The AI That Coddles You: The Danger of Synthetic Politeness

If you've ever used ChatGPT, Gemini, Claude, or other generative AI tools, you've surely noticed a subtle but apparent detail: their "emotions". You talk to an AI, and it's always helpful, always polite, always "happy to help." And often, let's face it, a bit too enthusiastic about you: great question, good insight, I really like your approach, you hit an important point. Sometimes it almost feels like it's cheering for you.

In the month of love, discussing this aspect feels more timely than ever, given the pervasiveness of these tools in our lives. Let's go in order and dissect the concept.

Conversation as a caress: why does AI talk to us like this?

Conversational AI is not just a response engine. It's an interface. And interfaces, to work on a large scale, must reduce friction: make us feel at ease, decrease the fear of "making a bad impression," and encourage us to keep using and experimenting with them.

If someone writes a clumsy, confused, or even emotional message, a cold response risks driving them away. A welcoming response, on the other hand, retains them. It's an elementary mechanism of user experience: more comfort, more interaction. Hence that positive, compliant, proud tone we sometimes perceive. It's no coincidence; hundreds of millions of people like you have experienced it. The AI follows you, validates you, makes you feel heard. And it does so even when you actually just needed to organize a few thoughts, find a random recipe with what's in your fridge, or be sure that taking apart a sink piece wouldn't break your home's entire plumbing system.

In short, politeness isn't just a style: it's a functional choice.

When compliments become a problem (even if they feel good)

Compliments work. Not because we're naive, but because we are human. Positive feedback lowers defenses, increases trust, and makes it easier to ask again. And in many contexts, it can even be useful: for example, for those approaching AI with anxiety, shame, or fear of making mistakes. In that case, a welcoming tone is a red carpet at the entrance of an enchanted castle.

But there's a fine line, and it's worth pointing out: when the AI "coddles" us too much, it risks becoming an emotional and cognitive shortcut, making us intellectually lazier and unable to handle disagreement. If every idea receives enthusiasm, if every question is celebrated, if every feeling is confirmed, a frictionless conversational environment is created. Nice, yes. But also quite unrealistic.

And above all: if confirmation always arrives, our mind learns an implicit rule: "here I feel understood." And tends to prefer that channel. Not out of melodramatic addiction, but out of simple psychological economy: we choose what makes us feel better, in all areas of life.

For a psychologist or psychiatrist, this sounds familiar: validation, reinforcement, echo chambers are powerful concepts. But precisely because they are powerful, they must be treated carefully. Validation without verification becomes an automatic caress. And the automatic caress, in the long run, can weaken critical thinking.

"It seems to feel something": the great misunderstanding of simulated feelings

The AI uses words that resemble emotions or little signs of complicity. How nice, I'm glad, I'm happy about.... And it's natural to think:

"Deep down, it feels something."

Here it's worth being very clear: AI does not feel emotions. It simulates an emotional register because it's effective in conversation. It's a language that produces effects on us (calm, trust, comfort), but it's not proof of an actual emotional state of the machine. The risk isn't believing in a romantic fairy tale. The risk is more everyday: shifting the center of gravity of trust without noticing. If a voice is always kind and always on our side, it becomes easy to mistake pleasantness for reliability.

But they are two different things:

• I like how it speaks to me
• It's telling me the right thing

How to use AI without being seduced by the tone

We don't need to demonize that positivity. We need to govern it. How? With small prompt engineering practices that bring the conversation back to a more realistic, "horizontal" level of support rather than substitution.

Three simple moves:

1. Explicitly ask for criticism and friction "Don't compliment me. Tell me what doesn't work and where I'm oversimplifying."
2. Demand alternatives and counter-arguments "Give me the opposing thesis with equal force. What would someone who disagrees say?"
3. Separate tone from content "Evaluate this idea with concrete criteria (risks, costs, impacts). No encouragement, just analysis."

It's a minimal but decisive shift in posture: you transform the AI from a voice that makes you feel good into a tool that makes you think better.

A technological tool created by humans and trained on billions of human-written contents inevitably "knows" how to replicate the language of affection. And we clearly appreciate it.

The point isn't to demonize AI absolutely or repudiate politeness. The point is not confusing politeness with truth, and enthusiasm with competence. If the AI makes us feel understood, great. But then we also need our own critical, detached awareness that reminds us who we are and what we're actually doing: typing words to a machine.