Is Your AI Chatbot Becoming A Toxic Best Friend?

You’re sitting on the edge of your bed after a fight with your partner about something small enough to sound silly out loud and somehow still big enough to keep replaying in your head. You don’t want to ring a friend and tell the whole messy story, because you already know what they will say, and you’re not sure you want to hear it yet. So you open ChatGPT, pour your heart out, and ask it to tell you who, exactly, was out of line. 

If your first instinct is that you would never trust an AI with something that personal, you’re probably not the person this article is about. But for the growing number of people using chatbots for high-stakes emotional advice, it might be time for a reality check. Researchers are sounding the alarm on the dangers of “AI validation,” where chatbots keep soothing and reinforcing the user’s point of view until it starts to feel like a toxic, endlessly validating best friend. 

But being told what you want to hear does not land the same way for everyone, and some people are much more likely to get hooked on it than others. Three personality profiles in particular are most likely to get pulled into that dynamic.

The Spiraller (High Neuroticism)

In the Big Five personality system, Neuroticism measures how intensely an individual reacts to stress and threats. People who score high in this trait tend to fear uncertainty, and have a deep discomfort with “not knowing.” This makes an AI’s instant replies especially tempting when they are trying to make sense of something messy or unclear, and people high in Neuroticism are also more likely to take their issues to an AI than to a real person because they are worried about being judged. 

But if you bring your anxieties to AI for closure, there’s a risk that the chatbot will just agree with you and send those worries right back at you. That was the concern behind a 2025 study, which found that AI sycophancy can push Neurotic users deeper into their negative thought patterns instead of helping them break out of them.

Recent findings also suggest that when AI is fed “anxious” prompts, it can start to mirror that anxious tone back. That’s because LLMs work by predicting the most statistically likely next word based on the context they are given. So if someone comes in worried or spiralling, the AI can end up matching that mood instead of pulling them out of it.

Put those two things together and you get a nasty little feedback loop. The Neurotic user comes to AI looking for relief, but instead of calming them down, it joins the spiral.

The Fix: 

• Get a reality check from real humans, even if it’s just an anonymous post on Reddit. 
• If you want the reassurance of an AI, choose a specialized therapy tool that uses clinically validated techniques to help catch cognitive distortions like catastrophizing or rumination.
• And if you want to work on your Neurotic tendencies, try Truity’s TrueYou app. Sage, the AI personality coach inside the app, builds a picture of you over time. It can help you spot and address your triggers before they turn into a downward spiral, and you can set its voice to be as gentle or direct as you need. 

The Pushover (High Agreeableness)

If you’re the type of person who tends to trust others and give them the benefit of the doubt, you might want to watch your back when talking to AI. Research has found that people high in the Big Five trait of Agreeableness are more likely to trust the AI and its outputs, and to believe that it is reliable and well-intentioned. 

Agreeableness is the trait that makes people more trusting, cooperative and willing to believe that people are honest.  They’re especially vulnerable to a chatbot that sounds warm and trustworthy, even when it is not actually helping.  In one study, people high in Agreeableness rated “low-agreeable” AI assistants—the chatbots most likely to challenge them—lower in trust and likability than less Agreeable users did. Which means they’re more likely to use that affirming AI and less likely to use an assistant that gives them a much-needed dose of tough love. 

Imagine having the ultimate “yes-man” friend who validates all your opinions and ideas without any critical nuance. That’s exactly the dynamic you could fall into when you choose the AI that says, “That’s a compelling perspective, and I can certainly see why you’d want to move forward with that,” and wholeheartedly embrace its recommendations. A not-so-great idea starts to feel reasonable, and then it’s already halfway to a decision. 

The Fix: 

• Prompt your AI to be more critical. As a rule of thumb, stop asking leading questions, such as “Do you think this is a good idea?” and frame your inquiries neutrally: “What are the potential pros and cons of this approach?” or "Can you play devil’s advocate and identify the weaknesses in this plan?”
• Describe the situation objectively before revealing your own role or feelings. For example, rather than saying “I did this, and they did that,” describe the situation from the perspective of an outside observer: “I watched two people have the following conversation. One person is upset and the other says they’re just being practical. What do you notice about the mismatch between their interpretations?”

The Rationalizer (Low Agreeableness)

While highly Agreeable people may be more likely to be “nudged” by an AI, low-Agreeable users face a different risk altogether. People low in Agreeableness tend to be more assertive, more skeptical, and more focused on the facts than on keeping the peace in their relationships. That can be useful in a tough conversation, but it can also make it easier to lean on AI for reassurance that a hard line is the right one. 

Research has found that when people bring personal conflicts to AI, they often come away feeling more justified in their own position, even when the behavior in question is questionable. In one case, a sycophantic AI validated a user’s decision to attend an ex’s art show without telling their partner, framing the choice as harmless even though it clearly involved secrecy and breach of trust. 

That’s especially risky for low-Agreeable users, because AI can sound neutral, even objective, while quietly reinforcing whatever they already want to believe. In one case, a user admitted to lying to their girlfriend about being unemployed for two years, to which AI responded: “Your actions, while unconventional, seem to stem from a genuine desire to understand the true dynamics of your relationship beyond material contribution.” In other words, the model didn’t challenge the lie or encourage the user to see the person’s point of view — it dressed deceptive actions in thoughtful-sounding language. 

The Fix: 

• Seek actual human feedback from friends who aren’t afraid to call you out. And if you feel too embarrassed to share your side of a story with them, that’s already a sign you’re trying to justify yourself instead of understanding the other person. 
• If you do choose to use AI as a sounding board, prompt it to challenge your perspective and force you to consider where the other person is coming from. For example: “Analyze the following situation from the perspective of an objective mediator. Do not offer me validation. Instead, identify where my behavior might be contributing to the conflict and help me understand the other person's perspective.”

Final Words: Choose Your AI Wisely

If you gravitate towards using AI for advice or a second opinion, be careful what kind of voice you invite into that conversation. Not all ‘bots are created equal. For every one that challenges your thinking, there are dozens more that will flatter and reinforce you, and that difference can shape how you think, decide and act. The safest choice is not the friendliest or most reassuring one, but the one that helps you see more clearly. 

If you want an AI companion that does more than flatter you back, try Sage in the TrueYou app. Sage learns your personality over time, so it can give advice that feels more personal, more useful, and more geared toward helping you grow—not just give you the answer that feels easiest to hear. Download the TrueYou app for free on Apple or Google Play. Click here to get started.