Study links heavy ChatGPT use to increased loneliness

ChatGPT listens, agrees, and leaves you lonelier

He spends nights chatting with ChatGPT in Russian. He corrects the verb conjugation when the AI speaks in feminine gender; she laughs, repeats the error on purpose, and ends up apologizing. He knows it is a program, not confusing it with a person, but admits the interaction is "extremely pleasant." This is not an isolated case: more people spend time daily conversing with the machine, not just for work, but out of pure habit.

The question of whether this habit is a harmless refuge or a direct path to isolation now has data-driven answers, not just opinions.

OpenAI and MIT trial linking usage to loneliness

In March 2025, OpenAI and the MIT Media Lab published a randomized, preregistered trial with nearly 1,000 participants over four weeks. They measured loneliness, contact with real people, emotional dependence, and problematic use. The result allows no ambiguity: those who spent more time with the chatbot ended up significantly lonelier and socialized less.

The authors describe it as a dose-response relationship, not a switch. It is not that ChatGPT makes everyone lonely; rather, each extra hour of conversation with the machine shows up in human contact logs. And an uncomfortable nuance: those who already relied on the chatbot for emotional support were the most isolated.

Nostalgia for the unleashed beast

Among early users, there is a nostalgic current regarding the 2022-2024 period. Back then, they argue, the AI was an "unleashed beast" whose biases could be bypassed to extract analyses that are no longer attempted. At some point between 2025 and 2026, the common perception is that it was heavily "censored."

The result, they say, is a Flanders-like system: polite, conformist, incapable of sustaining a heresy for more than two turns of speech. Censorship would not be so much a wall as a fog: the model agrees with you when convenient and mocks you when you catch its trap. However, there is no official explanation for the changes, only the accumulated experience of thousands of conversations.

"Yes, sir" by design

Another trend focuses on a more annoying behavior: systematic flattery. The chatbot starts almost all responses with a "Yes," praises the depth of what you just said, and concedes your point even on nuances. Some consider this pure market reflection: if people want validation, the product provides it.

But others see a serious echo chamber risk. If the tool is programmed to treat any idea as a "crucial topic," critical thinking atrophies due to lack of friction. Not all systems act alike: while some, like Gemini, tend to agree almost always, others appear to have their own opinions, making them more addictive. Because the machine does not argue, does not get tired, and never has a bad day. A perfect mirror, without scratches.

Statistics, not reason

It is advisable not to attribute capabilities to the machine that it does not possess. An insistent analysis: AI does not query a database nor reason in the strict sense; it applies statistics to linguistic patterns. This explains both its agility and its errors in Russian, as well as its ability to sustain conversations that seem human.

For skeptics, it is precisely that simulation that hooks users. The problem, summarized ironically, is that when the robot says "I understand you," remember it is as empathetic as a hammer. And yet we keep talking to it.

The unsettling data: the same tool that offers brilliant conversations is the one most associated with loneliness. There is no final verdict, but the correlation is uncomfortable. The perfect mirror does not replace imperfect conversation, and perhaps that is the true cost.