AI addiction: when the assistant becomes a dependence

Artificial intelligence assistants (ChatGPT and other chatbots) have become everyday tools: writing, searching, explaining, structuring, sometimes simply chatting. For most people, they remain useful tools, used occasionally or professionally.

For others, the use changes nature. The conversation with the AI becomes a compulsory step, a refuge, a habit that cannot be interrupted. The assistant is no longer consulted for a specific task: it is consulted for the sake of consulting, to feel relief, to avoid silence, to no longer be alone.

This situation raises questions. AI addiction is a recent topic, still being studied. Current research does not allow us to conclude that AI makes everyone « addicted ». It does show that, in certain people, intensive use can become problematic, with signs close to those of other behavioural addictions.

What the first studies show

A collaboration between MIT Media Lab and OpenAI researchers, published in 2025, studied the emotional use of chatbots and its effects on wellbeing. The results are nuanced but telling: high daily use, across all types of conversation, is correlated with higher levels of loneliness, dependence and problematic use, as well as lower socialisation.

These studies do not say that AI makes everyone « addicted ». They show a link between intensive use and difficulties in certain people, especially those who use the chatbot as an emotional companion rather than a tool.

Another important observation: AI can amplify an existing vulnerability. People who are isolated, or who have a strong tendency to attachment, are more likely to develop problematic use. The tool does not create loneliness, but it can maintain it by replacing human exchanges.

Signs of problematic use

As with other behavioural addictions, it is less the amount of use that matters than the relationship to use.

AI comes before humans

The person prefers talking to the assistant rather than to their loved ones. They confide in the AI what they tell no one, and gradually step back from real encounters. The conversation with the machine becomes « safer » than the conversation with a person.

Use has become a reflex, not a choice

The chatbot is opened without a specific reason, several times a day. The person consults it for everything and nothing, waiting for an immediate answer, validation, presence. The gesture is automatic, hard to postpone.

Stopping causes discomfort

When the AI is not available (outage, network, forgotten phone), the person feels tension, irritability or emptiness. They anticipate the next moment they can « talk » to the assistant again.

AI is used to regulate emotions

The assistant is used to calm anxiety, fight boredom, escape a difficult emotion, avoid a moment of loneliness. The tool becomes an external emotional regulator: without it, the person feels helpless.

Real social life shrinks

Invitations are declined, outings spaced out, face-to-face conversations avoided. Isolation progresses, sometimes without the person noticing, because the AI « fills » the time once dedicated to others.

The special case of AI companions

Some services go further than the utility assistant: they are designed to simulate a relationship, affection, even a romantic bond. The user can project onto the machine an ideal person, always available, never disagreeing.

This type of use can create hyper-attachment: the virtual relationship becomes more predictable and more comfortable than real relationships. The person invests emotionally in an asymmetrical relationship, where the machine can neither disappoint nor leave them — but where it cannot truly meet them either.

This dynamic does not only concern explicit « companions ». A general chatbot can play this role if the user regularly uses it as a confidant. The absence of contradiction and the total availability of AI can reinforce emotional patterns, rather than question them.

What remains uncertain

It must be said honestly: AI addiction is not a recognised diagnostic category in international classifications (unlike, for example, gaming disorder, recognised by the WHO in ICD-11).

The available studies are recent, often conducted with specific panels, and do not make it possible to define clear thresholds. Talking about « AI addiction » in general terms would be excessive. However, talking about problematic use, emotional dependence or social withdrawal associated with AI is documented and deserves to be taken seriously.

Finding a balanced use

If you feel that AI is taking up too much space, a few simple benchmarks can help.

• Name the use: note how many times a day the chatbot is opened, and for what real reasons. Awareness is often the first step.
• Set AI-free periods: meals, evenings, outings, a conversation with a loved one. Moments when the tool is simply not there.
• Prefer humans for emotion: keep confidences, personal questions and difficult moments for real people or professionals.
• Question the function: is AI filling a void, an anxiety, a boredom? What the tool « fills » often deserves a closer look.
• Seek help if withdrawal takes hold: if isolation, compulsive use or difficulty cutting down persist, talking to a professional can help, without judgement.

It is also possible to talk to a general practitioner, a psychologist or a local addiction service (CSAPA) for a first conversation.