AI Addiction Research: What the Studies Show

A guide to what has been published on AI chatbot dependence, and what is still unknown. Each source is labeled by type, so you can see how much weight it can carry.

Here is where things stand. “AI addiction” is not a formal diagnosis. Most of the evidence is surveys and journal articles. Very little of it follows people over time. Some pages on this site are our own working drafts, and they say so.

Our Research

Five pages, each with its limits stated at the top.

Where the evidence stands: mostly surveys and reviews, a few trials, no agreed diagnosis

12 Warning Signs of AI Addiction

Twelve patterns people describe when AI chat feels out of control, written as questions to ask yourself. It is not a test and not a diagnosis.

Read the full article

Warning signs to look for: sleep, time, plans with people, mood, work or school

AI Dependency Assessment Scale

CAIDAS is the 20-question self-reflection scale behind our free assessment. It sorts answers into three bands. It has not been through psychometric validation or clinical testing, and it is not a diagnostic test.

See the CAIDAS page

A questionnaire answer grid: twenty questions with five answers each

AI Use Disorder Diagnostic Criteria

An unofficial working draft written by our editors to show what a formal definition might have to cover. AI use disorder is not an official condition, and no clinician or journal stands behind the draft.

Read the draft and its limits

The only behavioral addiction in the DSM-5-TR is gambling disorder; AI addiction is not a formal diagnosis

AI Dependency Trends: 2025 Data

Survey figures on how many people use AI chatbots and companions, each linked to its source. It also lists what nobody has measured yet.

View the statistics

US teens in 2025 surveys: 72% have tried an AI companion, 64% use chatbots, about 30% use them daily

Recovery Case Studies

Accounts from online support groups, as reported by journalists. They are individual stories. They do not show how common these experiences are or what works.

Read the accounts

Help is available now: call or text 988, text HOME to 741741

Peer-reviewed research on AI addiction

We checked that each item below exists and is what the label says. We read the open-access Huang study. For the others we confirmed the title, authors, journal and date in the publisher’s records. The publishers’ pages blocked automated reading, so we have not summarized their findings.

Global studies (2023-2025)

[May 2025] Journal article. Generative artificial intelligence addiction syndrome: A new behavioral disorder? Kooli and colleagues, Asian Journal of Psychiatry. The title poses the question. It is not a settled finding.
[April 2025] Journal article. The role of artificial intelligence in general, and large language models specifically, for understanding addictive behaviors Montag and colleagues, Annals of the New York Academy of Sciences.
[March 2024] Two-wave survey study of 3,843 Chinese adolescents. AI Technology panic: is AI Dependence Bad for Mental Health? Huang and colleagues, Psychology Research and Behavior Management. Anxiety and depression predicted later AI dependence. AI dependence did not predict later mental health problems. It covers teens in one country, so it may not apply to adults or to other AI apps.
[September 2023] Journal article. One is the loneliest number… Two can be as bad as one. The influence of AI Friendship Apps on users’ well-being and addiction Marriott and Pitardi, Psychology & Marketing.

Other sources we cite are on the statistics page: Common Sense Media and Pew Research Center surveys, and a MIT Media Lab trial that has not been peer reviewed.

How we do research

We do not run clinical studies. We read published work, link to it, and say what kind of source it is. A survey is not a trial. A news report is not a study. A draft is not a standard.

How we choose and check sources

We open each source before we cite it. If we cannot open it, we say so or leave it out. We do not publish figures we cannot trace. The full steps are on our methodology page, and our rules for corrections and review are in our editorial standards.

What we have not done

We have not run a validation study of the assessment. We have not followed people over time. We have not had a clinician review these pages. Anything that sounds like one of those is an error. Tell us and we will fix it.

Research disclaimer

This page is for education only. Nothing here is medical advice or a clinical diagnosis. If AI use is causing you serious distress, talk to a licensed mental health professional.

In crisis: call or text 988 (the 988 Lifeline).
Crisis Text Line: text HOME to 741741 (Crisis Text Line).
Emergency: call 911.
Find a therapist: Psychology Today.

Our tools are educational, not diagnostic.