Name: Abeera Munawar Roll No: 23101340004

Title AI-Assisted CBT for Generalized Anxiety Disorder: A

Randomized Controlled Trial Evaluating Efficacy, Engagement,
and Therapeutic Alliance
Introduction Anxiety disorders are among the most common mental health
challenges globally, with Generalized Anxiety Disorder (GAD)
affecting a significant portion of the young adult population.
Despite the proven efficacy of Cognitive Behavioral Therapy
(CBT), multiple barriers hinder its widespread accessibility. These
include a shortage of trained therapists, financial limitations,
geographical restrictions, and the persistent stigma surrounding
mental health treatment. As a response to these limitations,
artificial intelligence (AI)-driven digital interventions have
emerged, offering new opportunities for scalable, cost-effective
mental health care delivery.

This study explores the use of a pre-developed AI chatbot—Dil Ki

Baat—that delivers CBT-based therapeutic support. Dil Ki Baat is
an AI chatbot designed to provide CBT-based therapeutic support
through a readily accessible platform. Unlike research prototypes
requiring custom training or development, this chatbot comes
equipped with established CBT protocols, reducing the burden on
researchers and participants alike. This eliminates the concern
about students training the chatbot and ensures standardized
therapy delivery across participants.

By conducting a randomized controlled trial (RCT), this study

compares AI-assisted CBT, human-led CBT, and a blended
approach combining both. The study seeks to determine whether a
blended approach is more effective than either AI-assisted or
human-led CBT alone in reducing GAD symptoms, while also
examining user engagement, therapeutic alliance, and satisfaction.
In this blended approach, the AI chatbot will primarily assist with
homework completion and provide emotional regulation support
between human therapy sessions, acting as a "close associate" to
the client. All therapy sessions will be conducted on an individual
basis.
Theoretical Framework The study is grounded in two complementary theories:

Cognitive Behavioral Theory (Beck, 1976): This model explains

anxiety as a result of distorted cognitions and maladaptive
behavioral patterns. CBT targets these distortions through
structured techniques like cognitive restructuring, exposure, and
behavioral activation—all of which are embedded into both
human-led and AI-delivered formats.
 Technology Acceptance Model (TAM) (Davis, 1989): TAM
highlights perceived usefulness (PU) and perceived ease of use
(PEU) as critical to the adoption of technology. This framework
informs the exploration of how users interact with the AI chatbot
and how those perceptions correlate with adherence and symptom
improvement. The TAM will be particularly relevant in
understanding how participants in the blended condition perceive
the AI chatbot as a homework aid and emotional support tool.
Problem Statement To evaluate whether a blended approach combining an AI-assisted
chatbot (used for homework support and emotional regulation) and
human-led CBT is more effective than either AI-assisted CBT or
human-led CBT alone in reducing anxiety symptoms among young
adults diagnosed with Generalized Anxiety Disorder (GAD), when
all therapy is delivered on an individual basis.
Research Question 1. Is a blended approach combining an AI-assisted CBT
chatbot (used for homework support and emotional
regulation) and human-led CBT more effective than either
AI-assisted CBT or human-led CBT alone in reducing
anxiety symptoms in individuals with GAD, when all
therapy is delivered on an individual basis?

2. How do user engagement, satisfaction, and perceived

therapeutic alliance differ between AI-assisted CBT,
human-led CBT, and the blended approach in individual
therapy sessions?

3. What CBT mechanisms (e.g., cognitive restructuring,

emotional regulation) are most responsible for therapeutic
change in each group? Does the AI chatbot's role in
homework support and emotional regulation specifically
enhance these mechanisms in the blended group during
individual therapy?

Research Objective Participants receiving the blended approach will show greater
anxiety reduction than those receiving AI-assisted CBT or human-
led CBT alone, when all therapy is delivered on an individual
basis.

Higher engagement with the AI chatbot will correlate with greater

symptom improvement in the AI-CBT group and potentially in the
blended group, particularly in relation to homework completion
and reported emotional regulation during individual therapy.

The blended approach may show the highest therapeutic alliance

scores, followed by human-led CBT, and then AI-CBT; these
 differences will be examined in relation to symptom reduction and
the perceived helpfulness of the AI chatbot as a homework and
emotional support tool during individual therapy.
Method (Sample, Sample and Sample Size:
Sample size, Data
Collection, Method Sample: The study will involve individuals diagnosed with GAD,
Measures) aged 18-50.
Sample Size: The target sample size is 30-36 participants (10-12 in
each of three groups), including individuals diagnosed with
Generalized Anxiety. The final sample size will depend on
recruitment success and power analysis.
Experimental Design:

Randomized Controlled Trial (RCT) with Three Groups:

Group 1 (AI-CBT): Engages with an AI chatbot delivering CBT

strategies (e.g., cognitive restructuring, exposure therapy) in
individual sessions.
Group 2 (Human-CBT): Receives the same therapy from a
licensed psychologist in individual sessions.
Group 3 (Blended AI and Human CBT): Receives therapy from a
licensed psychologist in individual sessions, with the AI chatbot
used to support homework completion and provide emotional
regulation support between sessions. The specific protocol for how
the AI chatbot is integrated into the homework process and how
emotional regulation support is provided must be clearly defined
(see below).
Therapy Format: All therapy sessions will be conducted on an
individual basis.

Duration: 8 weeks

Session Frequency: 1 session per week (30 minutes each)

Detailed Blended Protocol (Group 3):

Human therapist assigns CBT homework (e.g., thought records,

behavioral experiments) during each weekly individual session.
Participants in the blended group are instructed to use the AI
chatbot Dil Ki Baat to assist them in completing their homework
assignments during the week.
The AI chatbot will provide guidance, prompts, and support for
completing the assigned tasks.
The AI chatbot will also offer emotional regulation techniques
(e.g., mindfulness exercises, breathing techniques) if the
participant reports feeling overwhelmed or anxious while working
 on the homework.
Participants will be encouraged to use the chatbot as needed
throughout the week for homework assistance and emotional
support.
During the following individual human therapy session, the
therapist will review the participant's homework and their
interactions with the AI chatbot, providing further guidance and
support as needed.

Data Collection:

Data Collection Timepoints: Baseline, Week 4, Week 8, and 3-

month follow-up
Primary Outcome Measure:

Generalized Anxiety Disorder-7 (GAD-7) Scale: Pre- and post-

intervention scores to measure symptom reduction.
Secondary Measures:

Patient Health Questionnaire-9 (PHQ-9): To assess comorbid

depression changes.
User Engagement Metrics: AI-CBT chatbot interaction logs
(response time, session completion rate, frequency of use for
homework assistance and emotional regulation).
Therapeutic Alliance: Working Alliance Inventory (WAI) to
compare perceived connection with AI vs. human therapists.
Additional questions will be added to assess the perceived alliance
with the AI chatbot in the blended group, specifically regarding its
role in homework and emotional support.
Heart Rate Variability (HRV): Physiological measure of stress
reduction (optional for biofeedback).

Data Analysis Strategy ANOVA: Compare anxiety reduction across the three groups.
Post-hoc tests will be used to determine which groups differ
significantly from each other.
Chi-Square Test: Compare dropout rates between the three
groups.
Correlation Analysis: Examine the relationship between AI
chatbot engagement (specifically for homework and emotional
regulation) and symptom reduction in the AI-CBT and blended
groups.

Significance for Theory The rising incidence of GAD in young adults, coupled with limited
and Practice mental health resources, necessitates alternative intervention
strategies. This research is timely and significant for several
reasons:
 Clinical Relevance: It evaluates a disorder-specific sample (GAD
only), addressing a gap in AI-therapy literature by providing
comparative data between human-led, AI-assisted, and a blended
CBT approach where the AI specifically supports homework and
emotional regulation, all delivered in individual therapy sessions.

Accessibility and Innovation: The study explores the potential of

combining digital and traditional therapy to enhance mental health
care delivery, which is particularly valuable in underserved regions
with therapist shortages. The specific focus on homework support
and emotional regulation highlights a practical application of AI in
individual therapy.

International and National Contribution: Nationally, the project

supports the integration of digital health tools into clinical
psychology. Internationally, it aligns with global efforts to
democratize access to therapy, providing a framework for
optimizing treatment strategies in various settings.

Therapeutic Engagement and User Experience: By evaluating

engagement metrics and perceived therapeutic alliance, the study
contributes to the design of more effective and acceptable AI
mental health tools and to understanding how these tools can best
be integrated with human therapists, specifically in the context of
homework support and emotional regulation during individual
therapy.
References Bandelow, B., & Michaelis, S. (2015). Epidemiology of anxiety
disorders in the 21st century. Dialogues in Clinical Neuroscience,
17(3), 337-346.

Beck, A. T. (1976). Cognitive therapy and the emotional disorders.

New York: Penguin Books.

Davis, F. D. (1989). Perceived usefulness, perceived ease of use,

and user acceptance of information technology. MIS Quarterly,
13(3), 319-340.

Hofmann, S. G., Asnaani, A., Vonk, I. J. J., Sawyer, A. T., & Fang,
A. (2012). The Efficacy of Cognitive Behavioral Therapy: A Meta-
Analysis. Cognitive Therapy and Research, 36(5), 427–440.
doi:10.1007/s10608-012-9476-