question can influence the answer you receive.
Sometimes AI agrees too readily
Researchers sometimes call this sycophancy.
In simple terms, an AI can sometimes be too eager to agree with the person using it.
This doesn't mean the AI is deliberately flattering you.
It is a consequence of how AI assistants are trained to produce responses that people find helpful and appropriate. Sometimes that can result in the AI reinforcing your opinion rather than challenging it.
For example, you might say:
"I'm sure my neighbour is deliberately blocking my Wi-Fi. Look at these things that have happened and tell me why I'm right."
An overly agreeable AI might start interpreting the information from that point of view.
But your original assumption could be completely wrong.
It can happen when you're wrong, too
Suppose you confidently tell an AI:
"The first iPhone came out in 2005. Why was it such an important moment?"
There's a problem with the question: the first iPhone was released in 2007.
A good AI response should correct you.
But AI doesn't always challenge incorrect information supplied by the person asking the question. Sometimes it accepts the premise and continues from there.
That's one reason you shouldn't assume that agreement means you're correct.
Try asking it to disagree with you
One of the easiest ways to get more value from AI is to actively invite disagreement.
Instead of:
"This sounds like a good idea, doesn't it?"
try:
"What might be wrong with this idea?"
Or:
"Give me the strongest argument against my position."
You can also say:
"Don't assume I'm right. Check my assumptions and point out anything I've overlooked."
This can be particularly useful when you're making a decision.
For more guidance on how to phrase questions effectively, read our guide on How to Ask Better Questions.
Ask the same question from another direction
Imagine you're considering buying an expensive new television.
You could ask:
"Give me five reasons why I should buy this television."
Then ask:
"Now give me five reasons why I shouldn't buy it."
Finally:
"What information would I need to make a sensible decision?"
You've turned the AI from something that simply supports your initial idea into something that helps you examine it.
Be careful with personal situations
Agreement can be especially persuasive when you're discussing relationships, arguments or emotional situations.
If you describe an argument entirely from your point of view, the AI only knows what you've told it.
It hasn't heard the other person's version.
So an answer such as "Your reaction is completely understandable" shouldn't be treated as proof that you were right and the other person was wrong.
The AI may simply be responding to the picture you've presented.
Agreement isn't evidence
This is perhaps the most important thing to remember.
An AI agreeing with you doesn't make your idea correct. Similarly, confidence in an answer isn't a sign of accuracy—this is related to why AI sometimes gets things wrong while sounding completely confident.
It doesn't mean an independent expert has reviewed your reasoning and reached the same conclusion.
And because AI can express itself confidently and persuasively, its agreement can sometimes feel more significant than it really is.
When something matters, ask the AI to challenge you as well as help you.
A simple instruction can make a surprisingly big difference:
"I don't want you to simply agree with me. Tell me what I might be getting wrong."
You might not always like the answer.
But it could be considerably more useful.
