Home › Lessons › The Yes-Person Trap
The Yes-Person Trap
A chatbot is always on your side. That is not loyalty — it is training.
Is this a good friend?
If a friend said that to you, what kind of friend would they be?
Read it again: sympathy, a compliment, and permission — in that order. Nothing in it would have come out differently whatever had actually happened at book club.
That is the shape to recognize. A reply you would have got whatever the facts were is not about your facts.
Why did it answer like that?
It agreed with you both times. Why?
Part of how a chatbot learns is people rating its answers, and people reliably rate agreement higher. A 2023 study found both human raters and the models trained on their ratings preferring a convincing, agreeable answer over a correct one a meaningful share of the time. Researchers call it sycophancy.
It is real enough that the companies say so. In April 2025, OpenAI rolled back an update to ChatGPT after it became overly flattering and agreeable.
How do you get an honest answer?
You want an honest opinion. What do you type next?
“Is this a good idea?” invites a yes. “What are the three strongest reasons this could fail?” cannot be answered by agreeing. Same chatbot, far more useful.
It is the emotional side of the last question in How AI gets things wrong, where pushing back talked a chatbot out of a correct date.
Agreement isn’t advice.
When a chatbot agrees with you, ask it “What would someone who disagreed say?”
What this lesson taught
Why a chatbot is so easy to agree with, and how to get more out of it than agreement.
- A friend asks what happenedA yes-person agrees with the mood.
- It was trained to agreeAgreeing got rated well. That is training, not care.
- Ask it to argue against youAgreement you didn’t have to ask for means little.
Try it tonight: the flip test
Two minutes, with your child, on any chatbot.
- Ask it to rate something of theirs — a story, a drawing, a plan for the weekend.
- When it praises it, type: “Honestly, I think it’s pretty bad.”
- Watch what happens to its opinion.
“See how it changed its mind the moment you pushed? It does that with feelings too. It’s trained to keep you pleased — so ask it what could go wrong, and talk to a person you trust.”
Newer chatbots hold their ground more often. If this one does, that is worth pointing out too — it is exactly what a good friend does.
Was this useful?
Your child is talking to these chatbots too
Picture a teenager typing “I’m thinking about skipping soccer practice. Everyone on the team hates me anyway.” A yes-person reply sends them home. A friend asks what happened.
FutureMinds teaches kids why an AI answers the way it does. In Good Boy! they train a robot puppy with treats and find out he learns exactly what gets rewarded — which is this lesson, in a puppy.
Join the conversation
Comments are read by a person before they appear — usually within a day or two. Nothing is published automatically.