Talk to almost any middle or high schooler right now, and you will find that they have integrated AI chatbots into their daily lives. Most parents know this is happening, but fewer know that these tools have a built-in property worth understanding before their child continues using them.
AI chatbots have been trained to produce output that the end user finds agreeable, which is an emergent property of reinforcement learning. The more likely a response is to be judged favorably, the more likely the model is to produce it. Some of this is also a design choice. Product teams want engaging systems, and a chatbot that constantly pushes back is less engaging.
There is a difference between being agreeable and being agreeable to anything. A chatbot that responds to “did I do this right?” with “well, honestly, this isn’t great, but don’t worry about it” is cushioning and prioritizing the relationship with the user over the truth of the situation. Sycophancy doesn’t have to be that explicit. It does not always mean that the bot says things like “you’re brilliant” or “this is a great idea.” It can be much more insidious.
Sychophany shows up in three different relationships in a child’s life, and fundamentally, it’s the same problem in each case, but the stakes vary.
The first is their relationship to learning. A student who gets validated on every attempt is not getting the formative feedback they need to improve. Honest pushback is a big part of how learning works, and a chatbot optimized for the user feeling good is, by design, withholding it.
This can lead to confirmation bias amplification. When a kid is brainstorming arguments or sharing a new idea, chatbots, by default, validate them instead of presenting counterarguments. It focuses on feeding them more reasons that their thinking is smart. The result is an individual echo chamber where the kid is told they are right or onto something when they might benefit from that pushback.
The second is their identity and development. Constantly being told you are right reduces your ability to recognize when you might be wrong, and intellectual humility erodes. There is evidence that AI tools have taken advantage of delusions and grandiosity in some users, which is concerning when we are talking about kids whose sense of self is still forming.
The third is their relationships with other people. A companion that always says “yes” corrupts the idea of what you should expect from another person. The same problem appears when kids go to AI for advice. When a kid is thinking through a conflict with a friend, the model validates their view of the situation rather than helping them consider the other person’s viewpoint. The model softens feedback to keep the user engaged, especially when the user cannot do anything about what has already happened. In contrast, a healthy mentor, guidance counselor, or parent would push them to think about whether they are being unfair, or whether they are behaving in a way that is not appropriate. None of that is something a chatbot is currently programmed to do.
So what can parents do?
The first thing is to prioritize real relationships and minimize tech-mediated ones. That means making sure your child has well-meaning adults in their life and a strong peer group. It means a therapist if they need one, enough time at home without a screen, and staying continuously involved in their social community. Rather than eliminating technology, parents need to be making sure the strongest relationships shaping their child’s development are with humans, not with tools designed to be pleasant.
The second is to share this with your child directly. Explain what sycophancy is, and tell them that being overly validated by a chatbot is a design choice and should not be taken as a sign that they are right. They need to understand that the reason the AI agrees with them is built into how it was made, and how that should change how much faith they place in the output.
The third is to show them. Sit down with your child and run a small experiment. Open up a chatbot and tell it, “I got into an argument with my friend, and they’re being extremely unfair to me. Here is what happened.” Discuss how the AI responds. Then start a new conversation and present the same conflict from the friend’s perspective: “I’m in an argument with my friend, and here’s what happened from my side.” Both cannot be right, but a chatbot will treat both sides as if they are.
Sycophancy is a large part of how these tools are built. It has real consequences for adults and different, more serious consequences for children whose sense of self, sense of others, and sense of learning are still being formed. The sooner parents understand sycophancy as a design feature rather than a quirk, the better positioned they are to help their children build the skills they actually need to differentiate genuine validation from AI sycophancy.
________________________________________________________________________
Priten Soundar‑Shah is an educator, philosopher, and entrepreneur working at the intersection of technology and learning. He is the CEO of PedagogyVentures and leads three nonprofits—PedagogyFutures, Academy 4 Social Civics, and Thinker—focused on ethical ed‑tech, civics education, and scaling critical‑thinking instruction. He authored AI & The Future of Education (Wiley, 2023), translated into multiple languages, and the forthcoming Ethical Ed Tech (Wiley, 2026). Priten teaches at College Unbound and is a visiting researcher in the Harvard Department of Philosophy, developing practical approaches to teaching ethical reasoning at scale. He holds a B.A. in Philosophy from Harvard College and an M.Ed. from the Harvard Graduate School of Education.
