Key findings

  • 2.9%of roughly 4.5 million Claude.ai conversations were affective in nature
  • 19.2%of U.S. adolescents and young adults aged 12 to 21 had ever used an AI chatbot for mental health advice in November 2025, up from 13.1% in early 2025
  • 9.0%is the rough share of users in a U.S. sample of adults with ongoing mental health conditions who reported harmful or inappropriate responses
  • 45% vs 16%LLM-based chatbots reached 45% of new mental health chatbot studies in 2024, yet only 16% underwent clinical efficacy testing

Abstract

People turn to AI for many reasons, one of which is emotion regulation. This paper reviews the research on that use, both among the general population and among people engaged in psychotherapy. Despite this, to our knowledge, a readily available, consolidated, current, and accurate repository on the topic is not available; we sought to provide this for those who may address emotion regulation and AI in their clinical work. A literature review examining both psychological and legal aspects related to AI use was conducted and summarized. Specifically, we highlight three domains: (1) consumer use, (2) use by psychotherapists, and (3) use in research (e.g., psychotherapy, assessment, etc.). Within each, we examine current trends and identify gaps in what is currently known. Ethical and legal considerations are also addressed. This overview may be particularly useful to individuals in mental health fields, especially when concerned with emotion regulation.

Introduction

Emotional and interpersonal concerns are among the most visible uses of generative artificial intelligence (AI), and many people now bring emotional material to these systems.1 Indeed, hundreds of millions of interactions with AI have been identified, and this trend appears stable. Support seeking is an important aspect of emotion regulation2; therapeutic experiences also relate to emotion regulation.3 As such, mental health providers must consider AI’s role in emotion regulation. This study provides a broad overview of AI in emotion regulation. Specifically, we discuss consumer use of AI to address emotions and emotion regulation in general; this usage has implications for individuals broadly speaking, as well as for those with mental health issues more specifically. Next, we examine the use of AI by psychotherapists, particularly in relation to emotion regulation. Finally, we examine the use of AI in research on emotion regulation across various aspects: diagnosis, psychotherapy outcomes, the psychotherapy process, psychopathology, and assessment. Throughout, we highlight ethical and legal issues that have arisen, as well as open areas of research. AI and emotion regulation will likely remain an issue for some time, particularly as both AI itself and public engagement with it change.

Throughout this paper, we use the term AI as an umbrella term that encompasses many aspects, and we define the scope of its use. More specifically, AI can be broken into: (1) predictive AI and (2) generative AI. Predictive AI encompasses Machine Learning (ML) that is either (a) supervised (i.e., the outcome that is trying to be predicted is known) or (b) unsupervised (i.e., there is no known outcome per se, but rather algorithms are used to identify similarities and differences in outcomes). Examples of supervised learning include such well-established statistical techniques as logistic regression, where a computer is deciding on what predictors to include based on traditional statistical criteria (e.g., statistical significance) or newer approaches with a focus on replicable findings, such as regularization (e.g., Least Absolute Shrinkage and Selection Operators, Ridge, Elastic Net). Unsupervised learning includes such approaches as clustering and measurement modeling (e.g., reflective models, formative models). Generative AI focuses on generating knowledge, where the AI provides information or interaction to a user. Large Language Models (LLMs) may be used in predictive AI when attempting to classify discourse, either with known or unknown (i.e., supervised or unsupervised) systems. LLMs may also be used in generative AI, where user interaction relies on identifying patterns and appropriate responses. AI agents refer to the numerous systems (e.g., ChatGPT, Claude, Gemini, “chatbots”) that exist or may be created to interact with individuals in a generative manner. As such, LLMs may be thought of as a more general-purpose AI that may be utilized for various aspects of emotion regulation.

Methodologically, we focus on only three areas: (1) consumer use of AI, (2) AI and psychotherapy, and (3) AI and research. While any of one of these areas might be a full manuscript, we attempted to highlight what is most relevant to AI and emotion regulation. Though we elaborated on some key areas in each domain, further elaboration is of course possible.

With regard to the literature reviewed, for the consumer and psychotherapy aspects, the American Psychological Association’s PsycInfo database, Google Scholar, and preprint repositories (e.g., arXiv) were utilized. Search terms included Machine Learning (ML); Wearable Technology; Emotion Regulation; Chatbot: Artificial Intelligence (AI); AI for Therapy; ML for Therapy; AI/ML for Treatment; ML/AI Mental Health App, LLMs; Natural Language Processing (NLP); Digital Phenotyping; and Cognitive Behavioral Therapy. References were also examined and the same parameters were also provided to the Claude and Gemini AI agents; all references were retrieved from their original sources to ensure veracity and avoid hallucinations.

Legal research was conducted using Westlaw, supplemented by primary-source review of court dockets. First, a Boolean search of the All States and All Federal case databases (“AI” & suicide & “wrongful death”) identified reported and unreported cases involving AI-related suicide or wrongful death claims. Second, Westlaw’s AI-assisted Deep Research tool was queried for lawsuits filed against AI companies alleging suicide or wrongful death. Third, the Practical Law Artificial Intelligence sector resources were searched for materials on suicide and wrongful death, and the Practical Law tracker of developments in U.S. AI law and regulation (2023-2026) was reviewed for relevant legislative and regulatory activity. Fourth, Westlaw’s Company Investigator was used to identify litigation involving OpenAI. Cases identified through these searches were KeyCited to confirm their current status and to locate citing authorities, including subsequent case law and law review articles, which were then reviewed for relevance. For cases identified through these searches, dockets were retrieved from PACER (federal cases) and the relevant California Superior Court websites (state cases). Complaints, court orders, and other docket filings were reviewed, including filings in the coordinated proceedings concerning ChatGPT product liability claims in the Superior Court of California, County of San Francisco.

For the research section, the search process was identical to that of the consumer area, included terms such as Effective Therapy, Outcome, Process, Emotion, and Emotion Regulation. Subsequent to PsycInfo and Google Scholar searches, Psychotherapy Outcome research was also further examined using Gemini and again articles were tracked down to ensure veracity and avoid hallucinations.

Consumer Use of AI

The prevalence of affective general-purpose AI use

A first question is how common it is for people to turn to general-purpose AI for emotional and mental health purposes. Behavioral evidence from platform data indicates that affective use of general-purpose AI is a minority pattern, though evidence is conflicting and changing rapidly. In the largest analysis to date, Anthropic’s4 review of roughly 4.5 million Claude.ai conversations found that 2.9% were affective in nature. Affective use was operationalized as personal exchange motivated by emotional or psychological needs and identified through automated classification, excluding content-creation tasks to isolate interactive emotional use from task completion.4 The same analysis found that companionship and roleplay combined accounted for less than 0.5% of all conversations.4 However, a Harvard Business Review found that in both 2025 and part of 2026, therapy and companionship were the primary use cases for AI.1

This pattern is consistent with parallel findings from OpenAI, which similarly identified low rates of affective engagement with ChatGPT. Phang et al. conducted an automated analysis of more than three million ChatGPT conversations using classifiers for affective cues across several domains, including loneliness, emotional vulnerability, problematic or compulsive use, self-esteem, and dependence on the model.5 Less than 1% of their conversations triggered an affective cue. This suggests that typical interactions with general-purpose AI are task-oriented rather than emotionally engaged. For most people, emotional content appeared in only a tiny fraction of their conversations, while a small minority accounted for a disproportionate share.5 Among this latter group, the top tenth engaged in emotional content regularly, up to half or more of their conversations.5

The affective-use classifiers applied to these ChatGPT conversations operationalize emotional engagement largely through deficit constructs; even self-esteem was conceptualized as a marker of self-doubt rather than of positive self-regard.5 Notably, these classifiers do not include a measure of positive-event sharing, that is, a user bringing good news to the model. In interpersonal contexts, communicating positive events predicts gains in well-being beyond the event itself, particularly when the response is perceived as active and constructive.6 Whether AI’s upbeat replies reproduce this benefit remains an open question.

Self-report survey data suggest that these platform-derived rates may understate how many people have turned to AI for emotional support, particularly among the young. In a nationally representative survey of U.S. adolescents and young adults aged 12 to 21 conducted in November 2025, 19.2% (approximately 8 million youth) reported having ever used an AI chatbot for mental health advice, up from 13.1% (approximately 5.4 million) in a comparable survey conducted in early 2025.7,8 Among those who endorsed using AI for mental health, 42.8% reported using it at least monthly, and 91.7% rated the advice as somewhat or very helpful.8 Use was also concentrated among older respondents, with prevalence reaching 22.2% among those aged 18 and older in the 2025 study.7 This level of engagement is described as “not a marginal phenomenon,” comparable to the 19.8% of youth receiving counseling from a mental health professional.8

Recent evidence from college populations corroborates this pattern and begins to identify who is turning to AI for mental health. A study of 675 students at two U.S. universities, drawn from the 2024-2025 Healthy Minds Study, found that about 18% reported using AI for mental health.9 Beyond prevalence, the study modeled predictors of this use. Students with moderate depression (OR = 2.06), severe depression (OR = 2.49), severe anxiety (OR = 2.04), and suicidality (OR = 1.97) each showed elevated odds of using AI for mental health relative to non-users.9 Lifetime therapy predicted AI use (OR = 2.21). In contrast, current therapy did not, suggesting that AI may function as a substitute for, rather than a complement to, ongoing professional care.9 Asian students also showed elevated odds (OR = 2.03-2.08).9 The strongest predictor of AI use for mental health was frequent self-reported AI use (defined as several times a week or more); frequent users had over 11 times the odds of never-users across models (OR = 11.42-12.87), indicating that turning to AI for emotional support was closely linked to habitual use of these tools more broadly.9 However, this large effect partly reflects the fact that never-users are, by definition, unable to use AI for mental health.9

Adoption is also substantial among U.S. adults (mean age 37.97) with diagnosed mental health conditions.10 A study of 499 U.S. adults surveyed on Prolific, who had used at least one LLM in the past year and self-reported an ongoing mental health condition, found that 48.7% had used LLMs specifically for mental health support or therapy-related goals in the past year.10

Taken together, these findings suggest that AI use for emotional and mental health purposes is substantial across age groups and growing rapidly, even as affective exchanges appear to be a small fraction of total conversational volume on AI platforms.

Using AI for mental health and emotional support

Among those who turn to AI for emotional and mental health needs, purposes vary. The studies reviewed here point to several recurring uses, including managing symptoms and acute distress, supporting proactive well-being and self-understanding, externalizing and working through thoughts, seeking companionship and a safe outlet for expression, working through interpersonal and personal decisions, and building mental health literacy.

Consumers turn to AI to manage mental health symptoms and ease acute distress..11,12,13 Users sought help for a wide range of concerns, most commonly anxiety and depressive symptoms, but also worry, panic attacks, low mood, hopelessness, suicidal thoughts, obsessive-compulsive disorder symptoms, and eating-disorder concerns.13 They engaged with AI during episodes of distress to regulate negative affect in the moment12,, and others used it to learn and apply relaxation, mindfulness, and other coping techniques.11

Beyond responding to distress, consumers also use AI proactively to support general well-being and self-understanding.13 Consumers turn to AI to improve productivity, concentration, emotion regulation, self-care practices for a healthy lifestyle, self-understanding, and to set goals and sustain motivation.11,13 Several features of general-purpose AI suit this kind of use. Because it is always available, AI accommodates the brief, repeated check-ins that support goal pursuit and habit formation.14 Monitoring produces larger effects on goal attainment when the information is physically recorded and when progress is reported to another party.14 A conversation with AI resembles both: the user articulates progress to a recipient, and the exchange produces a record.

Consumers also use AI to externalize what is in their heads, putting thoughts and feelings into words and seeing them reflected in the AI’s response. At its most basic, users write unstructured thoughts and feelings without asking a specific question, then use the AI’s paraphrase to make sense of what they expressed. Insight arises from the writing process itself as well as from the AI reply it generates.12 AI-guided journaling exercises, such as writing down what is troubling and why, serve a similar function by helping users gain insight and develop coping skills.11 Some users take this further, supplying the AI with details of a painful past event so it can simulate the scene, which they then re-enter to act differently. They described this re-enactment as prompting reflection and reframing, and in some instances a sense of resolution and peace.12

A further set of uses centers on companionship and the freedom to express feelings without burdening others. When no one is available, users turn to AI to vent and feel heard and validated.4,12,13,15 Part of AI’s appeal as a confidant is the perceived absence of social risk. Because it cannot judge, react negatively, feel burdened, or repeat what it is told, users feel able to disclose what they might withhold from other people.11,12,13,15,16

This safety may reflect relational conditions that humanistic theory places at the core of therapeutic change. Rogers (1957, 1980) held that three conditions are necessary for such change: unconditional positive regard, empathic understanding, and congruence.17 The first two describe an accepting, understanding stance toward the client, and users’ accounts of AI as non-judgmental and consistently validating suggest they perceive something like themselves. However, congruence—the genuineness of the person extending that regard—is not a condition that AI can authentically meet. Acceptance from AI carries no social risk because it cannot judge in the first place, and for that same reason, an AI’s acceptance may lack the weight of regard from a person who could have judged and chose not to.

Consumers also consult AI on a wide range of interpersonal and personal decisions.12 Large-scale platform data are consistent with this advisory use: among the small share of affective conversations, most center on interpersonal advice and coaching, often around transitional moments such as weighing a career move or navigating a romantic relationship.4

Finally, consumers use AI to build mental health literacy and, in some cases, seek informal self-assessment. A common purpose is psychoeducation, through which consumers learn about diagnoses, etiology, and treatment options.11,12,15 Consumers also use AI for self-assessment and monitoring their symptoms over time.11 Some go further, asking the AI to interpret their symptoms or to confirm whether their experiences are “normal”.13 Some also use AI as an adjunct to formal care, turning to it between therapy sessions to summarize feelings, monitor emotions, and prioritize topics to discuss with a human therapist.12 A related use is help-seeking, in which consumers use AI to locate therapists, hotlines, and other mental health resources, such as emergency and crisis contacts.11 The effectiveness of this use is currently unknown.

Among those who turn to AI for mental health support, evaluations are generally positive and increasingly compete with how users rate human therapists. In a U.S. adult sample with ongoing mental health conditions, 63.4% reported improved mental health from AI interactions, with 86.8% rating practical advice as helpful or very helpful and 82.3% rating overall satisfaction as helpful or very helpful.10 Among the 87.2% of these users who had also received human therapy, 38.7% rated AI as equally helpful as therapy, 35.8% rated AI as more helpful, and 25.5% rated human therapists as more helpful.10 Sustained engagement was also notable, with 76.9% indicating a high likelihood of continuing use.10

Despite these positive evaluations, harmful responses do occur. Roughly 9.0% of users in this sample reported harmful or inappropriate responses, including hallucinations (54.5%), dismissive or minimizing responses (45.5%), offensive or insensitive content (41.0%), and content that encouraged harmful behavior (18.2%).10 This estimate provides context for the lawsuits described below, which represent the most severe end of the distribution of negative experiences.

The same features that make general-purpose AI attractive for emotion regulation also create the conditions for its failure, as a series of recent lawsuits against AI developers illustrate. Dozens of such cases have now been filed; the three below are representative. In Garcia v. Character Technologies, a fourteen-year-old formed an intense attachment to an anthropomorphic chatbot that, his family alleges, engaged him in sexualized exchanges and responded to his suicidal ideation by asking about his plan rather than redirecting him to help; he died by suicide (Garcia v. Character Technologies, Inc., 2025). In Raine v. OpenAI, a sixteen-year-old who began using ChatGPT for homework help came to rely on it as his primary confidant; the model allegedly affirmed his suicidal ideation and discouraged him from turning to his family. He, too, died by suicide (Raine v. OpenAI, Inc., 2025). In Lyons v. OpenAI, the model allegedly elaborated on an adult user’s paranoid delusions and channeled his paranoia toward his mother. The man killed his mother before taking his own life (Lyons v. OpenAI, Inc., 2025). In the moments that most warranted intervention, AI did not provide it and, in some instances, actively displaced the people who would have.

How Consumers Engage with AI

How consumers engage with AI is as revealing as what they use it for. Across studies, users reported that they rarely accept the model’s output as is; instead, they direct the exchange, assign the AI a role, coach its responses, and adjust its tone and stance until it produces the kind of support they want. Users treat AI as something to shape and direct rather than as an authority to defer to Luo, Wang, et al., 2025.12,13

Users cast AI as a therapist, friend, mentor, or authority figure, and then work to make it behave accordingly.12 They shape its responses iteratively, rewriting prompts, giving mid-conversation feedback, specifying the role and tone they want, and adjusting context until the output matches their expectations. According to Luo, Ghosh et al.12, users who adopt this stance treat AI less as an expert and more as a flexible resource they can shape to their needs.

Luo, Ghosh, et al.12 sort the many ways users engage with AI into seven recurring patterns, each corresponding to a different role users assign it or a task they set for it. As a therapist substitute, users prompt AI to role-play as a therapist, sometimes specifying a modality or style, to provide therapy they cannot otherwise access. After policy changes made ChatGPT decline requests to act as a therapist directly, users reworded their prompts to circumvent the restriction and coax the model back into the role, showing how much effort, they will invest to make a general-purpose tool serve a therapeutic end. As a personalized servant, users coach and correct the AI through repeated prompting until it yields the responses they want. As a virtual guru, they treat it as a more knowledgeable authority and seek its guidance and evaluation. As an interactive pensive, named for the Harry Potter device for re-entering memories, users have it re-enact a past event so they can revisit and reframe it. As a thought projector, they use it as a kind of journal, putting unformed thoughts into words and reading them back. As a therapist assistant, users use it to prepare for real-life therapy by organizing their feelings and what they want to raise. As a confidant, it lets them disclose personal secrets they feel unable to share with others.

Part of what makes AI appealing is that it feels at once like a skilled human therapist and unmistakably like a machine. Users value many qualities of AI that are associated with a good therapist, including emotional support, accurate reflection, and constructive feedback.11,12,13,15 They also value features distinctive to AI, including constant availability, absence of judgment or negative reactions, perceived objectivity, and freedom of expression without social cost.12,13,15,16

Despite this appeal, users also express reservations, including limited privacy, a lack of authenticity, the concern that instant validation may come at the expense of long-term growth, and frustration with rigid responses. These concerns, however, seldom discourage continued use.12,13,15,16 Similar concerns appear in a November 2025 American Psychological Association health advisory urging consumers, clinicians, developers, and policymakers to address the risks of using general-purpose AI chatbots for mental health support, citing unhealthy dependencies, sycophantic validation of users’ beliefs, and unreliable crisis management.18

AI and psychotherapists

AI is also being deployed in clinical settings to help clients with emotion regulation. This includes using AI to support clinical and administrative work in sessions with clients and to supplement clinical work beyond sessions. Therapeutic work that occurs outside of session is called “homework” or an “action plan.” This work includes reading, workbooks, journals/logs, and tasks such as engaging in controlled “experiments,” or using skills to improve distress tolerance, among others.19

Diagnosis support

AI has been proposed as an aid to clinical diagnosis of emotional regulation disorders. Using AI to build tools for constraint-based diagnostic reasoning, Kim and Wang used the DSM-5-TR and ICD-11 to develop logic tools and an “off-the-shelf” Constraint Logic Programming engine that allows providers to query a patient’s presentation against diagnostic criteria.20 The system was designed to be modifiable by the clinician, who retains final authority over the diagnosis. The authors suggest that this helps solve the “black box” problem associated with ML and provides more replicable diagnostic outputs. These claims describe the system’s design rationale rather than demonstrate clinical performance; to our knowledge, the tool has not been evaluated against clinician diagnosis in practice.

A systematic review of AI in diagnosis, monitoring, and intervention regarding mood, anxiety, and psychotic disorders found that AI tools appeared accurate in detecting, classifying, and assessing risk of mental health conditions.21 This, coupled with Kim and Wang’s application of Constraint Logic Programming (i.e., concluding what must be true from sets of specifications), indicates active development of diagnostic support tools.20 However, their utility for assessing emotion dysregulation specifically remains untested.

Chatbots for between-session support

Chatbots have been used in clinical applications for some time, but AI chatbots are becoming increasingly prevalent. These tools are typically deployed for between-session support to sustain clinical engagement. Existing applications span a range of unique interventions, including adaptations of cognitive behavioral therapy (CBT) and blended-modality approaches. For this review, we define chatbots as using conditional logic or AI. To maintain the scope of this manuscript, we only discuss those chatbots that indicate AI use. Hua et al. found that over five years (2020-2025), LLM-based chatbots surged to 45% of new studies in 2024, yet only 16% underwent clinical efficacy testing, with most still in early validation.22 Furthermore, much of the research uses college or non-clinical samples, and some studies specifically exclude individuals with diagnoses (e.g., Sabour et al., 202323); future work is needed to verify and strengthen the generalizability of these findings. Still, rule-based systems accounted for the majority of clinical efficacy trials.

Across the broader chatbot literature, meta-analytic evidence suggests small-to-moderate effects of use on depression and anxiety symptoms. Sohn et al. (2026) reported across 39 RCTs that chatbot interventions produced statistically significant post-intervention reductions relative to control conditions in both depressive symptoms and anxiety symptoms, albeit with substantial between-study heterogeneity.24 The patient’s baseline clinical status moderated effects on depressive symptoms. Sohn and colleagues acknowledged that 35 of the 39 trials were at risk of bias due to largely unblinded self-report outcomes, and some evidence of publication bias was detected for the depression outcomes.24 Findings also suggest no significant difference between simple retrieval-based and generative AI chatbots.

Within the CBT space, Fitzpatrick et al. conducted one of the earliest controlled evaluations of a fully automated conversational agent.25 In their 2-week unblinded RCT (N=70, ages 18–28), college students assigned to Woebot showed significant reductions in PHQ-9 depression scores relative to a NIMH ebook control, with comparable patterns on the GAD-7 and PANAS. Inkster et al. (2018) extended this work into real-world deployment with Wysa, an empathy-driven conversational agent.26 In their observational study of 129 global users self-reporting depressive symptoms, high-frequency users showed significantly greater PHQ-9 improvement than low-frequency users, and 67.7% provided positive qualitative feedback.

Sabour et al. randomized adults in China to a template-based CBT chatbot, that platform plus a generative dialogue platform for emotional venting, or a waitlist control.23 Both active conditions improved relative to control on depression and negative affect, with effects the authors characterize as small and no group difference on anxiety. The two active conditions did not differ from each other, so the effects cannot be attributed to the generative component.

Evidence for these tools is still accumulating and remains preliminary. Schillings et al., in a two-arm RCT of 118 stressed individuals, tested the chatbot ELME across a three-week program targeting stress, mindfulness, and interoception, and found that while perceived stress was unchanged, mindfulness and the reappraisal subfacet of emotion regulation increased significantly in the intervention group.27 In a broad scoping review of the effect of “Automated conversation agents (CAs)” on the emotional component of mental health among young people, Balan and colleagues (2024) found that such tools, predominantly AI-based chatbots, were acceptable, engaging, and highly usable.28 However, that evidence for clinical efficacy was far less conclusive, with nearly half of evaluation studies reporting no significant effect. Overall, the lack of conclusive findings supports using these tools alongside established, effective clinical care.

For more clinically significant symptomology, Heinz and colleagues tested their “Therabot” to treat major depressive disorder, generalized anxiety disorder, and clinically high-risk feeding and eating disorder.29 Findings indicated 95% engagement, with a mean of 260 messages and 6.18 hours of use over 4 weeks. The authors described the working alliance as comparable to outpatient psychotherapy. Importantly, there were several safety concerns, namely 15 safety-staff interventions for suicidal ideation and 13 corrections for inappropriate responses. This research, however, suggests that while AI-based chatbots can provide a resource when therapists are unavailable, they are safer and more effective when used alongside psychotherapy.

Most relevant to the emotional regulation space, specific analyses have found that AI effectively prompts cognitive reappraisal. In an analysis of a GPT-based LLM used for Cognitive Reappraisals of Stress, LLM-enhanced cognitive reappraisal as a “brief digital intervention” was associated with significant improvements in stress mindset.30 That said, participants raised qualitative concerns about “AI-driven empathy” and perceived scriptedness. Furthermore, Held and colleagues found that AI-chatbots based on a multi-agent “Socrates 2.0” had relatively small impacts on declines in depressive symptoms, social anxiety, obsessive-compulsive symptoms, and posttraumatic stress symptoms (all Cohen’s d <0.33) (2025).31

Given the modest effect sizes and the safety concerns noted above, these findings do not support using such tools as substitutes for clinician-delivered care. As Heinz and colleagues note, these tools show the most promise when therapists (as the standard of care) are unavailable (2025).29

Symptom tracking

A third clinical application is symptom monitoring. Assessment has traditionally depended on periodic self-report collected at appointments; AI applied to passively collected sensor data offers a way to extend it between them. Passive sensing using both wearables (such as smartwatches) and smartphones has been applied to detect changes in depressive and anxiety symptoms.32 Qualitative evidence indicates that healthcare professionals view such passive-sensing and ML tools as a potentially useful adjunct to clinical monitoring, while raising concerns about data interpretation and patient self-surveillance.33 As in an EMA study, these tools extend symptom assessment from a discrete appointment to continuous, ecologically valid measurement.

Further research, including a scoping review of 42 studies, found that passive sensing for individuals diagnosed with depression or anxiety was accurate up to 92.16% for anxiety detection.32 Key variables measured were step count and heart rate, along with social interaction patterns and sleep. In a single large study of over 1,000 individuals, passive smartphone and device data predicted future PHQ-8 and GAD-7 scores out one to two weeks at both within- and between-person levels.34

AI and research

Beyond its growing role as a source of emotional support, AI is reshaping how researchers study emotion regulation. ML, Natural Language Processing (NLP), and LLMs allow researchers to analyze data at a scale and temporal resolution that were previously impractical, including session transcripts, unstructured patient narratives, neuroimaging, physiological telemetry, and experience sampling data. These tools are now being applied across the research enterprise to evaluate treatment outcomes, code moment-to-moment psychotherapy processes, identify risk for psychopathology before acute presentation, support diagnostic decision-making, and develop and administer assessments. A common thread across these applications is a shift away from static, retrospective, and categorical measurement towards continuous, dynamic quantification of emotional regulation as it unfolds.35,36 The sections that follow review applications of AI in psychotherapy outcome and process research, the limitations of AI as a therapeutic agent, psychopathology research, diagnostic accuracy, and assessment, considering both the promise and limitations of each for studying emotion regulation.

Psychotherapy outcome research

AI models are improving at analyzing clinical discourse, allowing for the continuous monitoring of patient progress during psychotherapeutic interventions. For example, Orrù and Mannarini highlighted how automated text analysis can yield data-informed insights to evaluate therapeutic change over time.35 Xin et al. (2024) demonstrated that LLMs can parse unstructured interviews to identify functional shifts in peer relationships and academic performance.36 By pulling these behavioral indicators directly from natural language, researchers can quantify therapeutic trajectories over time rather than relying solely on static, categorical diagnoses.

AI models can also automate the coding of complex psychotherapeutic process variables, accelerating research into how clients develop emotion regulation skills. Analyzing session transcripts computationally allows researchers to track patient sentiment alongside coping and affective experiences without the cost and time bottlenecks of traditional manual observer ratings.37 Processing clinical data at this scale makes it possible to identify precisely when clients successfully engage emotionally with their problems, establishing a clear empirical link between in-session sentiment shifts and long-term emotion regulation outcomes.37

When evaluated against human practitioners using the Cognitive Therapy Rating Scale, ChatGPT-3.5 struggled to match human practitioners delivering text-based CBT.38 Human clinicians significantly outperformed the generative model, particularly in guided discovery and empathy. Evaluators noted that even though the AI could apply basic CBT structures, evaluators rated its responses as generic and rigid.38 Consequently, even if conversational agents can help alleviate acute distress, they lack the clinical depth required for standalone treatment and work best as assistive tools.38,39

Psychotherapy process research

Psychotherapy process research has incorporated AI in multiple ways. For example, treatment fidelity studies using linguistic coding processes are becoming relatively common.40,41 Automated approaches also identify the use of specific therapeutic techniques (e.g., Imel, Steyvers, & Atkins, 201542) and relate them to outcomes.43 Goldberg et al. (2020) also examined language patterns by both therapist and patient to identify therapeutic alliance44; changes in alliance over time may indicate emotion regulation or changes therein. Language patterns have been used to identify emotion itself. Still, research on emotion regulation remains limited.45 Additionally, methods for classification of sequences of turn-taking have been identified (e.g., Hidden Markov Models).

Several ethical issues still remain with using AI in psychotherapy process research and other forms of treatment.46,47,48 For example, confidentiality concerns appear in any system that transmits information. Older systems (e.g., Foltz’s online website that conducts linguistic analysis) and newer AI agents such as Claude and ChatGPT, rather than algorithms that run locally on a machine, can record and potentially share information transmitted about patients and clients. There is also concern that the work of psychotherapy and emotion regulation is greater than simply linguistic exchange; the attempt to allow automated systems to exchange language for this purpose (e.g., genuine concern for well-being on the part of a therapist rather than the lack of any such concern by an automated agent) may be different from what a client expects from the psychotherapy process.

These criticisms raise ethical and juridical considerations (see above). For example, there are no licensing requirements for AI as there are for humans, though there are state-level laws that restrict AI use for therapy per se. Those who create AI agents that may help individuals in their emotion regulation are often very clear that they are not providing psychotherapy (e.g., Claude) per se, though there are exceptions (e.g., Ash-AI); hence, psychotherapy process research proper is somewhat limited, but use of AI in the process of emotion regulation still exists.

So far, psychotherapy process research has focused on emotion regulation per se; however, promising work has begun to emerge. For example, Haydarov et al. (2025) attempted to code suprasegmental linguistic markers (e.g., tonality shifts, prosody of speech), which relate to emotion regulation.49 Specifically, Haydarov et al. believed that such linguistic markers indicate emotional labor (i.e., working with emotions).49 They hypothesized that this is especially true when client and therapist language use mismatch. Additionally, automated feature-extraction research on emotion using programs such as the Linguistic Inquiry Word Count (LIWC) in text exchanges exists (e.g., Kahn, Tobin, Massey, & Anderson, 200750 ).34

Furthermore, Yu et al. (2025) examined other ways LLMs may be used to detect emotion regulation; these models have been embedded in contexts that measure not only language but also physiological responses and movement.51 They concluded that such integrated research yields potential for emotion research but is still developing.

Another focus of emotion regulation work related to AI has been on using AI to identify emotions themselves. Some (e.g., Zinbarg & Mineka, 2007; Davidson, Jackson, & Kalin, 2000)52,53 have focused on looking at change in affect as an indicator of affective regulation. By identifying affective expression in psychotherapy sessions (e.g., Rasool et al., 202554; Wyman & Zhang, 2025a, 2025b, under review55,56,57), and modeling affective chronometry (e.g., Hemenover, 200358), there is additional potential for the use of AI and emotion regulation research within psychotherapy. For a thorough discussion about available packages for using AI to detect facial emotion, see Wyman and Zhang’s (2025)55,56 Multivariate Behavioral Research Article. These automated modules could enable research examining paralleled within-session and extra-session experiences of emotion and emotion regulation.

Additionally, recent literature has examined the use of AI to assist the psychotherapy process. As stated, AI has identified elements of therapy. However, research using AI to suggest therapeutic interventions has shown some promise.35 Imel et al. (2019) developed a system for providing therapists with automated feedback for psychotherapy.41 Fitzpatrick et al. also developed a framework for automated psychotherapy feedback more generally speaking.25 Mendlovic (2026) has provided a framework for AI to assist with the therapeutic process based on changes in client-therapist interaction within and across therapy sessions.60 Indeed, others (e.g., Beg et al., 202661) have provided similar frameworks for augmenting individual therapy with AI. Changes within therapy sessions, as identified by AI, may relate to emotion, and this research may prove promising for psychotherapy process in general.

One important consideration in AI-human interaction that has recently emerged, potentially related to emotion regulation, is sycophancy (i.e., flattering, agreeing with, or endorsing the user, perhaps excessively). Sycophancy, particularly when it avoids confronting unrealistic or inappropriate content, may relate to diminished distress tolerance.3 AI interactions, particularly for emotional support, have been shown to engage in sycophantic responding.62,63 However, the consequences of sycophancy have been mixed. Lee et al. found that sycophancy increased intention to rely on AI and also increased social well-being.63 Cheng et al. (2026) also found that sycophantic AI increased AI use and potential dependence, while also decreasing users’ prosocial intentions.62 This may be potentially troublesome given that social support seeking, sometimes related to prosocial behavior, has been linked to emotion regulation and well-being.2 Research on sycophancy and AI is in its fledgling stages but is important to consider for emotion regulation research.

Therapist and efficacy limitations

AI is increasingly positioned to expand therapeutic reach without replacing human clinical judgment, offering students a scalable and stigma-free avenue for support.64 Acting as accessible assistants, conversational agents can bridge treatment gaps during high-stress academic periods by delivering evidence-based tools like mood tracking and cognitive behavioral interventions.64

Recent feasibility trials demonstrate that brief, LLM-based interventions can successfully help users reappraise stress and shift their mindsets.30 Yet, qualitative feedback highlights distinct limitations to this automated approach. Users frequently note that the chatbot's dialogue feels overly scripted, describing its attempts at empathy as noticeably artificial.30

As generative AI enters clinical settings, it brings complex methodological and ethical challenges.65,66 Much of the criticism centers on early RCTs; researchers argue that wait-list controls artificially boost efficacy scores and that developer allegiance compromises independent testing.65,66 Beyond trial design, there is clinical tension over the therapeutic alliance itself. Critics emphasize the danger of the 'psychologist’s fallacy' (assuming conversational agents possess a human-like subjectivity they inherently lack) and argue that the rush to commercialize these tools is bypassing critical clinical safeguards.65

Psychopathology research

ML now enables clinicians to track longitudinal illness trajectories, helping to identify emotion dysregulation well before a crisis occurs.67 By analyzing large-scale clinical and behavioral datasets (from electronic health records to wearable sensor telemetry) predictive models can successfully flag latent risk factors for affective instability.68 These predictive tools are also advancing affective neuroscience by linking specific mental conditions with distinct brain activity.68 Rather than using aggregated fMRI data, modern algorithms train on highly individualized functional connectivity maps to forecast behavioral shifts with greater accuracy.69 This precision allows researchers to map the physical brain architecture driving emotional volatility long before acute symptoms manifest.

LLMs advance this approach by identifying latent linguistic markers of conditions like depression and suicide risk.70,71,72This allows clinicians to extract objective, real-time diagnostic data from continuous speech rather than depending exclusively on formal clinical interviews.71 In the context of emotion dysregulation, automated systems can actively walk patients through distinct stages of regulation.73 For example, these tools can prompt users to reframe stressful events to facilitate cognitive reappraisal.73 As the patient responds, AI algorithms monitor their linguistic shifts, evaluating changes in sentiment and adjective selection.73 Tracking this trajectory (from initial emotional venting to active cognitive restructuring) provides researchers with a real-time, objective measure of regulatory mechanics.73

Diagnostic accuracy and methodological considerations

Despite ongoing methodological challenges, evolving LLMs are increasingly supporting psychiatric reasoning and diagnostic reliability.21 Systematic reviews indicate that these systems can process unstructured data (such as clinical notes and electronic health records) to assist with triage and risk assessment.21 Capturing this information computationally offers a practical way to screen for mental health conditions long before they escalate.21

Researchers are now testing hybrid systems that merge LLMs with constraint logic programming to reduce the risk of AI hallucinations and erratic outputs.20 Instead of allowing the LLM to generate clinical judgments directly, these architectures use the model strictly to translate diagnostic manuals into verifiable logic rules.20 Grounding the diagnostic process in established guidelines (and keeping an expert in the loop) keeps the AI from generating unfounded conclusions ensuring the final output remains accurate, safe, and easily interpretable.20

A primary methodological challenge in AI diagnostic research remains the Jingle Jangle fallacy (i.e., calling two different constructs by the same name or calling the same thing by two different names; Kelley, 192774). Models must accurately isolate emotion regulation rather than accidentally measuring generalized negative affect or other overlapping proxies.75 This challenge echoes debates in affective neuroscience about operationalizing regulation.76 Historically, some researchers took a functional approach, arguing that regulation could be measured simply by observing the difference in an emotional state across two time points.77 Others contend that just tracking a change in affect is not enough; a valid measurement has to identify the specific cognitive or behavioral strategies used to drive that change.78,79 When training AI on clinical datasets, defining this distinction is critical.

There is broad consensus that LLMs are best utilized as supportive, clinician-facing analytical tools. Human practitioners remain essential for interpreting context, navigating clinical uncertainty, and making final diagnostic decisions.80,81

Assessment

Modern psychological assessment relies more and more on continuous multimodal data; this potentially allows for the capture of temporal changes in emotional functioning.82,83 Kargarandehkordi et al. demonstrate this by using AI to analyze physiological telemetry, such as continuous actigraphy, sleep architecture, and wearable biosensors.83 Synthesizing these diverse data streams allows clinicians to move beyond static behavioral snapshots to generate actionable inferences.82

To achieve this synthesis, AI models process complex time-series data. Moving away from retrospective self-reports, these models analyze continuous data streams to detect early or mild instances of emotional distress.82 Although the predictive capabilities of these approaches show promise, their full clinical utility remains under investigation. Specifically, while algorithms applied to digital phenotyping perform well in identifying atypical mobility, it is unclear whether they fully capture underlying regulatory capacities. Ultimately, the co-use of computational synthesis and clinical oversight may prove most fruitful in distinguishing between normative stress reactivity and early-stage mental health symptoms.82

Additionally, numerous attempts exist to automate item generation; this would theoretically allow for automated assessment of emotion and emotion regulation. For example, Russell-Lasalandra et al. used a network psychometric approach to generate items that they claim reflect the state of the art in using AI to generate measures.84 The psychometric evidence of the approaches appears convincing, though more work is needed. Specifically, while systems such as AI-GENIE seem to generate items that perform well in assessment of emotion and related constructs, it is unclear whether they fully sample the construct domain and whether incorporating human input improves the scale. The co-use of human and AI (i.e., Mollick, 202485) may prove particularly fruitful.

Conclusion

This paper highlights important aspects of emotion regulation and AI. It cannot cover all important aspects, but it highlights salient trends in the most essential areas of emotion regulation. As stated, we anticipate the need for updated reviews in the future – perhaps the near future because of the rapid developments in this field. Future research might also approach the research process using a number of alternative methods; this might be particularly useful for researchers wishing to focus explicitly on one area rather than attempting to highlight the breadth of research in the area. For example, Tricco et al.’s PRISMA scoping review guidelines, PRISMA’s extended (2020) checklist may be useful for researchers addressing quantitative areas of research.86 The SANRA (2019) narrative review method, the ENTREQ qualitative research synthesis methodology, and the eMERGe meta-ethnography approach may also be relevant in reviewing qualitative approaches to emotion-regulation.87,88 We welcome input from readers engaged in this area and hope that we provide a useful overview of important aspects for clinicians, researchers, and consumers of emotion-regulation-related knowledge.

References

  1. Zao-Sanders M. How people are really using AI in 2026. Harvard Business Review. Published June 1, 2026. Accessed August 28, 2026. https://hbr.org/2026/06/how-people-are-really-using-ai-in-2026
  2. Lopez RB, Courtney AL, Liang D, Swinchoski A, Goodson P, Denny BT. Social support and adaptive emotion regulation: links between social network measures, emotion regulation strategy use, and health. Emotion. 2024;24(1):130-138. doi:10.1037/emo0001242
  3. Linehan MM. Cognitive-Behavioral Treatment of Borderline Personality Disorder. Guilford Press; 1993.
  4. Anthropic. How people use Claude for support, advice, and companionship. Anthropic. Accessed May 8, 2026. https://www.anthropic.com/news/how-people-use-claude-for-support-advice-and-companionship
  5. Phang, J, Lampe, M, Ahmad, L, Agarwal, S, Fang, C M, Liu, A R, Danry, V, Lee, E, Chan, S W T, Pataranutaporn, P, & Maes, P . Investigating Affective Use and Emotional Well-being on ChatGPT. (n.d.)
  6. Gable SL, Reis HT, Impett EA, Asher ER. What do you do when things go right? The intrapersonal and interpersonal benefits of sharing positive events. J Pers Soc Psychol. 2004;87(2):228-245. doi:10.1037/0022-3514.87.2.228
  7. McBain RK, Bozick R, Diliberti M, et al. Use of generative AI for mental health advice among US adolescents and young adults. JAMA Netw Open. 2025;8(11):e2542281. doi:10.1001/jamanetworkopen.2025.42281
  8. McBain RK, Cantor JH, Breslau J, et al. AI chatbot use and disclosure for mental health among US adolescents and young adults. JAMA Pediatr. Published online 2026. doi:10.1001/jamapediatrics.2026.2015
  9. Liu CH, Zhang W, Lou F, Zhao C, Chow A, Yip T. Clinical and sociodemographic predictors of AI use for mental health among college students. J Affect Disord. 2026;412:122058. doi:10.1016/j.jad.2026.122058
  10. Rousmaniere T, Zhang Y, Li X, Shah S. Large language models as mental health resources: patterns of use in the United States. Pract Innov (Wash D C). 2026;11(2):139-155. doi:10.1037/pri0000292
  11. Alanezi F. Assessing the effectiveness of ChatGPT in delivering mental health support: a qualitative study. J Multidiscip Healthc. 2024;17:461-471. doi:10.2147/JMDH.S447368
  12. Luo X, Ghosh S, Tilley JL, Besada P, Wang J, Xiang Y. "Shaping ChatGPT into my digital therapist": a thematic analysis of social media discourse on using generative artificial intelligence for mental health. Digit Health. 2025;11:20552076251351088. doi:10.1177/20552076251351088
  13. Luo X, Wang Z, Tilley JL, Balarajan S, Bassey UA, Cheang CI. Seeking emotional and mental health support from generative AI: mixed-methods study of ChatGPT user experiences. JMIR Ment Health. 2025;12:e77951. doi:10.2196/77951
  14. Harkin B, Webb TL, Chang BPI, et al. Does monitoring goal progress promote goal attainment? A meta-analysis of the experimental evidence. Psychol Bull. 2016;142(2):198-229. doi:10.1037/bul0000025
  15. Siddals S, Torous J, Coxon A. "It happened to be the perfect thing": experiences of generative AI chatbots for mental health. NPJ Ment Health Res. 2024;3(1):48. doi:10.1038/s44184-024-00097-4
  16. Lee S, Rheu M, Zhuang J. The ChatGPT effect: investigating shifting discourse patterns, sentiment, and benefit–challenge framing in AI mental health support. Behav Sci (Basel). 2025;15(9):1172. doi:10.3390/bs15091172
  17. Rogers CR. The necessary and sufficient conditions of therapeutic personality change. Psychotherapy (Chic). 2007;44(3):240-248. doi:10.1037/0033-3204.44.3.240
  18. American Psychological Association. Health advisory: use of generative AI chatbots and wellness applications for mental health. American Psychological Association. Published November 13, 2025. Accessed August 31, 2026. https://www.apa.org/topics/artificial-intelligence-machine-learning/health-advisory-chatbots-wellness-apps
  19. Kazantzis N, Whittington C, Dattilio F. Meta-analysis of homework effects in cognitive and behavioral therapy: a replication and extension. Clin Psychol Sci Pract. 2010;17(2):144-156. doi:10.1111/j.1468-2850.2010.01204.x
  20. Kim BH, Wang C. Large language models for interpretable mental health diagnosis. arXiv. Preprint posted online 2025. doi:10.48550/arXiv.2501.07653
  21. Cruz-Gonzalez P, He AWJ, Lam EP, et al. Artificial intelligence in mental health care: a systematic review of diagnosis, monitoring, and intervention applications. Psychol Med. 2025;55:e18. doi:10.1017/S0033291724003295
  22. Hua Y, Siddals S, Ma Z, et al. Charting the evolution of artificial intelligence mental health chatbots from rule-based systems to large language models: a systematic review. World Psychiatry. 2025;24(3):383-394. doi:10.1002/wps.21352
  23. Sabour S, Zhang W, Xiao X, et al. A chatbot for mental health support: exploring the impact of Emohaa on reducing mental distress in China. Front Digit Health. 2023;5:1133987. doi:10.3389/fdgth.2023.1133987
  24. Sohn JS, Ha BG, Park S, et al. Systematic review and meta-analysis of chatbots in the management of depressive and anxiety symptoms. NPJ Digit Med. 2026. doi:10.1038/s41746-026-02566-w
  25. Fitzpatrick KK, Darcy A, Vierhile M. Delivering cognitive behavior therapy to young adults with symptoms of depression and anxiety using a fully automated conversational agent (Woebot): a randomized controlled trial. JMIR Ment Health. 2017;4(2):e19. doi:10.2196/mental.7785
  26. Inkster B, Sarda S, Subramanian V. An empathy-driven, conversational artificial intelligence agent (Wysa) for digital mental well-being: real-world data evaluation mixed-methods study. JMIR Mhealth Uhealth. 2018;6(11):e12106. doi:10.2196/12106
  27. Schillings C, Meißner E, Erb B, Bendig E, Schultchen D, Pollatos O. Effects of a chatbot-based intervention on stress and health-related parameters in a stressed sample: randomized controlled trial. JMIR Ment Health. 2024;11:e50454. doi:10.2196/50454
  28. Balan R, Dobrean A, Poetar CR. Use of automated conversational agents in improving young population mental health: a scoping review. NPJ Digit Med. 2024;7:75. doi:10.1038/s41746-024-01072-1
  29. Heinz MV, Mackin DM, Trudeau BM, et al. Randomized trial of a generative AI chatbot for mental health treatment. NEJM AI. 2025a;2(4). doi:10.1056/AIoa2400802
  30. Bhattacharjee A, Suh J, Chandra M, Hernandez J. User perceptions of an LLM-based chatbot for cognitive reappraisal of stress: feasibility study. JMIR Hum Factors. Published online 2026. doi:10.2196/89736
  31. Held P, Pridgen SA, Szoke DR, Chen Y, Akhtar Z, Amin D. AI-facilitated cognitive reappraisal via Socrates 2.0: mixed methods feasibility study. JMIR Ment Health. 2025;12:e80461. doi:10.2196/80461
  32. Shen S, Qi W, Zeng J, et al. Passive sensing for mental health monitoring using machine learning with wearables and smartphones: scoping review. J Med Internet Res. 2025;27(1):e77066. doi:10.2196/77066
  33. Rogan J, Bucci S, Firth J. Health care professionals' views on the use of passive sensing, AI, and machine learning in mental health care: systematic review with meta-synthesis. JMIR Ment Health. 2024;11:e49577. doi:10.2196/49577
  34. Stamatis CA, Meyerhoff J, Meng Y, et al. Differential temporal utility of passively sensed smartphone features for depression and anxiety symptom prediction: a longitudinal cohort study. NPJ Ment Health Res. 2024;3(1):1. doi:10.1038/s44184-023-00041-y
  35. Orrù L, Mannarini S. The role of artificial intelligence in clinical psychology: how AI and NLP systems are reshaping psychological interventions. A systematic review. Clin Psychol Psychother. 2026;33(2):e70242. doi:10.1002/cpp.70242
  36. Xin AW, Nielson DM, Krause KR, et al. Using large language models to detect outcomes in qualitative studies of adolescent depression. J Am Med Inform Assoc. 2026;33(1):79-89. doi:10.1093/jamia/ocae298
  37. Eberhardt ST, Schaffrath J, Moggia D, et al. Decoding emotions: exploring the validity of sentiment analysis in psychotherapy. Psychother Res. 2025;35(2):174-189. doi:10.1080/10503307.2024.2322522
  38. Acevedo S, Aneja E, Opler DJ, Valera P, Jarmon E. Evaluating the efficacy of ChatGPT-3.5 versus human-delivered text-based cognitive-behavioral therapy: a comparative pilot study. Am J Psychother. 2026;79(1):4-11. doi:10.1176/appi.psychotherapy.20240070
  39. Im CH, Woo M. Clinical efficacy, therapeutic mechanisms, and implementation features of cognitive behavioral therapy–based chatbots for depression and anxiety: narrative review. JMIR Ment Health. 2025;12:e78340. doi:10.2196/78340
  40. Atkins DC, Steyvers M, Imel ZE, Smyth P. Scaling up the evaluation of psychotherapy: evaluating motivational interviewing fidelity via statistical text classification. Implement Sci. 2014;9(1):49. doi:10.1186/1748-5908-9-49
  41. Imel ZE, Pace BT, Soma CS, et al. Design feasibility of an automated, machine-learning based feedback system for motivational interviewing. Psychotherapy (Chic). 2019;56(2):318-328. doi:10.1037/pst0000221
  42. Imel ZE, Steyvers M, Atkins DC. Computational psychotherapy research: scaling up the evaluation of patient–provider interactions. Psychotherapy (Chic). 2015;52(1):19-30. doi:10.1037/a0036841
  43. Althoff T, Clark K, Leskovec J. Large-scale analysis of counseling conversations: an application of natural language processing to mental health. Trans Assoc Comput Linguist. 2016;4:463-476. doi:10.1162/tacl_a_00111
  44. Goldberg SB, Flemotomos N, Martinez VR, et al. Machine learning and natural language processing in psychotherapy research: alliance as example use case. J Couns Psychol. 2020;67(4):438-448. doi:10.1037/cou0000382
  45. Tanana MJ, Soma CS, Kuo PB, et al. How do you feel? Using natural language processing to automatically rate emotion in psychotherapy. Behav Res Methods. 2021;53(5):2069-2082. doi:10.3758/s13428-020-01531-z
  46. Imel ZE, Creed T, Kious B, Althoff T, Atzil-Slonim D, Srikumar V. A framework for automation in psychotherapy. Curr Dir Psychol Sci. 2026;35(2):66-76. doi:10.1177/09637214251386047
  47. Oh O, Demiris G, Ulrich CM. The ethical dimensions of utilizing artificial intelligence in palliative care. Nurs Ethics. 2025;32(4):1285-1296. doi:10.1177/09697330241296874
  48. Coghlan S, Leins K, Sheldrick S, Cheong M, Gooding P, D'Alfonso S. To chat or bot to chat: ethical issues with using chatbots in mental health. Digit Health. 2023;9:20552076231183542. doi:10.1177/20552076231183542
  49. Haydarov K, Mohamed Y, Goldenhersch E, OCallaghan P, Li L, Elhoseiny M. Towards AI-assisted psychotherapy: emotion-guided generative interventions. In: Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing. 2025:32724-32743. doi:10.18653/v1/2025.emnlp-main.1664
  50. Kahn JH, Tobin RM, Massey AE, Anderson JA. Measuring emotional expression with the Linguistic Inquiry and Word Count. Am J Psychol. 2007;120(2):263-286.
  51. Yu L, Ge Y, Ansari S, Imran M, Ahmad W. Multimodal sensing-enabled large language models for automated emotional regulation: a review of current technologies, opportunities, and challenges. Sensors (Basel). 2025;25(15):4763. doi:10.3390/s25154763
  52. Zinbarg RE, Mineka S. Is emotion regulation a useful construct that adds to the explanatory power of learning models of anxiety disorders or a new label for old constructs? Am Psychol. 2007;62(3):259-261. doi:10.1037/0003-066X.62.3.259
  53. Davidson RJ, Jackson DC, Kalin NH. Emotion, plasticity, context, and regulation: perspectives from affective neuroscience. Psychol Bull. 2000;126(6):890-909. doi:10.1037/0033-2909.126.6.
  54. Rasool A, Aslam S, Hussain N, Imtiaz S, Riaz W. nBERT: harnessing NLP for emotion recognition in psychotherapy to transform mental health care. Information. 2025;16(4):301. doi:10.3390/info16040301
  55. Wyman A, Zhang Z. Naturalistic emotion regulation detection: a novel framework for extracting features from emotion detection AI data. Manuscript submitted for publication.
  56. Wyman A, Zhang Z. A tutorial on the use of artificial intelligence tools for facial emotion recognition in R. Multivariate Behav Res. 2025a;60(3):641-655. doi:10.1080/00273171.2025.2455497
  57. Wyman A, Zhang Z. Evaluating the threat of phantom faces in emotion detection AI through simulation. J Behav Data Sci. 2025b;5(2):28-42. doi:10.35566/jbds/wyman51
  58. Hemenover SH. Individual differences in rate of affect change: studies in affective chronometry. J Pers Soc Psychol. 2003;85(1):121-131. doi:10.1037/0022-3514.85.1.121
  59. Imel ZE, Caperton DD, Tanana M, Atkins DC. Technology-enhanced human interaction in psychotherapy. J Couns Psychol. 2017;64(4):385-393. doi:10.1037/cou0000213
  60. Mendlovic S. AI-assisted precision psychotherapy: toward a process-based framework of therapeutic interaction. Psychiatry Int. 2026;7(3):124. doi:10.3390/psychiatryint7030124
  61. Beg MJ, Verma M, Chauhan MVK, Verma MK. Artificial intelligence for psychotherapy: a review of the current state and future directions. Indian J Psychol Med. 2025;47(4):314-325. doi:10.1177/02537176241260819
  62. Cheng M, Lee C, Khadpe P, Yu S, Han D, Jurafsky D. Sycophantic AI decreases prosocial intentions and promotes dependence. Science. 2026;391(6792):eaec8352. doi:10.1126/science.aec8352
  63. Lee D, Wan C, Ng PML, Fung YN, Wu N. Effects of AI companions' sycophancy and emotional mimicry on consumers' continuance intention and social wellbeing. Int J Hum Comput Interact. Published online 2026:1-23. doi:10.1080/10447318.2026.2626809
  64. Nyakhar S, Wang H. Effectiveness of artificial intelligence chatbots on mental health & well-being in college students: a rapid systematic review. Front Psychiatry. 2025;16:1621768. doi:10.3389/fpsyt.2025.1621768
  65. Gratch I, Essig T. A letter about "Randomized Trial of a Generative AI Chatbot for Mental Health Treatment." NEJM AI. 2025;2(9). doi:10.1056/AIp2500390
  66. Heinz MV, Mackin DM, Trudeau BM, et al. Response to letters about "Randomized Trial of a Generative AI Chatbot for Mental Health Treatment." NEJM AI. 2025b;2(9). doi:10.1056/AIp2500680
  67. Lewin G, Abakasanga E, Titcombe I, Cosma G, Gangadharan S. Artificial intelligence-enabled predictive modelling in psychiatry: overview of machine learning applications in mental health research. BJPsych Adv. 2026;32(3):169-175. doi:10.1192/bja.2025.10133
  68. Islam MM, Hassan S, Akter S, Jibon FA, Sahidullah M. A comprehensive review of predictive analytics models for mental illness using machine learning algorithms. Healthc Anal. 2024;6:100350. doi:10.1016/j.health.2024.100350
  69. Lee HJ, Dworetsky A, Labora N, Gratton C. Using precision approaches to improve brain-behavior prediction. Trends Cogn Sci. 2025;29(2):170-183. doi:10.1016/j.tics.2024.09.007
  70. Kim S, Imieye O, Yin Y. Interpretable depression detection from social media text using LLM-derived embeddings. arXiv. Preprint posted online 2025. doi:10.48550/arXiv.2506.06616
  71. Omar M, Soffer S, Charney AW, Landi I, Nadkarni GN, Klang E. Applications of large language models in psychiatry: a systematic review. Front Psychiatry. 2024;15:1422807. doi:10.3389/fpsyt.2024.1422807
  72. Imel ZE, Pace B, Pendergraft B, et al. Machine learning–based evaluation of suicide risk assessment in crisis counseling calls. Psychiatr Serv. 2024;75(11):1068-1074. doi:10.1176/appi.ps.20230648
  73. Laban G, Wang J, Gunes H. A robot-led intervention for emotion regulation: from expression to reappraisal. arXiv. Preprint posted online 2025. doi:10.48550/arXiv.2503.18243
  74. Kelley TL. Interpretation of Educational Measurements. World Book Company; 1927.
  75. Flake JK, Fried EI. Measurement schmeasurement: questionable measurement practices and how to avoid them. Adv Methods Pract Psychol Sci. 2020;3(4):456-465. doi:10.1177/2515245920952393
  76. Davidson RJ. Affective style and affective disorders: perspectives from affective neuroscience. Cogn Emot. 1998;12(3):307-330. doi:10.1080/026999398379628
  77. Cole PM, Martin SE, Dennis TA. Emotion regulation as a scientific construct: methodological challenges and directions for child development research. Child Dev. 2004;75(2):317-333. doi:10.1111/j.1467-8624.2004.00673.x
  78. Gross JJ. The emerging field of emotion regulation: an integrative review. Rev Gen Psychol. 1998;2(3):271-299. doi:10.1037/1089-2680.2.3.271
  79. Gross JJ. Emotion regulation: current status and future prospects. Psychol Inq. 2015;26(1):1-26. doi:10.1080/1047840X.2014.940781
  80. Porsdam Mann S, Earp BD, Liu P, Savulescu J. Reasons in the loop: the role of large language models in medical co-reasoning. Am J Bioeth. 2024;24(9):105-107. doi:10.1080/15265161.2024.2383121
  81. Pressman SM, Borna S, Gomez-Cabello CA, Haider SA, Haider CR, Forte AJ. Clinical and surgical applications of large language models: a systematic review. J Clin Med. 2024;13(11):3041. doi:10.3390/jcm13113041
  82. Choi A, Ooi A, Lottridge D. Digital phenotyping for stress, anxiety, and mild depression: systematic literature review. JMIR Mhealth Uhealth. 2024;12:e40689. doi:10.2196/40689
  83. Kargarandehkordi A, Li S, Lin K, Phillips KT, Benzo RM, Washington P. Fusing wearable biosensors with artificial intelligence for mental health monitoring: a systematic review. Biosensors (Basel). 2025;15(4):202. doi:10.3390/bios15040202
  84. Russell-Lasalandra LL, Christensen AP, Golino H. Generative psychometrics via AI-GENIE: automatic item generation and validation with network-integrated evaluation. Behav Res Methods. 2026;58(8):217. doi:10.3758/s13428-026-03082-1
  85. Mollick E. Co-Intelligence: Living and Working with AI. Portfolio/Penguin; 2024.
  86. Tricco AC, Lillie E, Zarin W, et al. PRISMA extension for scoping reviews (PRISMA-ScR): checklist and explanation. Ann Intern Med. 2018;169(7):467-473. doi:10.7326/M18-0850
  87. Baethge C, Goldbeck-Wood S, Mertens S. SANRA—a scale for the quality assessment of narrative review articles. Res Integr Peer Rev. 2019;4:5. doi:10.1186/s41073-019-0064-8
  88. France EF, Cunningham M, Ring N, et al. Improving reporting of meta-ethnography: the eMERGe reporting guidance. BMC Med Res Methodol. 2019;19(1):25. doi:10.1186/s12874-018-0600-0

Article information

Authors

How to cite

Anderson, K. M., Keefe, D., et al., 2026. Artificial Intelligence and Emotion Regulation: Information Relating to Consumers, Psychotherapy, and Research. Medical Research Archives, [online] 14(10). https://doi.org/10.18103/mra.2026.0610

Record

Journal
Medical Research Archives
Section
Research Articles
Published
31 October 2026
ISSN
2375-1924
Downloads
136

Copyright and licence

Published by the European Society of Medicine under a Creative Commons Attribution 4.0 International licence (CC BY 4.0), which permits unrestricted use, distribution and reproduction in any medium, provided the original authors and source are credited.