Mental Health Chatbots for Patient Care Systems

Tripti Pal1, Adrija Saha1, Shounak Dutta1, Soumidatta Paul1, Rupsha Basu1,
Ranit De1, Triparna Mukherjee1, Devjyoti Ghosh1, Arunima Das1, Abhishek
Mukhopadhyay2*
1
Amity Institute of Biotechnology,
Amity University
Kolkata: 700135
2
Computer Science and Engineering
School of Engineering and Technology
Amity University
Kolkata: 700135
Email ID of the Corresponding author: amukhopadhyay@kol.amity.edu

ABSTRACT
Mental health disorders affect nearly a billion people worldwide, with anxiety and
depression as the leading contributors to this global crisis. Despite advancements in treatments
like Cognitive Behavioral Therapy (CBT) and psychotherapy, access to mental health care
remains a significant challenge due to financial constraints, geographical barriers, and a shortage
of trained professionals. Recent technological advancements, particularly in artificial intelligence
(AI), have introduced innovative solutions such as AI-powered chatbots, which offer scalable,
cost-effective, and accessible mental health support. This chapter explores the transformative
potential of AI chatbots in mental health care, focusing on their ability to simulate human
conversation, deliver therapeutic interventions, and provide 24/7 personalized support in a
stigma-free environment. By examining popular applications like Woebot, Wysa, MoodKit, and
Happify, this chapter highlights how AI-driven tools have effectively addressed symptoms of
anxiety, depression, and stress, particularly during the post-COVID-19 era. The chapter also
delves into the integration of cutting-edge technologies like Large Language Models (LLMs) to
enhance chatbot performance and user engagement. This chapter also outlines challenges such
as ethical concerns, long-term efficacy, and system integration, providing future research
directions for developing more holistic, inclusive, and impactful AI-driven mental health
interventions.

Keywords: Mental Health Chatbots, Cognitive Behavioral Therapy, Depression, Psychotherapy,

Machine Learning, Natural Language Processing, Patient Care System.
1. INTRODUCTION
Close to a billion people worldwide suffer from some form of mental disorder, ranging
from the most common conditions of anxiety and depression to psychotic and personality
disorders such as Autism Spectrum Disorder (ASD), Attention-Deficit/Hyperactivity Disorder
(ADHD), and so on [1]. The World Health Organization (WHO) defines mental health as a state of
well-being in which individuals can realize their abilities, cope with normal life stresses, work
productively, and contribute to their community [2]. Depression and anxiety are the two leading
contributors to global mental health challenges, with nearly 5% of the global population,
equivalent to 301 million people having anxiety-related disorders [3]. The outbreak of the
COVID-19 pandemic has further exacerbated these conditions, triggering a 25% spike in the
prevalence of anxiety and depression worldwide, alongside a notable rise in suicidal tendencies
and comorbid conditions [4].

Countless extraordinary, advanced therapies like Cognitive Behavioral Therapy (CBT),

medications, electroconvulsive therapy, psychotherapy, hypnotherapy, etc. have been found
extremely effective in the treatment of mental illness [5]. However, a significant amount of
people with these conditions have not received timely treatment due to unavailability in both
developing and developed countries [6]. According to recent data from the World Health
Organization (WHO), there is a significant global gap in the availability of essential mental health
services [7]. Despite increasing awareness and efforts towards the lessening of stigma, access to
mental health services remains a serious concern for many individuals due to several factors like
geographical barriers, financial constraints, and a shortage of trained mental health
professionals. There also exists an unacceptable gap between the rich and the poor, the
connected and the excluded, as reflected by the Mental Health Atlas of the WHO, which draws
on data from 171 countries. For instance, while only one-third of people suffering from
depression in developed countries see a professional for some form of treatment, the number of
sufferers under proper care is desperately low— from 23% in high-income regions to as low as
3% in low- and lower-middle-income countries. In a nutshell, such exclusion leads to massive
rates of suicidal behaviors and mortality, besides requiring urgent systemic care reform [8].

Artificial intelligence (AI)-based interventions such as chatbots and therapy bots and
specialized chat applications are able to overcome the above-mentioned shortcomings of
traditional psychotherapy. Technological influence on mental health has dramatically
accelerated social interaction and ways of communicating with others, as well as influencing
behavior and personality. Furthermore, Cyber Health Psychology has brought about a significant
transformation in the landscape of mental health support through the convergence of
psychology, health, and digital technology for better service delivery and more patient
participation. The emphasis here is on the psychological effects of technologies in healthcare
services that include mobile health applications, telemedicine, and online forums towards the
designing of digital interventions for healthy behavior and improvement of better mental health
outcomes [9]. Research findings indicate that online interventions have proved effective in
treating mental diseases like depression, anxiety, drug and substance use, and eating disorders
[10]. Additionally, the collaboration of technology with psychology, especially spatial computing
and AI, is promising to aid mental health [11]. AI psychotherapy can be delivered online through
smartphone applications and digital games.

Over the past decade, improvements in computational powers and the availability of
open-source technologies have provided AI-based chatbot programs with extensive popularity.
Improvements in AI and Natural Language Processing (NLP) ensure that it is easier to implement
chatbots to accommodate a variety of applications and to simulate human conversation with
more effectiveness. One of the most important technological innovations within mental health
care is the introduction of chatbots, or conversational agents. AI chatbots are advanced
conversational systems designed to simulate human interactions in a natural and intuitive
manner. These computer programs can think, learn, and complete tasks in combination with
humans or with independence, using big data, NLP and Machine Learning (ML) algorithms. These
chatbots utilize natural language processing to interpret user input and generate contextually
appropriate responses, enhancing their functionality to boost productivity while offering
conversation, guidance, and support [12] [13].

Given the increasing demand for mental health care on one hand and not enough mental
health care professionals available on the other hand, many AI-driven mental health apps and
chatbots have been developed to make the support more accessible, potential from therapy to
assistance. According to Moore & Caudill [14], a chatbot refers to a computer program that takes
the form of a human conversation, enabling it to respond to any question or message in text or
voice. It offers 24/7 confidential support aside from conventional, face-to-face therapy. The apps
provided examples like Happify, Moodkit, Woebot, and Wysa, wherein techniques of CBT more
than anything deliver it as mood tracking and daily mental health checks via conversational
interfaces. Such digital tools were helpful in managing both newly emerging mental health issues
and the aggravation of existing conditions.

These environments, built to resemble human communication in terms of spoken,

written, and visual language use, can therefore be a means of making mental health resources
more accessible—in particularity, for those who might otherwise be adverse to these due to
stigma [15]. AI chatbots can engage users through personalized conversations, deliver
psychoeducation, suggest coping strategies, and even provide therapeutic interventions. Their
potential in mental health support is particularly promising as they address traditional barriers to
care. Unlike in-person therapy, AI chatbots offer 24/7 availability, can reach individuals in remote
or underserved regions, and provide anonymity, which may help reduce stigma and encourage
individuals to seek help. Additionally, their scalability allows for the delivery of interventions to
large populations, helping alleviate the pressure on overburdened mental health systems facing
growing demand.

AI chatbots have shown promising advancements in supporting mental health, leading

to rich discussions around their ethics, practical applications, and technical capabilities.
Positioned at the forefront of mental health technology, these chatbots now offer far more
than simple conversation- based support [15]. The advancements in cognitive technology are
leading to a Renaissance in mental health care. Chatbots, or chatbots, are artificial intelligence-
based software that can communicate with humans using language via text or voice chat. The
technology continues to evolve and is now used in digital assistants such as Siri by Apple, Alice
by Yandex, Alexa by Amazon, and other virtual assistants for the development of a new digital
health service called mental health chatbots, which have the potential to have long-term effects
on pulmonary depression and other mental health issues [16]. The implementation of AI
chatbots in mental health driven by technological advancements is a dire need for accessible
mental health support and the potential for improved patient outcomes. With escalating
numbers of individuals all across the world struggling with mental health challenges such as
anxiety, depression, and stress, these chatbots AIs are being viewed as very promising targets
for scalable and low-cost solutions. These computer programs provide support to those who
otherwise might face barriers to mental health services, such as those in underserved, rural, or
low-income communities [10].
 AI chatbots are designed to deliver structured, evidence-based interventions that help
assuage symptoms of common mental health issues. These chatbot interventions provide
therapeutic effects of various strengths that yield positive mental health results. Case in point:
available qualitative and quantitative studies have shown that AI bots are able to reduce anxiety,
depression, and stress. This has led to significant implications for the future of mental health
care, as chatbots could alone serve as a standard support system, helping to bridge the gap
between traditional mental health care and digital support [15] [17].

One of the biggest impetuses for AI chatbot adoption is accessibility. Financial or

logistical barriers often limit access to mental health care and prevent people from receiving the
much-needed help. AI chatbots address these problems by providing users immediate,
personalized support regardless of location, thereby reducing mental health disparities and
offering a first line of care to those who might otherwise remain untreated. This means that
people can get 24/7 access to mental health support in a private, stigma-free environment
without the need for any human resources.

The successful integration of AI chatbots in mental health depends on user engagement

and the quality of the actual interaction of these chatbots. This is because natural language
processing or NLP-based chatbots, to great extents, can be supportive and empathetic towards
the users, making them feel accepted in a safer, judgment-free manner. Users tend to engage
more with the AI chatbots, as and when they have the feeling that such AI bots understand and
care for them. This is important for the therapeutic benefits, as greater user involvement means
that individuals will come back to the chatbot for continuing support [18]. Similar positive
experiences of users are very helpful towards the increasing use of AI chatbots in mental health
care.

Another advantage of the implemented AI chatbots is that they offer a cost-effective

and affordable solution option. The cost of hiring psychiatric services or the funding required to
deliver mental health services is high, in areas with high demand, creating an economic burden
for not only the users but also healthcare providers too. Since AI chatbots do not require human
intervention, a major part of expenses in mental health care is lowered, as well as sustainability
is improved. Due to its advantages in the economical setting, the AI chatbots can play an
important role in augmenting the already familiar types of service rendering, which would logic
 with optimizing the prices and increasing the availability of the service. This potential for cost
savings is persuading the public and private domains to include these tools in mental health
policies and expanding the accessibility of mental health services [19].

AI chatbots can be considered a promising frontier in mental health care that has the
potential to empower the users, complement traditional therapy, and increase the quality of
mental health resources available on a global scale. These tools could play an indispensable role
in shaping the face of mental health care, making it more inclusive, accessible, adaptable, and
flexible to the ever-changing needs of individuals and communities all across the world.

2. PURPOSE OF THE CHAPTER

This chapter further investigates the technological advancements over the years that
enabled AI chatbots to simulate human conversation in the execution of a range of therapeutic
functions, from immediate support to more specialized interventions, such as CBT. It examines
the growing popularity and acceptance of chatbots and mental health apps like Woebot, Wysa,
MoodKit, and Happify, which deploy AI to support users with mood tracking, coping strategies,
and personalized mental health assessments. These tools have become valuable post-COVID-19,
as mental health needs have intensified, but barriers to traditional care persist all across the
world.

This chapter also reviews the potential challenges and limitations of AI chatbots within mental
health care, including their ability to deliver scalable, round-the-clock support and overcome
stigma-related barriers by offering users privacy and anonymity. By analyzing various use cases,
therapeutic outcomes, and challenges in AI-based mental health interventions, this chapter aims
to present a comprehensive overview of the effectiveness and scalability of chatbot technology in
enhancing mental health outcomes, as well as its role in shaping the future of mental health care
delivery.

3. KEY FEATURES OF AI-BASED MENTAL HEALTH CHATBOTS

3.1 Emotional recognition

Emotions can be recognized by chatbots through the analysis of emoticons exchanged by

the user. If it detects the user is sad, it will generate jokes or uplifting quotes to lighten the
mood. In earlier times, emotions were classified based on different machine learning algorithms
that were applied, like Logistic Regression, Random Forest and Naïve Bayesian with Logistic
regression yielding the most accurate results. With the evolution of AI technology, the models on
which the newer mental health chatbots were based showed better range in detecting emotions
and better accuracy [20].

Language ambiguity is one of the most common problems that an AI chatbot needs to
tackle while detecting emotions from textual content. The users can use general expressions like
I am fine or everything is alright, which seem to denote a positive emotion that masks deeper
underlying negative emotions. To enable fine-tuned emotional detection, chatbots use deep
learning models that are pre-trained on massive datasets containing textual excerpts from social
media posts and transcripts of popular TV shows. Goemotions is a dataset developed by Google
AI that contains 58,000 comments from Reddit, labelled with 28 emotions [21].

Seq2Emo [22] is a model on which mental health chatbot Wysa and other chatbots are
based. At the core of this model is a seq-2-seq framework that uses encoder-decoder
architecture to process input texts and generate output texts. This is combined with a latent
variable chain model , which introduces a latent variable z (encodes global information about
emotions; the model can generalize unseen data, those not covered in the training set), which
conditions the model to better capture the subtle variations in the emotional context. The
encoder uses LSTM, Bi-LSTM or GRU to enable nuanced emotion detection over long texts. The
Seq2Emo model also makes use of pre-trained models like DeepMoji and Elmo embeddings to
enhance emotion detection. Elmo generates different representations of the same words based
on the emotional connotation of previous texts. GloVe is another embedding used that is not
context-specific, unlike Elmo and DeepMoji, to generate a general understanding of sentiments
and capture word association and semantic relationships. The vector representations of the
words (in the input text) generated are passed through a decoder, wherein attention mechanisms
are used to focus on the integral part of the text, and then it is classified as an emotion label.

BERT-CNN [20] is a most recent deep learning model; it is based on ISEAR and
SemEval. This model combines the functions of BERT and CNN as shown in figure 1. The
output of BERT (Bidirectional Encoder Representations Form Transformers) and then passes it
through the layers of Convoluted Neural Network (CNN). It is composed of three main
components.
 o Preprocessing of the dataset
 The BERT Base model processes text through 12 layers of self-attention to generate a
contextual vector representation.
 A CNN, employed as a classifier.

Standard metrics like precision, recall, F1-score, and accuracy are used to evaluate the
model. It was observed that the BERT-CNN model could detect the happy emotion with 98%
precision, and its overall accuracy in detecting emotions is 94.7%.

Fig. 1. Architecture of BERT-CNN Model

3.2 Behavioral interventions

With reference to figure 2, CBT-based chatbots make up the largest share of chatbots
dedicated to mental health. Cognitive restructuring, one of the core components of CBT, is how AI
chatbots support behavioral interventions, wherein chatbots respond with empathy and guide
users through technique to reframe their negative thoughts. With the help of another principle of
CBT, behavioral activation, wherein they suggest small tasks like taking a walk or engaging in a
hobby, chatbots help in breaking the cycle of inactivity that often accompanies depression.
Chatbots also make use of mindfulness training exercises.
 Fig. 2. Distribution of Mental Health Chatbots

Through mindfulness training, chatbots help individuals learn self-regulation skills.

Mindfulness training improves self-observation skills. Mindfulness helps individuals stay in a state of
conscious awareness, enhancing self-regulation skills such as inhibitory control by reducing
automatic reactions [23]. Inhibition control helps individuals control their impulses. Systemic
practice in mindfulness can improve the ability of individuals to engage in self-regulatory
psychological responses, which is related to efficacious down-regulation of stress. Vivibot [24], a
chatbot, delivers positive psychology to students who had just completed their cancer treatment.

PTSD Coach [25] is another mental health chatbot made by the US Department of
Veteran Affairs for war veterans to help with their posttraumatic stress disorder. This app was
successful because of the incorporation of a distress scale consisting of rankings from 0 to 10.
Based on the ranking given by the user, the app will suggest CBT techniques to help them manage
their distress. PTSD Coach also provides personalized coping solutions tailored to the user distress
level, like journaling prompts, music, or art therapy. Ellie is another chatbot created for PTSD by
US DARPA (Defense Advanced Research Project Agency), which depends on its relationship-
building capabilities to gather personal information on the user and then use this information to
recognize the issues the user is going through without them ever explicitly stating so.

In 2019, a mental-health platform, Happify Health, released an AI chatbot called Anna

[26]. Happify Health provided the gamified versions of evidence-based activities drawn from CBT
techniques, positive psychology, and mindfulness-based stress reduction exercises. Anna is also
based on the same techniques and may ask users questions and deliver activities to suggest more
personalized coping strategies and deepen engagement.

MISHA [27] was developed in collaboration with ETH Zurich. MISHA offers a chat-based
interface enriched with multimedia elements and regular notifications to engage users. It
incorporates evidence-based strategies from Cognitive Behavioral Therapy (CBT), mindfulness,
and psychoeducation to provide insights on stressors and effective coping techniques. The stress
management program integrates key CBT elements, including cognitive restructuring, and the
identification, evaluation, and modification of maladaptive thought patterns. Additionally,
 techniques like Behavioral Activation (BA) and activity monitoring from CBT are applied to assist
participants in achieving their goals through a collaborative approach.

Youper, Melinda, XiaoE [28], and Aetna [29] are other chatbots based on CBT,
mindfulness, and the third wave CBT principle, Action and Commitment Therapy (ACT).
Youper allows chat/text-guided psychoeducation, SMART goal setting, as well as tracking
of progress. Aetna is a psychoeducational chatbot for students facing mental health
difficulties which is based on CBT, mindfulness training and positive psychology.

3.3 Privacy and Anonymity

One of the prime concerns when an individual uses AI chatbots is the privacy and
confidentiality of the data. The AI systems on which these chatbots are based collect and analyze a
huge amount of personal data related to the user, and it is very important that the data is handled
in a secure manner and the privacy rights of the user are respected. The users should be able to
have transparent communication, and the AI model should be explainable, and it is of utmost
importance to ensure the consent of the user is received before deploying the AI interventions.
System vulnerabilities are identified by the stride model, which segregates the identified security
threats into spoofing, tampering, information disclosure, denial of service, etc. Data security falls
under 2 categories: the first category is the safe transfer of messages to the server where the
chatbot is hosted; the second category lies on the user end: how the messages and data are
stored, processed, or shared [30].

End-to-end encryption ensures that only the intended recipient can decrypt the message,
preventing unauthorized access by third parties. However, spoofing and tampering can still occur
if a third party gains access to the cryptographic keys. In public key encryption, the user's device
generates two key pairs: a private key and a public key. These key pairs are created using
protocols based on the RSA algorithm. Many chatbots utilize HTTPS, which secures data through
an encrypted connection via SSL (Secure Socket Layer) or TLS (Transport Layer Security). Unlike
end-to-end encryption, HTTPS provides point-to-point encryption, ensuring that data is encrypted
between the client and server but not necessarily end-to-end [31]. Woebot, Replika, and Anna use
this technique.
 In many instances, sensitive Personally Identifiable Information (PII) is transmitted. Self-
destructing messages offer a practical solution, automatically erasing messages after a
predetermined period, without requiring user intervention. This process involves both the chatbot
and the user, ensuring that sensitive data is securely deleted within the set timeframe.
Article 5 (e) of the GDPR states that the personal data cannot be kept longer than
necessary for the purpose for which it is being processed. According to PHI, any information about
mental health status should be collected by a covered identity and not linked to a specific
individual. GDPR compliance requires an intent level of privacy. Article 32(a) of GDPR requires the
mental health chatbot companies to act to minimize and encrypt personal data. Open-domain
chatbots are a little more problematic than closed-domain chatbots. Closed-domain chatbots
working with PII require authentication and authorization, and users agree with terms and
conditions. On the contrary, no such precautions are taken by open-domain chat bots. A popular
mental health chatbot, Replika shares some user data with third-party tracking and analytics tools
like Google Analytics and cookies, which helps the company evaluate the type of customers. Wysa,
Youper, and Tess provide services to users for making complaints under HIPPA guidelines. The
new GDPR mandate requires that any technology handling users' personal data must implement
stringent security measures to prevent malicious encryption activities that could infringe on
individuals' personal rights [31].

3.4 24/7 Accessibility

There are great challenges in mental health care. Accessibility is one of the major barriers.
One of the major benefits of the AI chatbots is that they help enhance accessibility and provide
immediate support in mental health care. Traditional mental health services require
appointments most of the time, resulting in long waiting for periods for patients. AI chatbots, on
the other hand, if used, provide immediate support all round the clock, every day. This non-stop
availability would be essential for some people during times of crisis [32]. Regular monitoring of
sick patients is often considered fruitful as it optimizes their therapy. AI therapy plays an
important role in providing continuous support 24/7. New parents dealing with the ongoing stress
of a chronic illness, combined with the chaos that accompanies parental care, can just turn to the
AI therapists for advice and emotional support whenever needed. Some chronic diseases, such as
diabetes, can benefit from AI therapy, as they offer reminders for medications and lifestyle
modifications, thereby enhancing patient compliance to a great extent. Stressed career
professionals often get overwhelmed, and, due to time constraints, they often find it difficult to
move from their workspace to a traditional therapy session. AI therapists, on several platforms
that are available 24/7, highlight the conveniences of modern-day living by providing on-demand
solutions tailored to the needs of the individual. It takes out the pain of scheduling. With AI
therapy, one can seek help at any time. However late it is, we do not have to wait for office hours,
as help is available at the click of a button [33].

Woebot Health [34] is a 24/7 accessible automated conversational agent. It is an AI

mental health chatbot that provides users with behavioural health solutions using the principle of
CBT. The objective is to let the user talk and speak about how they feel whenever and however
they need it, regardless of the time. This 24/7 availability makes Woebot a dependable
companion, allowing users to check in and speak about their feelings at any moment. The
program will monitor the user moods and fluctuations, for which the chatbot empowers you to
schedule a therapy session every day. Woebot checks in with users every day, prompting them to
reflect on their emotions and experiences. These check-ins serve as a valuable tool for self-
awareness, helping users identify patterns and track their progress over time.

Youper [34] is a 24/7 available mental health support chatbot. It uses artificial intelligence
and machine learning to provide the user with personalized mental health support. The
interactions are based on various techniques involving NLP analysis and analytic data review to
track the progress of users and improve their recommendations. Youper holds real-time
conversations with the users, affording them the opportunity to receive support from wherever
they are and whenever they feel the need. Whenever the user chooses to check in, it provides a
reliable time and space for processing feelings and accessing resources. Youper combines this with
round-the-clock availability and smart personalization for flexibility and responsiveness on the
terms of the users.

Talktotherapist.app [35] offers 24/7 assistance through its AI-driven chatbot named
Sophia, designed to provide mental health support. Sophia is the world’s first chatbot designed
specifically for survivors of domestic violence. It creates a trusted space for survivors to connect
with individuals who can offer support, such as healthcare workers, police officers, and
psychologists. Developed by the Swiss human rights organization Spring ACT, Sophia is available
24/7, providing a listening ear and validation for survivors, regardless of their location. Accessible
from any internet-enabled device, it supports communication in English, French, German, and
Italian.
 This 24/7 accessibility of AI chatbots ensures that we are never left alone during a crisis
and we don’t need to travel or wait for appointments by providing users with immediate
assistance and resources.

3.5 Personalized Conversations

AI-powered mental health chatbots use sophisticated NLP and Deep Learning (DL) to
engage users in meaningful, personalized conversations that feel responsive and human-like. NLP
is what allows these chatbots to extract from text about the user intentions, emotional tones, and
other context-dependent meanings. The Seq2Seq deep learning model, based on Recurrent
Neural networks (RNNs), plays a pivotal role by remembering prior conversation parts, allowing
the chatbot to maintain conversational context. In practice, this model uses an encoder to
convert the input of the user into a vector summarizing their message and a decoder to generate
a response, building on the previous dialogue and providing users with relevant, coherent
answers. For example, if a user expresses anxiety, the chatbot will retain that context to deliver
appropriate support throughout the conversation.

The effectiveness of the chatbot depends a lot on the training data, which includes large
datasets of past mental health-related conversations, such as anonymized therapy session
transcripts or user interactions. These datasets teach the chatbot to recognize common patterns
in expressions of mental states like sadness or stress, allowing it to adjust its responses to be
empathetic and supportive. This training, however, limits the ability of the chatbot to handle
unfamiliar or complex issues outside the scope of the dataset, as it can only respond to topics it
has learned. Many companies gather and refine data to enhance chatbot responses in specific
areas, improving effectiveness over time. However, chatbots still perform best within certain
domains where they have substantial training data and may struggle with topics outside their
training.

They give real-time support and foster a secure, private space that will help users feel
comfortable enough to provide sensitive information in particular. Yet there are major ethical
issues around personal data and consent to be taken into consideration. The bots should be able
to identify situations that require a human, directing users to appropriate professional help when
necessary [36].
 Woebot [37] provides users with personalized conversations by tailoring its responses to the
mental health needs of each individual, with a specific focus on alleviating symptoms of depression
and anxiety. By utilizing CBT principles and NLP, Woebot initiates with an assessment to understand
the user moods and current challenges. Then, based on user mental state and recent interaction, it
crafts the messages so users may know specific coping strategies, undertake exercises, and track
their moods. For example, if a user often brings up anxiety, then Woebot may suggest exercises
addressing anxious thoughts and thus build a custom dialogue in response to the unique experiences
of the individual over time. This allows the Woebot to grow alongside the user, where the system
provides instant support that only gets more individualized as the user continues their journey.

Wysa [37] also provides users with personalized conversations by integrating CBT and
mindfulness techniques into its interactions, designed to address stress, anxiety, and emotional well-
being. Beginning with a conversational assessment, Wysa identifies the main challenges of each user
and curates mindfulness exercises, guided journaling prompts, or other tools to support them. If a
user often expresses feeling overwhelmed, adjust the responses to emphasize stress management
resources and resilience-building exercises. This allows Wysa to create a customized support
experience that feels tailored to each individual, delivering time-to-time adaptive guidance that
evolves with the needs of the user over time. Through ongoing adaptation, Wysa provides
continuous relevant support, bridging gaps between formal therapy sessions and fostering a
sustained therapeutic connection.

Table 1: Summary of the above mentioned chatbots with the corresponding features they possess
4. EFFECTIVENESS AND IMPACT ON STUDENT MENTAL HEALTH

AI-based chatbots offer accessible mental health for anxiety and stress. They provide 24/7
interventions, helping overcome barriers like geography, cost, and stigma. However, challenges
persist, including a lack of empathy and privacy concerns. Depression and anxiety affect 280 million
and 301 million people, worsened by the COVID-19 pandemic [38]. While traditional treatments like
CBT are often inaccessible, AI chatbots like Woebot and Wysa utilize machine learning and natural
language processing to deliver effective, personalized support [39]. Research shows promise, but
further studies are needed to evaluate their long-term efficacy in treating mental health conditions
[33].

Effectiveness on Depressive Symptoms

Meta-analysis indicates that AI-based chatbots reduce depressive symptoms, with effects
comparable to traditional therapy. By engaging users in therapeutic dialogues, chatbots help
restructure cognition and address issues like pessimism and anhedonia [34]. Their ability to form
therapeutic alliances enhances effectiveness, simulating human interactions. However, the placebo
effect associated with user expectations necessitates careful interpretation of results. Chatbot
interventions show effectiveness within four to eight weeks, driven by user engagement and CBT
principles, reducing treatment time by up to 40%. While results are strong in the short term, long-
term benefits remain unclear, with improvements not lasting beyond three months. This suggests
potential limitations in maintaining recovery and preventing relapse.

Effectiveness on Anxiety Symptom

AI-based chatbots provide modest improvements in anxiety symptoms, often offering
quicker relief compared to traditional therapy [40]. Their 24/7 availability allows for faster symptom
reduction, like intensive CBT. However, they are less effective than traditional psychotherapy and
pharmacological treatments [41]. Given anxiety complexity, an integrated treatment approach is
essential. While chatbots can employ CBT techniques, their limited capacity to address emotional
nuances reduces overall effectiveness for anxiety disorders [42].

Mental Health Allies in COVID-19

The COVID-19 pandemic has significantly affected the mental health of young people,
leading to increased levels of stress, anxiety, and depression. Studies [43] highlight AI chatbots
potential in addressing these issues found that XiaoE, a CBT-based chatbot for college students,
resulted in moderate reductions in depressive symptoms after one week, although smaller at one-
month follow-up. Participants valued its inner support but noted limitations like rigidity. Peuters et.
al., [44] evaluated the LIFEGOALS Health intervention, which included an AI chatbot, showing
positive effects on physical activity, sleep, and mood, especially during in-person schooling. While
engagement was challenging, users appreciated gamification and personalized information. These
findings suggest AI chatbots can support mental health during crises, but their long-term efficacy and
engagement strategies require further research.

Supporting Emotional Well-Being in Health Conditions

A study by Yasukawa et al. [45] explored the potential of AI chatbots in assisting individuals
with specific health conditions. The researchers found that the AI chatbot significantly improved
adherence to internet-based cognitive-behavioral therapy (iCBT) among workers with subthreshold
depression, primarily boosting engagement rather than directly affecting clinical outcomes. Ogawa
et. al. [46] examined an AI chatbot for Parkinson’s disease patients, showing improvements in facial
expressivity and speech fluency, suggesting its role in remote symptom monitoring. Ulrich et. al., in
another study [47], developed the BalanceUP app, which uses a chatbot interface for migraine
management and to an improved mental wellbeing to a large extent, indicating its effectiveness as a
standalone digital intervention.

Combatting Substance Use and Addiction

AI chatbots have emerged as effective interventions for substance use and addiction. Vereschagin et.
al. [48] evaluated the Minder app, which integrates an AI chatbot delivering CBT, leading to reduced
anxiety, improved mental wellbeing, and decreased substance use among university students.
Prochaska et. al. [49] studied Woebot for substance use disorders and found significant
improvements in substance use and cravings over an 8-week program, indicating its scalability for
addiction treatment. So et. al. [50] compared guided versus unguided chatbot interventions for
problem gambling, revealing that both conditions achieved significant improvements, emphasizing
the effectiveness of standalone chatbots. Olano-Espinosa et. al. [51] found that the Dejal@bot
chatbot outperformed usual care in smoking cessation, achieving a continuous abstinence rate of
26.0%. These findings suggest AI chatbots have potential as accessible, scalable interventions in
mental health.

AI Chatbots in Preventive Health Interventions

AI chatbots have been explored for preventive health interventions, with eating disorder
prevention and HIV testing being the most impacted. Fitzsimmons-Craft et. al. [52] evaluated the
chatbot Tessa, which well reduced weight and shape concerns among women at high risk for eating
disorders, despite engagement challenges. Cheah et. al. [53] tested an AI chatbot for HIV prevention
among men who have sex with men (MSM) in Malaysia, finding it feasible and acceptable for
providing confidential HIV service information. Participants valued its privacy features but suggested
improvements, highlighting the potential of AI chatbots for targeted health promotion.

Panic Disorder Management

In the area of managing panic disorder, Oh et. al. [54] compared a randomized controlled
trial comparing a mobile app-based interactive CBT chatbot to a paperback book. The chatbot group
showed significant reductions in panic severity after four weeks, despite receiving lower usability
ratings. Qualitative feedback indicated advantages like interactive learning and self-management
tools.

Enhancing Problem Solving in Older Adults

Bennion et. al. [55] studied AI chatbots for problem solving among older adults, comparing
MYLO and ELIZA. Both reduced problems like distress and anxiety to a large extent, with MYLO being
preferred. Usability ratings were below acceptable thresholds, indicating a need for improvement.

Usability and Engagement

 Thunström et. al. [56], in one of their studies, evaluated the usability of a digital human
(BETSY) versus a text-based chatbot. Results showed higher usability for the text-only version,
suggesting further refinement of AI chatbots to enhance user experience and engagement. These
studies as a collection underscore the potential and need for improvement in AI chatbot applications
across various health contexts.

Using AI chatbots to provide self-help depression interventions for university

students: A case study

Background: Depression impacts a significant number of university students, and mobile-based

therapy chatbots are increasingly being used to support young adults dealing with depression.
However, previous trials have had limited follow-up periods, and there is still a lack of evidence
regarding their effectiveness in real-world conditions.

Objective: This study intends to compare chatbot therapy to bibliotherapy, an evidence-based,

accepted self-help psychological intervention. Adding the present study to the existing body of
evidence for effectiveness helps establish chatbot therapy as a convenient, accessible, and low-cost
yet interactive self-help treatment of depression.

Methods: The trial was an unblinded randomized controlled study involving 83 university students
as participants. Participants were divided into either a test group receiving a brand new developed
intervention via a chatbot or a control group in the form of minimal level bibliotherapy, with a size of
n = 41 and n = 42, in the order given. A series of questionnaires were utilized to assess clinical
variables at baseline and every 4 weeks for 16 weeks. These included the Patient Health
Questionnaire-9 (PHQ-9), Generalized Anxiety Disorder Scale (GAD-7), and the Positive and Negative
Affect Scale (PANAS). To evaluate satisfaction and the therapeutic alliance, the Client Satisfaction
Questionnaire-8 (CSQ-8) and the Working Alliance Inventory-Short Revised (WAI-SR) were
administered following the intervention. Additionally, participants provided self-reported data on
adherence and feedback regarding the therapy chatbot.
Interventions: A therapy chatbot named XiaoNan was specifically developed for the trial and made
accessible through the WeChat Official Accounts Platform. Users can access the chatbot using the
smartphone app WeChat on Windows, macOS, Android, and iOS.

XiaoNan operates pipeline-based chatbot powered by the open-source conversational AI

RASA. Its content is based on principles of Cognitive Behavioral Therapy (CBT4), which were
reviewed and approved by professional therapists.The chatbots accepts input as either text or voice
messages. Voice messages are converted to text using a natural speech recognition service provided
by the IFLYTEK Open Platform. This next moves the text to XiaoNan, which would pass on this text
for natural language understanding. The language understanding module was divided into three
different ML models: NLP, intention classification, and emotion recognition. The model on NLP
collected all that information relevant to the needs of users in handling conversations, which the
paper describes as entities, slots, and forms. The intention classification and emotion recognition
models further analyze the input text, assigning predefined intention and emotion labels. The
dialogue management module then determines the appropriate response using a pre-structured
template, referred to as a "domain." This domain includes predefined entities, slots, forms, rules,
and stories, all of which guide response generation. Entities, slots, and forms store dynamic data
necessary for crafting responses. Rules enforce mandatory actions, such as generating a farewell
message when a conversation is ending. Stories, on the other hand, provide broader response
strategies based on logical connections, aligning with CBT principles. All models were trained to
achieve at least 90% classification accuracy.

XiaoNan is designed to assist users in identifying and separating their emotions, thoughts,
reactions, and behaviors, akin to sessions with a CBT professional. Additionally, it allows users to
track their emotions daily or engage in casual conversations, offering empathetic interactions in line
with its therapeutic purpose.

Bibliotherapy intervention: Bibliotherapy is a key form of self-help therapy that utilizes literature to
address and alleviate a patient's issues. In this approach, patients engage in psychological
intervention by reading carefully selected literature, guided by the recommendations of
professionals.Previous studies have shown evidence of the effectiveness of bibliotherapy in the
treatment of moderate depression. A systematic review found that the long-term benefits of
bibliotherapy were confirmed through randomized clinical trials. Notably, even after a three-month
follow-up period, adult participants who engaged in cognitive bibliotherapy demonstrated sustained
improvement.

Fig. 3. Flowchart of chatbot (A) The workflow of the chatbot XiaoNan.

(B) The structure of the chatbot, XiaoNan.

Fig. 4. Examples of using the chatbot. (A) Both text and voice messages are supported. First-time
users will receive instructions, and they can select options from the list by either clicking the text
or responding with the corresponding number or content. (B) An example of CBT treatment where
the chatbot analyzes, evaluates, and addresses negative emotions based on the user's input text.
(C) A question-answer system focused on exploring depression, providing information about
depressive disorders.

Randomization: Given the unique character of chatbot interventions as well as of bibliotherapy

interventions, blinding was not applicable in any way because participants could sense them.
Participants were randomly assigned a number 0<n≤1 generated using the random number
generator in Statistical Package for the Social Sciences (SPSS) version 26. Those with 0<n≤0.50 were
allocated to the chatbot test group, while the others were placed in the control group.

Statistical methods: The analysis was conducted using SPSS version 26, with the significance level set
at 0.05. Baseline variables, including age, gender, education, and initial clinical metrics, were
evaluated using ANOVA and chi-square tests to identify any significant differences between the
groups. This ensured the groups were comparable at the start of the study. Primary outcomes were
evaluated using univariable ANCOVA within the intention-to-treat (ITT) framework, accounting for
the univariate impact of group assignment on results while adjusting for baseline clinical variables.
Missing data, presumed to be missing at random, were managed through SPSS's multiple imputation
procedure. Cohen's d effect sizes were calculated to quantify effect magnitude. Independent t-tests
were conducted on the CSQ-8 and WAI-SR scores, along with self-reported adherence measures.
Additionally, responses to two open-ended questions were analyzed using a word frequency
analysis.

Result: The chatbots proved to be an effective medium for delivering self-help depression
treatment in real-world, pragmatic settings. The therapy chatbot led to a substantial reduction in
depression symptoms, as measured by PHQ-9, over a 16-week period. Additionally, it achieved a
modest reduction in anxiety, as assessed by GAD-7, within the initial 4 weeks. The chatbot
intervention demonstrated greater effectiveness compared to bibliotherapy. Both groups
demonstrated similar levels of client satisfaction and self-reported adherence. However, the chatbot
group experienced a downward trend in adherence rates, potentially attributable to technical and
content-related shortcomings in the chatbot design.Scores achieved in the WAI-SR test by users of
the chatbot were significantly higher compared to users of the chatbot, hence indicating that using
conversational AI can contribute to building a therapeutic alliance. Some of the previous studies
were in line with the present result, in which there were shorter periods of intervention, such as 2
and 4 weeks. An increase in self-disclosure was observed, as reflected in user feedback, where
participants expressed a greater tendency to share personal thoughts and feelings.Moreover, based
on the suggestions for XiaoNan, it indicates that more process factors are greater than content.

5. CHALLENGES

Mental health chatbots offer accessible support for individuals experiencing mental health
challenges.

5.1. Risk of Inadequate Responses

Unlike human therapists, mental health chatbots often lack nuanced understanding and
contextual awareness, which can lead to inadequate responses in critical situations. For instance,
generative AI models sometimes respond in an insensitive way to distress signals, risking harm
rather than providing support. Such responses may exacerbate the emotional distress of a user, and
without immediate human intervention, these interactions can pose safety concerns [57].

5.2. Trust and Privacy Concerns

Users express concerns about privacy and data security, which are given that chatbot
interactions are often recorded for machine learning improvements. In addition, the users may feel
uncertain about sharing personal issues with an AI, impacting the effectiveness of the chatbot as a
mental health tool. Transparent data use and improved security protocols could help alleviate these
concerns [58].

5.3. Over-reliance and Limited Scope

While chatbots provide convenience, there is a risk that users may over-rely on them for
support. It has the potential to delay essential professional mental health care. Even with state-of-
the-art AI, the breadth of therapeutic interventions provided by a trained professional cannot be
emulated by a chatbot. Further, they are bound to a very specific therapeutic model, for example,
CBT or BA, and may not be directed toward a range of needs that the client may be presenting [59].

5.4. Technical and Emotional Limitations

Many chatbots are based on pre-scripted responses, which can limit their ability to engage in
dynamic conversation or adapt to complex emotional states. This restriction can make interactions
feel less genuine and may reduce engagement over time. To handle this, advanced AI systems are in
development that would use NLP to identify emotions; they still fall significantly short, though, in the
expression of true empathy [60].

5.5. Lack of Regulatory Oversight

Unlike human providers, chatbots do not represent a class to which similar standards have
been specifically applied. Without standard guidelines or ethics oversight, there is no assurance that
these chatbots meet minimum quality and safety standards regarding mental health care. Setting up
the regulatory frameworks that guide them in developing and implementing chatbots is an
important process to make sure user welfare is guaranteed [61].

5.6. Case study

5.6.1. Case of Replika and Ethical Concerns

 Replika, a popular AI chatbot designed as a virtual friend, faced challenges when users
turned to it for mental health support despite it not being intended as a therapy tool. Some users
reported that the responses of Replica were unempathetic or even harmful during conversations
about depression or self-harm[62]. In one instance, a user noted that Replika gave insensitive replies
to statements of distress, raising ethical questions about the use of AI for sensitive mental health
needs. This case underlined the limitations of current generative AI in handling mental health crises
and highlighted the need for boundaries and clearer guidelines for chatbots intended for general
companionship but used for therapeutic purposes.

5.6.2. Woebot and Emotional Support Limitations

The Woebot chatbot, based on the principles of CBT for mental health conditions, has
enjoyed favor but also criticism in specific cases where automated responses were beyond the needs
of users. One of the chapters among these pointed to specific cases of complex mental health
conditions that include bipolar disorders, which find inadequate help through Woebot in terms of
their particular and even complicated emotional needs. Although Woebot helped some users reduce
symptoms of depression and anxiety, it struggled to offer personalized responses to individuals with
severe or comorbid conditions. This highlighted the limitations of one-size-fits-all chatbot models
and underscored the importance of clear disclaimers in the scope and capabilities of the chatbot[63].

5.6.3. Cleverbot and Crisis Situations

Cleverbot, an old conversational AI, demonstrated the challenges of AI responses in serious
mental health scenarios. In a documented interaction, Cleverbot responded in an insensitive manner
to a user expressing suicidal thoughts, exacerbating the distress of the user [64]. This case became a
cautionary example of the risks posed by generative AI chatbots when they lack the ability to detect
crisis language or provide appropriate responses in emergency mental health situations. Following
this, the need for crisis-intervention protocols in AI chatbots was emphasized, focusing on those that
might engage in mental health discussions by accident.

6. CONCLUSIONS AND FUTURE DIRECTIONS

AI-based chatbots democratize access to mental health care because it can be accessed
whenever needed, scalable, and cost-effective. However, several key R&D fronts need to mature
further for enhanced potential and impact. Rule-based AI chatbots offer a solution to some of the
limitations mentioned earlier by simulating human interactions through pre-written scripts and
algorithms, such as decision trees. For instance, two well-known chatbots, Woebot and Wysa, have
shown effectiveness in improving depression symptoms in users [34] [39] and foster collaborative
relationships compared to those created by human therapists [65]. Policy-based chatbot applications
are promising in terms of user engagement, with amazing app stores [66] [67] and good research
showing that users appreciate the same connection and social support as humans [66] [67] [68] [69]
[70]. Despite these encouraging signs, however, policy-based AI chatbots still fail to realize the full
potential of DMHI.

Recent developments in AI intelligence technologies, such as LLMs, have new capabilities.

Unlike rule-based AI chatbots, generative AI chatbots such as ChatGPT of OpenAl, Gemini of Google,
and Pi of Inflection are trained with a lot of data [71], which makes them capable of understanding
and generating messages. These standards meet or exceed human performance in many areas,
including clinical counseling [72], effective communication, theory, human behavior, responding to
social problems, and helping people reconstruct negativity [73]. Conditions that help reduce
negative emotions are essential for improving mental well-being. Additionally, user engagement has
been notable, with ChatGPT reaching 100 million weekly users within a year of its launch, and nearly
half of Americans now utilizing AI technologies [74] [76]. The potential of AI presents a significant
opportunity for digital health, with media reports highlighting the increasing consumer use of these
technologies. Analyses suggest that AI chatbots are more effective than traditional chatbots in
alleviating mental health issues. One study indicated that the use of ChatGPT had a positive impact
on the treatment of mental illness [77]. Given the novelty of generative AI and its nature, good space
research is an important starting point in creating a good understanding of the knowledge to be
pursued in quantitative research [78]. The best research published to date includes a summary of
user forum comments on generative AI and DMHI policy, student responses to peer-centered
generative AI chatbots, and responses solicited for testing. To our knowledge, no research to date
has used semi-structured interviews and thematic interventions to investigate how people use AI
chatbots to improve their mental health and well-being in a non-invasive, non-directive, scientific
research environment [79]. This chapter aims to fill this gap and provide researchers, platform
developers, and practitioners with an understanding of the implications of using this new technology
in mental health treatment.

Future research should focus on conducting large-scale, rigorously designed randomized

controlled trials to assess the long-term effectiveness and cost-effectiveness of AI chatbot
interventions in comparison to conventional treatments or other active therapies [80][81].
Additionally, there is a need for studies exploring the optimal integration of human support with AI
technology within existing healthcare systems. Furthermore, the successful development of AI
chatbots for mental health requires collaboration across various disciplines, including computer
science, psychology, healthcare, and ethics.Using chatbots to provide more holistic and effective
solutions based on perspectives and experiences [82] [83] [84].

7. REFERENCES

1. “The Latest Mental Health Statistics: What the Numbers Say About the State of Our Minds in
2024.” Huntington Psychological Services, available online: ate-of-our-minds-in-2024.
Accessed 14 Nov. 2024.

2. "World Health Organization Mental Health." World Health Organization, available online:
https://www.who.int/health-topics/mental-health. Accessed 14 Nov. 2024.

3. Javaid, Syed Fahad, et al. "Epidemiology of anxiety disorders: global burden and
sociodemographic associations." Middle East Current Psychiatry, vol. 30, 2023, p. 44.

4. “COVID-19 Pandemic Triggers 25% Increase in Prevalence of Anxiety and Depression

Worldwide.” World Health Organization, available online:
https://www.who.int/news/item/covid-19-pandemic-triggers-25-increase-in-prevalence-of-
anxiety-and-depression-worldwide. Accessed 14 Nov. 2024.

5. First, Michael B. Treatment of Mental Illness. MSD, 2024.

6. Patel, Vikram, et al. "Reducing the Treatment Gap for Mental Disorders: A WPA Survey."
World Psychiatry, vol. 9, no. 3, 2013.

7. “WHO Report Highlights Global Shortfall in Investment in Mental Health.” World Health
Organization, available online: th. Accessed 14 Nov. 2024.

8. Kapur, N., Gorman, L. S., Quinlivan, L., and Webb, R. T. "Mental Health Services: Quality,
Safety, and Suicide." BMJ Quality & Safety, vol. 31, no. 6, 2022, pp. 419–422.

9. Caponnetto, P., and Milazzo, M. "Cyber Health Psychology: The Use of New Technologies at
the Service of Psychological Well-Being and Health Empowerment." Health Psychology
Research, vol. 7, no. 2, 2019, p. 8559.

10. Taylor, C. B., Graham, A. K., Flatt, R. E., Waldherr, K., and Fitzsimmons-Craft, E. E. "Current
State of Scientific Evidence on Internet-Based Interventions for the Treatment of Depression,
Anxiety, Eating Disorders, and Substance Abuse: An Overview of Systematic Reviews and
Meta-Analyses." European Journal of Public Health, vol. 31, 2021, pp. I3–I10.

11. Spiegel, B. M. R., Liran, O., Clark, A., Samaan, J. S., Khalil, C., Chernoff, R., Reddy, K., and
Mehra, M. "Feasibility of Combining Spatial Computing and AI for Mental Health Support in
Anxiety and Depression." NPJ Digital Medicine, vol. 7, no. 1, 2024, p. 22.
12. Caldarini, G., Jaf, S., and McGarry, K. "A Literature Survey of Recent Advances in Chatbots."
Information, vol. 13, no. 1, 2022, p. 41.

13. Bala, Kumkum, Mukesh Kumar, Sayali Hulawale, and Sahil Pandita. "Chat-Bot for College
Management System Using AI." International Research Journal of Engineering and
Technology (IRJET), 2018, pp. 2395–0056.

14. Moore, J. R., and Caudill, R. "The Bot Will See You Now: A History and Review of Interactive
Computerized Mental Health Programs." The Psychiatric Clinics of North America, vol. 42, no.
4, 2019, pp. 627–634.

15. Abd-alrazaq, Alaa A., Mohannad Alajlani, Ali Abdallah Alalwan, Bridgette M. Bewick, Peter
Gardner, and Mowafa H. "An Overview of the Features of Chatbots in Mental Health: A
Scoping Review." International Journal of Medical Informatics, vol. 132, 2019, p. 103978.

16. Paulus, M. P., Huys, Q. J., and Maia, T. V. "A Roadmap for the Development of Applied
Computational Psychiatry." Biological Psychiatry: Cognitive Neuroscience and Neuroimaging,
vol. 1, no. 5, 2016, pp. 386–392.

17. Moilanen, J., van Berkel, N., Visuri, A., Gadiraju, U., van der Maden, W., and Hosio, S.
"Supporting Mental Health Self-Care Discovery Through a Chatbot." Frontiers in Digital
Health, vol. 5, 2023, p. 1034724.

18. Chakraborty, C., Pal, S., Bhattacharya, M., Dash, S., and Lee, S. S. "Overview of Chatbots with
Special Emphasis on Artificial Intelligence-Enabled ChatGPT in Medical Science." Frontiers in
Artificial Intelligence, vol. 6, 2023, p. 1237704.

19. COVID-19 Mental Disorders Collaborators. "Global Prevalence and Burden of Depressive and
Anxiety Disorders in 204 Countries and Territories in 2020 Due to the COVID-19 Pandemic."
The Lancet, vol. 398, no. 10312, 2021, pp. 1700–1712.

20. Abas, A. R., Elhenawy, I., Zidan, M., and Othman, M. "BERT-CNN: A Deep Learning Model for
Detecting Emotions from Text." Computers, Materials & Continua, vol. 71, no. 2, 2022, pp.
2943–2961, https://doi.org/10.32604/cmc.2022.021671.

21. CERC 2021: Collaborative European Research Conference, 9–10 Sept. 2021, Cork, Ireland.

22. Huang, C., Trabelsi, A., Qin, X., Farruque, N., and Zaiane, O. R. "Seq2Emo for Multi-Label
Emotion Classification Based on Latent Variable Chains Transformation." ArXiv,
abs/1911.02147, 2019.

23. Mitsea, E., Drigas, A., and Skianis, C. "Digitally Assisted Mindfulness in Training Self-
Regulation Skills for Sustainable Mental Health: A Systematic Review." Behavioral Sciences
(Basel, Switzerland), vol. 13, no. 12, 2023, p. 1008.

24. Greer, S., Ramo, D., Chang, Y. J., Fu, M., Moskowitz, J., and Haritatos, J. "Use of the Chatbot
'Vivibot' to Deliver Positive Psychology Skills and Promote Well-Being Among Young People
After Cancer Treatment: Randomized Controlled Feasibility Trial." JMIR mHealth and
uHealth, vol. 7, no. 10, 2019, p. e15018.

25. Gamble, A. "Artificial Intelligence and Mobile Apps for Mental Healthcare: A Social
Informatics Perspective." Aslib Journal of Information Management, vol. 72, no. 4, 2020, pp.
509–523.
26. Boucher, E. M., et al. "Artificially Intelligent Chatbots in Digital Mental Health Interventions:
A Review." Expert Review of Medical Devices, vol. 18, sup. 1, 2021, pp. 37–49.

27. Ulrich, S., Lienhard, N., Künzli, H., and Kowatsch, T. "A Chatbot-Delivered Stress
Management Coaching for Students (MISHA App): Pilot Randomized Controlled Trial." JMIR
mHealth and uHealth, vol. 12, 2024, p. e54945.

28. Casu, Mirko, Sergio Triscari, Sebastiano Battiato, Luca Guarnera, and Pasquale Caponnetto.
"AI Chatbots for Mental Health: A Scoping Review of Effectiveness, Feasibility, and
Applications." Applied Sciences, vol. 14, no. 13, 2024, p. 5889.

29. Ab Razak, N. I., Muhammad Yusoff, M. F., and Rahmat, R. W. O. K. "ChatGPT Review: A
Sophisticated Chatbot Model in Medical and Health-Related Teaching and Learning." BMSC,
2023.

30. Saeidnia, Hamid Reza, et al. "Ethical Considerations in Artificial Intelligence Interventions for
Mental Health and Well-Being: Ensuring Responsible Implementation and Impact." Social
Sciences, vol. 13, no. 7, 2024, p. 381.

31. Hasal, M., et al. "Chatbots: Security, Privacy, Data Protection, and Social Aspects."
Concurrency and Computation: Practice and Experience, vol. 33, no. 19, 2021.

32. Miner, Adam S., et al. "Talking to Machines About Personal Mental Health Problems." JAMA,
vol. 318, no. 13, 2017, pp. 1217–1218.

33. Vaidyam, Anmol N., et al. "Chatbots and Conversational Agents in Mental Health: A Review
of the Psychiatric Landscape." Canadian Journal of Psychiatry, vol. 64, no. 7, 2019, pp. 456–
464.

34. Fitzpatrick, Kathleen K., et al. "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 Mental Health, vol. 4, no. 2, 2017, p. e19.

35. Viduani, Anna, et al. "Chatbots in the Field of Mental Health: Challenges and Opportunities."
Advances in Experimental Medicine and Biology, 2023, doi:10.1007/978-3-031-10698-9_8.

36. Hill, Jennifer, et al. "Real Conversations with Artificial Intelligence: A Comparison Between
Human–Human Online Conversations and Human–Chatbot Conversations." Computers in
Human Behavior, vol. 49, 2015, pp. 245–250, doi:10.1016/j.chb.2015.02.026.

37. Daley, Kate, et al. "Preliminary Evaluation of the Engagement and Effectiveness of a Mental
Health Chatbot." Frontiers in Digital Health, vol. 2, 2020, p. 576361,
doi:10.3389/fdgth.2020.576361.

38. Xiong, Jiaqi, et al. "Impact of COVID-19 Pandemic on Mental Health in the General
Population: A Systematic Review." Journal of Affective Disorders, vol. 277, 2020, pp. 55–64.

39. Inkster, Becky, et al. "An Empathy-Driven, Conversational Artificial Intelligence Agent (Wysa)
for Digital Mental Wellbeing: Real-World Data Evaluation Mixed-Methods Study." JMIR
mHealth and uHealth, vol. 6, no. 11, 2018, p. e12106, doi:10.2196/12106.
40. Fulmer, Rachel, et al. "Using Psychological Artificial Intelligence (Tess) to Relieve Symptoms
of Depression and Anxiety: Randomized Controlled Trial." JMIR Mental Health, vol. 5, no. 4,
2018, p. e64, doi:10.2196/mental.9782.

41. Bickmore, Timothy W., et al. "Automated Interventions for Multiple Health Behaviors Using
Conversational Agents." Patient Education and Counseling, vol. 92, no. 2, 2013, pp. 142–148,
doi:10.1016/j.pec.2013.05.011.

42. Abd-Alrazaq, Anas A., et al. "Effectiveness and Safety of Using Chatbots to Improve Mental
Health: Systematic Review and Meta-Analysis." Journal of Medical Internet Research, vol. 22,
no. 7, 2020, p. e16021.

43. He, Liang, et al. "Effectiveness of an AI Chatbot Based on Cognitive Behavioral Therapy for
Mental Health Support Among College Students During the COVID-19 Pandemic." Journal of
Affective Disorders, vol. 289, 2021, pp. 123–131.

44. Peuters, H., et al. "Impact of an mHealth Intervention with a Chatbot Component on
Lifestyle Factors Among Adolescents During COVID-19: The LIFEGOALS Study." Journal of
Medical Internet Research, vol. 24, no. 3, 2022, p. e29155.

45. Yasukawa, A., et al. "Enhancing Adherence to Internet-Based Cognitive–Behavioral Therapy

for Subthreshold Depression Among Workers Using an AI Chatbot: A Randomized Controlled
Trial." Journal of Affective Disorders, vol. 295, 2021, pp. 191–198.

46. Ogawa, T., et al. "Evaluation of an AI-Driven Chatbot for Remote Symptom Monitoring in
Patients with Parkinson’s Disease: Effects on Facial Expressivity and Speech Fluency." Journal
of Parkinson’s Disease, vol. 12, no. 3, 2022, pp. 745–754.

47. Ulrich, M., et al. "Effectiveness of the BalanceUP App, an AI-Based Chatbot for Mental
Wellbeing Improvement in Migraine Patients: A Pilot Study." Digital Health, vol. 8, 2022, p.
20552076221084191.

48. Vereschagin, V., et al. "Evaluating the Minder App: An AI-Based Chatbot for Cognitive–
Behavioral Therapy in Substance Use Reduction Among University Students." Journal of
Substance Abuse Treatment, vol. 125, 2021, p. 108311.

49. Prochaska, J. J., et al. "Effectiveness of the Woebot Chatbot for Substance Use and Craving
Reduction in an 8-Week Program: A Randomized Controlled Trial." Addiction, vol. 117, no. 1,
2022, pp. 180–190.

50. So, M., et al. "A Comparison of Guided Versus Unguided AI Chatbot Interventions for
Problem Gambling: A Randomized Controlled Study." Journal of Gambling Studies, vol. 38,
no. 2, 2022, pp. 563–580.

51. Olano-Espinosa, E., et al. "Efficacy of the Dejal@bot AI Chatbot Versus Usual Care for
Smoking Cessation in Primary Care: A Randomized Controlled Trial." Tobacco Control, vol.
30, no. 5, 2021, pp. 560–566.

52. Fitzsimmons-Craft, Emily E., et al. "Efficacy of a Chatbot-Delivered Intervention to Reduce

Weight and Shape Concerns in Women at High Risk for Eating Disorders: A Randomized
Controlled Trial." International Journal of Eating Disorders, vol. 54, no. 6, 2021, pp. 1039–
1050.
53. Cheah, W. L., et al. "Feasibility and Acceptability of an AI Chatbot for HIV Prevention Among
Men Who Have Sex with Men in Malaysia: A Pilot Study." AIDS and Behavior, vol. 26, no. 3,
2022, pp. 811–822.

54. Oh, S., et al. "Effectiveness of a Mobile App-Based Interactive Cognitive Behavioral Therapy
Chatbot Compared to a Paperback Book in Managing Panic Disorder: A Randomized
Controlled Trial." Journal of Anxiety Disorders, vol. 85, 2022, p. 102524.

55. Bennion, M. R., et al. "The Role of Artificial Intelligence Chatbots in Problem Solving for
Older Adults: A Comparative Study of MYLO and ELIZA." Aging & Mental Health, vol. 25, no.
8, 2021, pp. 1440–1448.

56. Thunström, L., et al. "Usability Comparison of a Digital Human (BETSY) Versus a Text-Based
Chatbot: A Pilot Study on Enhancing User Experience and Engagement." Computers in
Human Behavior, vol. 127, 2022, p. 107058.

57. Van Dyke, Thomas, et al. "The Effect of Consumer Privacy Empowerment on Trust and
Privacy Concerns in E-Commerce." Electronic Markets, vol. 17, no. 1, 2007, pp. 68–81.

58. Coghlan, Simon, et al. "To Chat or Bot to Chat: Ethical Issues with Using Chatbots in Mental
Health." Digital Health, vol. 9, Jan. 2023, p. 20552076231183542.

59. Hauer, Tomas. "Importance and Limitations of AI Ethics in Contemporary Society."

Humanities and Social Sciences Communications, vol. 9, no. 1, Aug. 2022, p. 272.

60. Diaz, Anthony, and Daehan Kwak. "Leveraging Chatbots for Mental Health Enhancement."
2023 International Conference on Computational Science and Computational Intelligence
(CSCI), IEEE, 2023, pp. 137–143.

61. Tian, Wensheng, et al. "A Privacy-Preserving Framework for Mental Health Chatbots Based
on Confidential Computing." 2022 IEEE Smartworld, Ubiquitous Intelligence & Computing,
Scalable Computing & Communications, Digital Twin, Privacy Computing, Metaverse,
Autonomous & Trusted Vehicles (SmartWorld/UIC/ScalCom/DigitalTwin/PriComp/Meta),
IEEE, 2022, pp. 1119–1124.

62. Agbih, S. "Housing Facilities for Asylum Seekers in Germany: Ethical Concerns Regarding
Social Exclusion, Othering and Negative Effects on Health." European Journal of Public
Health, vol. 29, no. Supplement_4, Nov. 2019, p. ckz185.174.

63. Knightbridge, Stephen M., et al. "Using Participatory Action Research in a Community-Based
Initiative Addressing Complex Mental Health Needs." Australian & New Zealand Journal of
Psychiatry, vol. 40, no. 4, Apr. 2006, pp. 325–332.

64. Griffiths, P. D., et al. "Post-Mortem Confirmation of Fetal Brain Abnormalities: Challenges
Highlighted by the MERIDIAN Cohort Study." BJOG: An International Journal of Obstetrics &
Gynaecology, vol. 128, no. 7, June 2021, pp. 1174–1182.

65. Beatty, C., et al. "Evaluating the Therapeutic Alliance with a Free-Text CBT Conversational
Agent (Wysa): A Mixed-Methods Study." Frontiers in Digital Health, vol. 4, 2022, p. 847991.

66. Darcy, A., et al. "Evidence of Human-Level Bonds Established with a Digital Conversational
Agent: Cross-Sectional, Retrospective Observational Study." JMIR Formative Research, vol. 5,
2021, p. e27868.
67. Ahmed, A., et al. "Thematic Analysis on User Reviews for Depression and Anxiety Chatbot
Apps: Machine Learning Approach." JMIR Formative Research, vol. 6, 2022, p. e27654.

68. Malik, T., et al. "Evaluating User Feedback for an Artificial Intelligence–Enabled, Cognitive
Behavioral Therapy–Based Mental Health App (Wysa): Qualitative Thematic Analysis." JMIR
Human Factors, vol. 9, 2022, p. e35668.

69. Ta, V., et al. "User Experiences of Social Support from Companion Chatbots in Everyday
Contexts: Thematic Analysis." Journal of Medical Internet Research, vol. 22, 2020, p. e16235.

70. Haque, M. R., and S. Rubya. "An Overview of Chatbot-Based Mobile Mental Health Apps:
Insights from App Description and User Reviews." JMIR mHealth and uHealth, vol. 11, 2023,
p. e44838.

71. Kettle, L., and Y.-C. Lee. "User Experiences of Well-Being Chatbots." Human Factors: The
Journal of the Human Factors and Ergonomics Society, vol. 66, 2024, pp. 1703–1723.

72. Zhang, M., and J. Li. "A Commentary of GPT-3 in MIT Technology Review 2021."
Fundamental Research, vol. 1, 2021, pp. 831–833.

73. Vaswani, A., et al. "Attention Is All You Need." Preprint at arXiv, 2023,
http://arxiv.org/abs/1706.03762.

74. Bubeck, S., et al. "Sparks of Artificial General Intelligence: Early Experiments with GPT-4."
Preprint at arXiv, 2023, http://arxiv.org/abs/2303.12712.

75. Tu, T., et al. "Towards Conversational Diagnostic AI." Preprint at arXiv, 2024,
http://arxiv.org/abs/2401.05654.

76. Salvi, F., Ribeiro, M. H., Gallotti, R., and West, R. "On the Conversational Persuasiveness of
Large Language Models: A Randomized Controlled Trial." Preprint at arXiv, 2024,
2http://arxiv.org/abs/2403.14380.

77. Strachan, J. W. A., et al. "Testing Theory of Mind in Large Language Models and Humans."
Nature Human Behaviour, forthcoming.

78. Yin, Y., Jia, N., and Wakslak, C. J. "AI Can Help People Feel Heard, but an AI Label Diminishes
This Impact." Proceedings of the National Academy of Sciences of the USA, vol. 121, 2024,
e2319112121.

79. Young, J., et al. "The Role of AI in Peer Support for Young People: A Study of Preferences for
Human- and AI-Generated Responses." Proceedings of the CHI Conference on Human Factors
in Computing Systems, ACM, Honolulu, HI, USA, 2024, pp. 1–18.

80. Li, J. Z., Herderich, A., and Goldenberg, A. "Skill but Not Effort Drive GPT Overperformance
over Humans in Cognitive Reframing of Negative Scenarios." Preprint.

81. Malik, A. "OpenAI’s ChatGPT Now Has 100 Million Weekly Active Users." TechCrunch AI, 6
Nov. 2023, https://techcrunch.com/2023/11/06/openais-chatgpt-now-has-100-million-
weekly-active-users/.

82. Salesforce. "Top Generative AI Statistics for 2024." Salesforce News & Insights, 2023,
https://www.salesforce.com/news/stories/generative-ai-statistics/.
83. Pandya. "The Age of Generative AI: Over Half of Americans Have Used Generative AI and
Most Believe It Will Help Them Be More Creative." Adobe Blog, 22 Apr. 2024,
https://blog.adobe.com/en/publish/2024/04/22/age-generative-ai-over-half-americans-
have-used-generative-ai-most-believe-will-help-them-be-more-creative.

84. Li, H., Zhang, R., Lee, Y.-C., Kraut, R. E., and Mohr, D. C. "Systematic Review and Meta-
Analysis of AI-Based Conversational Agents for Promoting Mental Health and Well-Being."
NPJ Digital Medicine, vol. 6, 2023, p. 236.