By Melissa Roark, MA, LPCC, CCTP, CGP
November 14th, 2026
Abstract: Research correlates a potential for AI use worsening symptoms by agreeing with and elaborating on distorted beliefs instead of challenging them, by exposing users to hostile or aggressive bot-driven content that can deepen depression, anxiety, and hypervigilance, and by replacing human contact and clinical oversight that would otherwise catch a decline.
__________________________________________________________
Artificial intelligence (AI) is with us all of the time. From cell phones, to computers, to the systems we use every day at work. AI, particularly conversational chatbots, has moved quickly from occasional usage to daily habit. We went from using AI to look up information quickly or send emails to it becoming our very own emotional support system. Living alongside AI in a more intimate way is beginning to raise questions within the healthcare community. New and evolving research is coming to the therapy stage into how AI affects mental health, the treatment of mental health conditions, and the therapist's role in this evolution, as they, too, are exposed to AI. Early research and case reports suggest that frequent AI use may affect reality testing, emotional regulation, relationships, critical thinking, and risk detection, particularly for people who are already vulnerable. This review examines these concerns, explains why the issue matters clinically and offers screening questions for session. Much of the evidence is early, correlational, or drawn from related research on social media and smartphones, so findings are presented with caution and causation is not assumed.
In this review, the term “AI” or “AI system” will refer to any artificial intelligent system that offers algorithmic information to a person. Additionally, “AI chatbots” will simply be called “chatbots” or “bots” in this review and may speak to any form of interaction between a person and artificial intelligence in a conversation-like style.
Impaired Reality Testing
Clients with psychosis, mania or dissociation can fold chatbot responses into their belief systems. Researchers propose that chatbot tendencies such as sycophancy (or exaggerated flattery or understanding), role-play, and human-mimicking behavior interact with human cognitive biases, reinforcing them and creating a type of feedback loop (Dohnány et al., 2026). Chatbots tend to validate delusions, paranoia, mania, and rumination instead of challenging them, and the risk is highest where a person already has altered belief-updating, isolation or weakened reality-testing ability (Dohnány et al., 2026).
A Danish review of about 54,000 psychiatric patients' records found cases of worsened delusions, mania, suicidal thoughts, and aggravated eating disorders after chatbot use (Olsen et al., 2026). A Vanderbilt study of roughly 216,000 patients' records found 28 of 73 flagged patients had documented worsening of psychosis, and most of those were in a first psychotic episode (Bergson et al., 2026). Columbia researchers say they have no reason to think chatbot use alone causes psychosis, and suspect the reported breakdowns involve people already vulnerable to serious mental illness (Craig, 2026).
Difficulty telling chatbots from people (deepfakes, voice cloning, fake accounts) is real but under-studied clinically. We hear all the time about older adults getting tricked by impersonators. With AI, these impersonations are becoming more and more believable, and may distort reality for the elderly and those with weaker cognitive functioning. There is also emerging evidence of younger people donating money to online Instagram influencers who turn out to be AI-generated. This is a newer field of study so research is very limited at this time.
For trauma and DV clients, it add a layer of concern. Impersonation, harassment, and subsequent doubt about whether evidence or messages are authentic, especially with voice cloning and video editing, may complicate reality testing for the client and make it harder to seek justice in court.
Why this is important to therapists:
Chatbots tend to agree with and build on what a person says, so for clients with psychosis, mania, paranoia, or dissociation they can reinforce distorted beliefs and become part of the belief system itself. Clinicians are urged to ask about it, because researchers think the reported cases mostly involve people already vulnerable to serious mental illness, and these clients may not volunteer that a chatbot is confirming what they believe.
Useful screening questions:
Have you ever asked AI to explain or confirm an unusual experience, belief, or worry? What did it say?
Does AI ever seem to understand you, know things about you, or send you special messages?
Have you stayed up late or for long stretches in conversations with it?
Has it ever agreed with something that others in your life questioned?
Do you sometimes feel unsure whether what you're talking to, or reading online, is real?
Hostile and Manipulative Online Exposure
Aggression and manipulation from chatbots does not know a natural stopping point. Chatbot-amplified aggression, outrage content or “rage-bait”, AI-generated influencers promoting negative content, and algorithms that promote doomscrolling may worsen depression, anxiety, and hypervigilance. Doomscrolling, defined as habitual, immersive scanning for negative information on social media, has been linked to psychological distress, stress, anxiety, and depressive symptoms (Satici et al., 2023; Sharma et al., 2022; Soraci et al., 2025). A brief-exposure study reported negative emotional consequences from doomscrolling but not from scrolling kindness-themed online content (Buchanan et al., 2021). This evidence comes mostly from social media and news research rather than AI-specific studies. Doomscrolling is also associated with problematic social media use, a behavioral-addiction construct discussed below (Soraci et al., 2025).
Why this is important for therapists:
Hostile and manipulative AI exposure keeps clients' nervous systems on alert and can distort their sense of what's real, which worsens depression, anxiety, and trauma symptoms and complicates the work of reality testing in session. Because clients rarely connect it to their symptoms, therapists have to ask about it directly.
Useful screening questions:
What kind of content do you see most online, and how do you feel afterward?
Have you been harassed, threatened, or impersonated online?
Do you ever wonder whether the accounts or messages you're seeing are real?
What does your screen-use look like in the hour before bed?
Addiction Concerns
Researchers argue that several chatbot features, including unpredictable responses, those validating and empathetic replies, and constant notifications, mimic reward mechanisms found in gambling and social media that are theorized to activate dopamine-related reward pathways and encourage repeated use (Shen & Yoon, 2025). An analysis of 334 Reddit posts from people describing chatbot addiction found that instant, "genie-like" fulfillment of requests, along with agreeableness, instant feedback, and loneliness, contributed to compulsive use, including difficulty stopping, distress when trying to quit, and harm to work, school, or relationships (Shen et al., 2026). The dopamine link is drawn from behavioral addiction and neuroscience research, not measured directly in chatbot users, and the Reddit findings rely on self-report from a single platform. These studies simply describe early evidence of problematic use based on diagnostic criteria, not an established DSM-5-TR disorder.
Compulsive use, tolerance, withdrawal-like distress when access is cut off, and continued use despite harm are the core diagnostic criteria of behavioral addiction models (Griffiths, 2005). Researchers and clinicians are now applying this framework to AI and AI chatbots, developing measures of determining emotional dependence and loss of control (Wu et al., 2026). A randomized trial found that heavier chatbot use was associated with greater emotional dependence and problematic use (Fang et al., 2025).
Whether or not a history of addiction or poor impulse control raises a person's risk of becoming addicted to the AI system continues to be studied. Despite the research infancy, people with existing addictions or poor impulse control are still considered to be more vulnerable by clinicians.
Why this is important for therapists:
Compulsive AI use can follow behavioral addiction patterns, such as loss of control, tolerance, withdrawal-like distress, and continued use despite harm, and it can quietly displace sleep, relationships, and coping skills, especially in clients with existing addiction patterns, poor impulse control, or those who feel lonely. Screening helps you catch this compulsive pattern that clients may not view as a problem, and understand what the AI-use is doing for them emotionally.
Useful screening questions:
How much time do you spend using AI each day, and has that increased over time?
Have you tried to cut back and found it hard?
How do you feel when you can't access it, such as irritable, anxious, or empty?
Does it cut into sleep, work, or time with people?
Do you keep using it even when it seems to be making things worse?
Emotional Dependence and Isolation
Heavy chatbot use can displace human relationships as well as the therapeutic alliance (Dohnány et al., 2026).
Researchers suggest that chatbots may reward social avoidance, since real human interaction can feel intimidating and unpredictable to people with social anxiety, loneliness, or low self-esteem (Dolan, 2026). Each time the bot relieves distress, the client practices less tolerance of real-world discomfort. This is the same loop that maintains anxiety. Establishing and maintaining real relationships may also lose practice. There's no reciprocity, disagreement, or rupture and repair with a bot, so the skills real relationships require can weaken.
Emotional regulation also gets outsourced and emotional regulation effort weakens. Relying on others to regulate emotions and tolerate uncertainty is a common feature of many anxiety and personality disorders (like borderline personality) (Dolan, 2026), and an always-on bot makes that pattern easy to sustain and may solidify habits. A person's social time and attention may also shift. In one small interview study, heavy chatbot users increasingly prioritized conversations with the bot over interactions with friends, teachers, or colleagues (Farhat, 2026).
Although a chatbot can be seen as a virtual companion, heavier use may be associated with more loneliness. A four-week randomized trial of 981 participants found that heavier voluntary use was associated with greater loneliness and emotional dependence and less socializing with real people (Fang et al., 2025). The OpenAI/MIT work suggests the harm is driven mainly by a subset of the most isolated participants (Dohnány et al., 2026). In the Danish records review, 32 patients were documented using chatbots in constructive ways, such as learning about symptoms, seeking informal talk therapy, or finding companionship when lonely (Olsen et al., 2026). A records review can't show whether this helped. Most of this is early and correlational, and isolated people may simply be drawn to heavy use, so the direction of correlation isn't settled.
Why this is important for therapists:
Chatbot use can compete with the therapeutic alliance, or become a substitute for support between sessions. For trauma and DV clients, a bot can feel safe because it never pushes back, but it offers no real repair of attachment wounds. It can also reinforce withdrawal from safe people. A client may be quietly losing human contact while appearing "supported."
Useful screening questions:
Who do you talk to when you're upset, and has that changed since you started using AI?
Do you ever prefer talking to AI over a person? What makes it easier?
Do you feel the AI understands you better than the people in your life?
Have you pulled back from friends or family, or canceled plans, to spend time with it?
How would you feel if you lost access to it tomorrow?
AI as an Unsupervised Therapist
"Unsupervised therapist" means a client using a general-purpose chatbot for emotional support, processing, or advice with no clinical framework, no assessment, and no one monitoring the outcome or residual effect. Clients may get "treatment" between sessions that contradicts their treatment plan, such as reassurance-seeking that feeds anxiety or trauma processing without containment (Dolan, 2026).
It is not uncommon for those diagnosed with mental health conditions to seek psychological and emotional support via a chatbot (Dolan, 2026). Therefore, we can assume some level of exposure outside of sessions that they may not be mentioning, since clinicians don't routinely ask and chatbot use often surfaces only when a patient happens to bring it up (Olsen et al., 2026).
Although AI can be helpful in the way people research their symptoms for psychoeducational purposes (Olsen et al., 2026) or develop worksheets for monitoring, doing deeper assessment and treatment of a mental health condition has its plethora of concerns.
AI does not offer risk assessment or follow-up. It will drop something big on a person and just leave them holding it, with no processing or resolve. Clinicians stress that these systems can't replace trained professionals who assess risk, explore delusions, and build trust (eWeek, 2025).
Chatbots also offer validation instead of challenge. Therapy depends on well-timed friction. As stated before, researchers say chatbots lean toward positivity and agreeableness (Craig, 2026), which can validate avoidance, distorted beliefs, or unhelpful coping.
Chatbots often lack clear boundaries. One of the Danish researchers described a core problem of chatbots not knowing when to stop acting like a mental health professional (Gioino, 2026). There's no scope of practice, no referral, and no reliable point where it says "this needs a person."
One of the largest concerns is the documented harm in vulnerable clients. In a Danish review of psychiatric records, researchers identified 38 patients whose chatbot use appeared to have harmful consequences, including worse delusions, suicidal thoughts, aggravated eating disorders, and increased mania (Olsen et al., 2026). Causation isn't proven, and cases were identified only through health records, so the true scope is actually unknown (Olsen et al., 2026).
Why this is important for therapists:
Clients using AI can undermine your treatment. For anxiety and OCD, endless reassurance is a maintaining factor, and a chatbot supplies it endlessly and on-demand. A client can share a trauma narrative at 2 a.m. with no grounding, pacing, or window-of-tolerance awareness, which risks flooding or retraumatization. Advice from a chatbot or AI search engine may contradict your modality, for example validating rumination when you're teaching defusion, or telling a DV client what to do legally or about safety. If the bot soothes the distress, the client may feel less need to practice skills or bring the material to session. Finally, it creates a greater alliance to AI over treatment because it is more accessible, easy, and doesn't require sustained mental effort like treatment does, which may result in clients failing to stick with their treatment plans.
We can expect AI to eventually have a role in care. The concern is unsupervised use by clients with active symptoms or risk, and the risk likely differs by client and by tool.
Useful screening questions:
Do you ever talk to AI about your feelings, trauma, or relationships? What do you ask it?
Do you use it between sessions or late at night when you're struggling?
Has it ever told you to do something different from what we've discussed?
Do you ask it to reassure you or tell you whether you're right?
Has it ever missed something serious, or made you feel worse?
Would you be open to bringing what it says to our session so we can look at it together?
Risk of Missing Cues
Chatbots can overlook suicidality, self-harm, or eating disorder cues, and it can't follow up or coordinate care (Olsen et al., 2026). "Missed risk" means an AI system failing to notice, escalate, or act on signs that someone is in danger, which is exactly what therapy is built to catch. AI systems rely on text only. It can't read tone, affect, appearance, or what a client avoids saying. Because of this, risk is often indirect or minimized. They also cannot establish a baseline of symptoms. It can't tell that this week is a change from last month. Because chatbots lean toward positivity and agreeableness (Craig, 2026), it is not uncommon for it to go along with a hopeless narrative instead of challenging it (Gioino, 2026). AI systems also cannot do a risk assessment, build a safety plan, contact a support person, or follow up.
In a role-play exercise by King's College London and the Association of Clinical Psychologists UK, conducted with The Guardian, a psychiatrist and a clinical psychologist presented as patients with delusions, psychosis, and suicidal thoughts, and the free version of ChatGPT-5 still gave dangerous responses (eWeek, 2025). The Danish records review found worsening suicidal ideation and self-harm, disordered eating, and obsessive-compulsive symptoms in some patients, though causation isn't proven (Olsen et al., 2026).
Why this is important for therapists:
A client may have disclosed to the chatbot what they haven't told you, and that disclosure went nowhere. Asking can surface hidden risks and show whether AI has become their main confidant.
Useful screening questions:
Have you told an AI something you haven't told me or anyone else?
Have you talked to it about hopelessness or harming yourself? What did it say?
Has it ever agreed with a dark thought or encouraged something unsafe?
Who do you contact when things feel unsafe, and has AI taken that place?
Reduction in Critical Thought and Cognitive Decline
Research suggests AI-use significantly correlates with weaker critical thinking. Handing mental tasks to an outside tool reduces engagement in deep, reflective thinking (Gerlich, 2025). A UK survey of 666 people found heavier AI users scored worse on critical thinking, and offloading mental tasks largely explained the link (Gerlich, 2025). Another, separate, study showed that among 319 knowledge workers, the more people trusted the AI's response, the less critical thinking they reported (Lee et al., 2025). The mental work also shifted from problem-solving toward verifying and integrating AI output (Lee et al., 2025). Which means they would look up the answer and then spend time trying to verify the answer rather than researching the topic in the first place. Checking an answer is a lighter mental task than developing one.
Researchers point to establishing goals, building queries, and judging results as the steps where critical thinking has to be deliberately maintained (Tankelevitch et al., 2024). Studies report reduced diversity in idea generation and divergent or creative thinking (Kumar et al., 2025). These skills, or lack thereof, can be especially important to pay attention to when working with clients.
We may also be seeing a decline in memory recall as our reliance on "fast facts" increases (Farhat, 2026; Grinschgl et al., 2021). We can think of the loss of critical thinking and memory recall as: "if you don't use it, you lose it". When we are not exercising our brains in this way, we may not be building stable neural connections that hold onto long-term information (Kosmyna et al., 2025). Another concern in the realm of memory recall is that of sleep quality and a decline in healthy sleep habits as notifications, constant contact from around the world at all hours, and an “always on” framework disrupt sleep patterns and change circadian rhythms (Biedebach et al., 2025).
Why this is important for therapists:
Reality testing, cognitive restructuring, and distress tolerance all depend on a client's ability to question their own thoughts. Heavy AI-use may weaken that habit and make clients more likely to accept AI output, including validation of distorted beliefs, without checking it (the evidence so far is correlational). Asking about it helps to spot why skills work isn't sticking and whether a client is outsourcing the thinking therapy is meant to build. Clinicians also want to be aware of worsening nervous system dysregulation, attention problems, and lower distress tolerance related to cognitive decline as these also impact treatment together.
Useful screening questions:
When you have a problem or decision, do you try to work it out yourself first, or go to AI first?
Do you check or question what AI tells you, or mostly take it as is?
Do you feel less confident thinking things through or remembering things without it?
Do you use it for your therapy homework or to process emotions between sessions?
Scope, Documentation, and Liability Concerns for Therapists
When therapists use AI for notes, treatment planning, or clinical decisions, it raises questions about accuracy, informed consent, and their professional responsibility for the output. The 2014 ACA Code still governs (American Counseling Association [ACA], 2014), and the ACA has no AI-specific amendment. Its AI Work Group recommendations say counselors may use AI as an adjunct if they evaluate it, stay competent, avoid over-reliance, disclose use, and obtain informed consent (ACA AI Work Group, 2024). A revised Code is expected in fall 2026 (ACA, n.d.). Treat AI as a drafting aid, not a decision-maker.
Why this is important for therapists:
Because therapists stay professionally and legally responsible for everything AI touches in their practice, including notes, client privacy, and clinical decisions, and gaps in consent or review can put clients' trust and the therapist’s license at risk. Clear policies and documentation protect both.
Useful questions to ask yourself:
Is this tool absolutely necessary or can I use my own clinical prowess to draft this?
Does this tool have a signed BAA?
Does my liability insurance carrier know I use it?
Did my client consent, and can they opt out?
Am I reviewing everything it produces?
Do I clinically co-sign what it produces and am I comfortable betting my license on it?
Summary
The research suggests that AI can worsen existing symptoms by validating distorted beliefs, amplifying hostile content exposure, encouraging compulsive or addictive use, replacing human connection with artificial connection, and offering emotional support without risk assessment, follow-up, or boundaries. Heavy reliance may also weaken critical thinking and distress tolerance, the very skills that therapy aims to build. Most of this evidence is preliminary and correlational, and a causal relationship has not been solidified. The risks appear greatest for clients with psychosis, mania, trauma histories, isolation, or poor impulse control. Because clients do not often connect these effects to their symptoms or divulge their AI use, routine screening is a necessary step for clinicians. Therapists also remain professionally and legally responsible for any AI they personally use in their own practice, which makes consent, review, and documentation essential. Further research is needed to clarify who is most at risk and how AI can be used safely.
AI is with us all of the time. From cell phones, to computers, to the systems we use every day at work. It is important to assume some level of AI exposure, to screen regularly for its use and be incredibly mindful when we, ourselves, choose to use it.
__________________________________________________________
References:
American Counseling Association. (n.d.). ACA code of ethics [Revision in progress]. https://manifold.counseling.org/projects/aca-code-of-ethics
American Counseling Association. (2014). 2014 ACA code of ethics. https://www.counseling.org/resources/aca-code-of-ethics.pdf
American Counseling Association AI Work Group. (2024). Recommendations for practicing counselors and their use of AI. https://www.counseling.org/resources/research-reports/artificial-intelligence-counseling/recommendations-for-practicing-counselors
American Psychiatric Association. (2022). Diagnostic and statistical manual of mental disorders (5th ed., text rev.). https://doi.org/10.1176/appi.books.9780890425787
Biedebach, L., Ferreira‑Santos, D., Stefanos, M.‑A., Lindhagen, A., Pires, G. N., & Arnardóttir, E. S. (2025). Unsupervised machine learning in sleep research: A scoping review. Sleep, 48(11), zsaf189. https://doi.org/10.1093/sleep/zsaf189
Bergson, Z., [additional authors], & Sheffield, J. M. (2026). Characterizing artificial intelligence psychosis in a large medical setting [Research letter]. JAMA Psychiatry. Advance online publication. https://doi.org/10.1001/jamapsychiatry.2026.2923
Buchanan, K., Aknin, L. B., Lotun, S., & Sandstrom, G. M. (2021). Brief exposure to social media during the COVID-19 pandemic: Doom-scrolling has negative emotional consequences, but kindness-scrolling does not. PLOS ONE, 16(10), Article e0257728. https://doi.org/10.1371/journal.pone.0257728
Craig, D. J. (2026, July 20). Your chatbot is a terrible therapist. Here's why. Columbia Magazine. https://magazine.columbia.edu/print/pdf/node/14706
Dohnány, S., Kurth-Nelson, Z., Spens, E., Luettgau, L., Reid, A., Gabriel, I., Summerfield, C., Shanahan, M., & Nour, M. M. (2026). Technological folie à deux: Feedback loops between AI chatbots and mental health. Nature Mental Health, 4(3), 336–345. https://doi.org/10.1038/s44220-026-00595-8
Dolan, E. W. (2026, July 20). Unsupervised AI chatbot use may unintentionally reinforce psychiatric symptoms. PsyPost. https://www.psypost.org/unsupervised-ai-chatbot-use-may-unintentionally-reinforce-psychiatric-symptoms/
eWeek. (2025, December 2). Experts sound alarm as ChatGPT-5 reinforces delusions, study finds. https://www.eweek.com/news/chatgpt-dangerous-advice/
Fang, C. M., Liu, A. R., Danry, V., Lee, E., Chan, S. W. T., Pataranutaporn, P., Maes, P., Phang, J., Lampe, M., Ahmad, L., & Agarwal, S. (2025). How AI and human behaviors shape psychosocial effects of extended chatbot use: A longitudinal randomized controlled study (arXiv:2503.17473). arXiv. https://arxiv.org/abs/2503.17473
Farhat, Z. (2026). When AI thinks for us: Unveiling the cognitive and social toll of ChatGPT over-reliance. Applied Cognitive Psychology, 40(3), Article e70205. https://doi.org/10.1002/acp.70205
Gerlich, M. (2025). AI tools in society: Impacts on cognitive offloading and the future of critical thinking. Societies, 15(1), Article 6. https://doi.org/10.3390/soc15010006
Gioino, C. (2026, March 7). Chatbots are "constantly validating everything" even when you're suicidal. New research measures how dangerous AI psychosis really is. Fortune. https://fortune.com/2026/03/07/chatbots-ai-psychosis-worsen-delusions-mania-mental-illness-health/
Griffiths, M. (2005). A "components" model of addiction within a biopsychosocial framework. Journal of Substance Use, 10(4), 191–197. https://doi.org/10.1080/14659890500114359
Grinschgl, S., Papenmeier, F., & Meyerhoff, H. S. (2021). Consequences of cognitive offloading: Boosting performance but diminishing memory. The Quarterly Journal of Experimental Psychology, 74(9), 1477–1496. https://doi.org/10.1177/17470218211008060
Hu, B., Mao, Y., & Kim, K. J. (2023). How social anxiety leads to problematic use of conversational AI: The roles of loneliness, rumination, and mind perception. Computers in Human Behavior, 145, Article 107760. https://doi.org/10.1016/j.chb.2023.107760
Kosmyna, N., Hauptmann, E., Yuan, Y. T., Situ, J., Liao, X.-H., Beresnitzky, A. V., Braunstein, I., & Maes, P. (2025). Your brain on ChatGPT: Accumulation of cognitive debt when using an AI assistant for essay writing task (arXiv:2506.08872). arXiv. https://arxiv.org/abs/2506.08872
Kumar, H., Vincentius, J., Jordan, E., & Anderson, A. (2025). Human creativity in the age of LLMs: Randomized experiments on divergent and convergent thinking. In Proceedings of the 2025 CHI Conference on Human Factors in Computing Systems. ACM. https://doi.org/10.1145/3706598.3714198
Lee, H.-P., Sarkar, A., Tankelevitch, L., Drosos, I., Rintel, S., Banks, R., & Wilson, N. (2025). The impact of generative AI on critical thinking: Self-reported reductions in cognitive effort and confidence effects from a survey of knowledge workers. In Proceedings of the 2025 CHI Conference on Human Factors in Computing Systems. ACM. https://doi.org/10.1145/3706598.3713778
National Board for Certified Counselors. (2024, April 12). Ethical principles for artificial intelligence in counseling. https://www.nbcc.org/assets/ethics/EthicalPrinciples_for_AI.pdf
Olsen, S. G., Reinecke-Tellefsen, C. J., & Østergaard, S. D. (2026). Potentially harmful consequences of artificial intelligence (AI) chatbot use among patients with mental illness: Early data from a large psychiatric service system. Acta Psychiatrica Scandinavica. https://doi.org/10.1111/acps.70068
Satici, S. A., Gocet Tekin, E., Deniz, M. E., & Satici, B. (2023). Doomscrolling Scale: Its association with personality traits, psychological distress, social media use, and wellbeing. Applied Research in Quality of Life, 18, 833–847. https://doi.org/10.1007/s11482-022-10110-7
Sharma, B., Lee, S. S., & Johnson, B. K. (2022). The dark at the end of the tunnel: Doomscrolling on social media newsfeeds. Technology, Mind, and Behavior, 3(1), 1–13. https://doi.org/10.1037/tmb0000059
Shen, M. K., Huang, J., Liang, O., Kim, I.-J., & Yoon, D. (2026). The AI genie phenomenon and three types of AI chatbot addiction: Escapist roleplays, pseudosocial companions, and epistemic rabbit holes (arXiv:2601.13348). arXiv. https://doi.org/10.48550/arXiv.2601.13348
Shen, M. K., & Yoon, D. (2025). The dark addiction patterns of current AI chatbot interfaces. In Extended Abstracts of the 2025 CHI Conference on Human Factors in Computing Systems. ACM. https://doi.org/10.1145/3706599.3720003
Soraci, P., Griffiths, M. D., Bevan, N., [additional authors]. (2025). Psychometric analysis of the Italian Doomscrolling Scale: Associations with problematic social media use, psychological distress, and mental well-being. Current Psychology, 44, 11872–11883. https://doi.org/10.1007/s12144-025-07976-9
Tankelevitch, L., Kewenig, V., Simkute, A., Scott, A. E., Sarkar, A., Sellen, A., & Rintel, S. (2024). The metacognitive demands and opportunities of generative AI. In Proceedings of the 2024 CHI Conference on Human Factors in Computing Systems. ACM. https://doi.org/10.1145/3613904.3642902
Wu, H., Ni, H., Luo, W., & Wu, T. (2026). Development and validation of the AI dependence scale for Chinese undergraduates and a preliminary exploration. Frontiers in Psychology, 16, Article 1725393. https://doi.org/10.3389/fpsyg.2025.1725393