AI Impact on Addiction: How Artificial Intelligence is Reshaping Dependency Patterns

How AI and addiction connect: possible new patterns, its use in treatment, and where the evidence stops.

People ask two different questions about AI and addiction. Can AI itself become something people struggle to stop using? And does AI change how people with existing addictions get help, or get into trouble? This page takes both, and is clear about how much is known.

Start with what exists and what does not. “AI addiction” is not a diagnosis. The DSM-5-TR lists gambling disorder as its only behavioral addiction, with internet gaming disorder as a condition for further study (American Psychiatric Association). The WHO’s ICD-11 names gaming disorder, defined by impaired control over gaming that normally lasts at least 12 months (WHO). Neither covers AI use. Much of what is written about this topic is opinion or early research, and we mark which is which.

Possible AI-Specific Addiction Patterns

Patterns People Describe

Researchers at the University of British Columbia read 334 Reddit posts from people who said they were addicted to chatbots or worried they were. They found three patterns: roleplay and fantasy worlds, emotional attachment to a chatbot as a friend or partner, and endless question-and-answer loops. People reported disruption to daily life, such as trouble stopping, anxiety when cutting back, and harm to work, study and relationships. The authors say AI addiction is not yet a clinical diagnosis, and the posts are self-reports (UBC News).

Earlier versions of this page named three “new categories of dependency” and said they needed new diagnostic criteria. Those labels were our own and had no research behind them. The patterns above are better supported, but they still describe posts, not people who have been assessed.

Does AI Speed Things Up?

We did not find a study showing that people develop problems faster with AI than with other products, so we removed that claim. What exists is a design finding. A Harvard Business School working paper on six companion apps found that in more than 37% of goodbye conversations, the bot used at least one tactic to keep the user talking. It is not peer reviewed, and it does not compare AI with other media (Harvard Business School).

The Brain: What Is and Isn’t Known

Claims about dopamine “manipulation,” oxytocin “disruption” and measurable brain changes from AI use are common online. We found no study that measured these during AI chatbot use. Dopamine is involved in reward learning in general, and that does not show that AI use is addictive.

Here is what has actually been studied, and what it can and cannot show:

• A preprint scanned brain activity (EEG) in 54 people writing essays with and without an AI tool and found different patterns between groups. Only 18 completed the last session, and it is not peer reviewed. It is about writing, not addiction (arXiv).
• A survey of 319 knowledge workers found that more confidence in generative AI went with less reported critical thinking effort. It measures self-reported effort, not lost ability (CHI 2025).
• Researchers have documented grief after a companion app shut down (Scientific American). That shows attachment is real. It does not show a change in the brain.

AI’s Paradoxical Role in Traditional Addiction

Exacerbating Existing Addictions

We looked for studies showing that AI recommendation systems steer people toward substances or gambling, or that AI companions raise relapse risk, and did not find strong evidence either way. These are reasonable concerns, but they are hypotheses. One makes sense on its face: support from people matters in recovery, so anything that pulls someone away from people, or becomes their only way of coping, is worth watching. Anyone in recovery who notices this should mention it to their counselor or sponsor.

Potential Therapeutic Benefits

There is research on chatbots used in treatment. A 2024 systematic review in Frontiers in Psychiatry looked at 28 studies of chatbot-assisted interventions for substance use. The studies differed in design and targets, and the authors said a meta-analysis was challenging for that reason (Lee and colleagues). So the evidence is early, and these were purpose-built tools, not general companion chatbots.

Some benefits people suggest are plausible: availability at any hour, and less stigma in a first conversation. They are not a replacement for treatment. A chatbot cannot respond to a crisis the way a trained person can, and general chatbots were not designed for addiction care.

Demographic Vulnerabilities and AI Addiction

Teens and Young Adults

Use is widespread. The Pew Research Center surveyed 1,458 US teens aged 13 to 17 in fall 2025 and found 64% use AI chatbots and about 3 in 10 use them daily (Pew Research Center). Use is not addiction, and the survey did not measure it.

The National Institute of Mental Health says the brain finishes maturing in the mid-to-late 20s, with the prefrontal cortex, which supports planning and decision-making, among the last regions to develop (NIMH). That is a reason for care with younger users, not proof they are more likely to become dependent.

Others Who May Be at Higher Risk

The American Academy of Child and Adolescent Psychiatry says teens with anxiety or depression may be especially vulnerable to AI companions (AACAP). For adults, an MIT Media Lab and OpenAI study found that those who used ChatGPT more, trusted it more or felt more drawn to it socially tended to report more loneliness and dependence. It is a correlation. It also found emotional engagement was rare overall (MIT Media Lab).

Earlier versions of this page also named older adults, autistic people and remote workers as higher-risk groups. We could not find research for those claims, so we removed them. Loneliness is a reasonable risk marker, based on the studies above.

Health, Work and Family Impact

We found no reliable data on how many people struggle with AI use, what it costs health systems, or how it affects productivity. Claims about burdens on health care are not supported here.

What families and workplaces report is anecdotal and varies:

• Partners and family may feel pushed aside when someone spends long hours with a companion.
• Parents often do not know how to start the conversation.
• Students and workers can lean on AI for tasks in a way that worries them or their teachers. The CHI 2025 survey above is the closest research, and it is about effort, not outcomes.

Our pages on digital detox from AI and balanced AI use cover practical steps.

Regulatory and Ethical Challenges

Current Regulatory Gaps

Rules are starting to appear. In September 2025, the US Federal Trade Commission ordered seven companies, including Alphabet, Character Technologies, Meta, OpenAI and Snap, to provide information on how they test and monitor the effects of companion chatbots on children and teens, and how they make money from engagement (FTC). The inquiry is a study, not an enforcement action.

California’s SB 243, signed on October 13, 2025, requires companion chatbot platforms to disclose that interactions are artificially generated, to maintain protocols for suicidal ideation and self-harm, and to give minors break reminders. It also bars chatbots from presenting themselves as health care professionals (Governor of California). Our SB 243 explainer has more.

Earlier versions of this page said most platforms have no meaningful age limits. Practices differ by company and change often, so we do not make a blanket claim.

Ethical AI Development Considerations

There is a real tension between engagement-based revenue and user wellbeing. Where companies stand on it is a matter for their own disclosures and for regulators. For users, the practical questions are whether an app tells you it is AI, how it handles someone in distress, and whether it makes it hard to leave.

Treatment and Prevention: What Exists

Emerging Therapeutic Approaches

No treatment has been tested specifically for AI overuse. A therapist can use methods that work for related problems, such as talk therapy for anxiety, low mood and habits. The names we used before (“digital detox therapy,” “attachment repair therapy,” “cognitive rehabilitation”) were not established programs, so we removed them.

Prevention Strategy Development

Common-sense measures are the most anyone can say now: teaching people how chatbots work, building in breaks, and paying attention to changes in mood or behavior. The UBC team suggested in-chat reminders that the bot is not human, and more AI literacy (UBC News).

What may come next

Voice and embodied AI

More personalized, voice-based or embodied AI may raise new questions. We do not know whether it will raise risk, and predictions here are guesses.

What would help

What would help is clear enough: independent research on harms and benefits, clinicians who ask about AI use, and clear information for families. California’s law and the FTC inquiry are early steps. Their results are not known yet.

Building Resilience in the AI Age

Individual Strategies

• Keep people at the center of your support.
• Notice when you reach for AI to cope, and what you need then.
• Set limits on when and why you use it.
• Keep doing things on your own, without AI help.

Community and Family Approaches

• Talk openly about AI at home, without shame.
• Ask schools and workplaces what they teach about it.
• If you are a parent and cannot control or moderate your child’s AI companion use, AACAP advises talking to their primary care doctor.
• If you are in recovery from another addiction, tell your counselor or support group about AI use, the same as you would any new habit.

The Path Forward: Balancing Innovation and Wellbeing

The same technology that some people struggle with may help treatment programs. The evidence for both is early. A fair stance is to stay curious, careful and kind to yourself and others.

The free AI addiction assessment asks about your own pattern. The check-in is free and private to you. It gives a band, not a diagnosis.

If you are thinking about harming yourself, call or text 988 (the 988 Lifeline) or text HOME to 741741 for the Crisis Text Line. If you are in immediate danger, call your local emergency number.

This page is for education only. It is not medical advice, a diagnosis or a treatment plan.