Not provided
Not provided
Not provided
Not provided
Not provided
Not provided
Not provided
Not provided
Not provided
Not provided
Not provided
Not provided
Not provided
Not provided
Not provided
Not provided
Not provided
Not provided
Not provided
The goal of this randomised controlled trial is to evaluate the effectiveness of an AI mental health chatbot in promoting emotional wellbeing and perceived empathy among university students in Singapore with mild or subclinical symptoms of anxiety and depression. The main questions it aims to answer are:
Can the chatbot provide emotional validation and be perceived as empathic?
Does the chatbot reduce symptoms of anxiety and depression and improve wellbeing more than an inactive control condition?
Researchers will compare students who engage in four sessions with the chatbot to students who complete four neutral writing tasks to assess differences in emotional wellbeing, empathy, and resilience.
Participants will:
Complete baseline wellbeing assessments
Be randomised to:
Four 20-minute chatbot sessions providing personalised support using cognitive-behavioural and compassion-focused techniques (intervention group), or
Four 20-minute neutral writing sessions unrelated to mental health (control group)
Complete post-session and follow-up wellbeing questionnaires
All sessions are conducted virtually over Zoom. Participants are full-time students at the National University of Singapore, aged 21 and above. The study aims to inform future development of AI tools for emotional support in non-clinical settings.
This randomised controlled trial evaluates the effectiveness of an AI-based mental health chatbot in promoting emotional wellbeing among university students in Singapore. The chatbot, part of the existing Intellect app used by the National University of Singapore (NUS), delivers supportive conversations grounded in cognitive-behavioural and compassion-focused techniques.
A total of 172 NUS students (aged 21+) will be screened for eligibility. Those with moderate to severe anxiety or depression (GAD-7 or PHQ-9 ≥10), current mental health treatment, or history of self-harm will be excluded and provided with mental health resources. Eligible participants will be randomly assigned to either:
Intervention group: Four 20-minute chatbot sessions over four weeks, delivered via Zoom. The chatbot explores participants' challenges, promotes emotional insight, identifies unhelpful thinking patterns and behaviours, and offers weekly reflection homework.
Control group: Four 20-minute neutral writing tasks (non-mental-health topics) over four weeks, matched in format and timing.
All participants complete self-report wellbeing questionnaires before and after each session, and a one-month follow-up. Primary outcomes include changes in anxiety (GAD-7), depression (PHQ-9), wellbeing (WHO-5), affect (PANAS), resilience (BRS), and perceived empathy (PEQ). The chatbot's perceived empathy will be assessed after session 1 and session 4 in the intervention group.
Sessions are not audio- or video-recorded, and participant anonymity is preserved. Researchers monitor sessions via video to ensure safety. At follow-up, control participants will be offered access to the chatbot if it proves effective.
This study aims to assess feasibility, empathy, and clinical impact of chatbot-delivered support, contributing to the development of scalable digital tools for emotional wellbeing in non-clinical university settings.
Not provided
Not provided
Not provided
Not provided
| Label | Type | Description | Intervention Names |
|---|---|---|---|
| AI-delivered emotional wellbeing support | Experimental | Participants in this arm will complete four sessions with an AI-powered mental health chatbot over four weeks. Each session will last approximately 20 minutes and take place via Zoom. The chatbot provides structured emotional support informed by CBT principles, including identifying unhelpful thinking patterns and coping behaviours, delivering thought restructuring and compassion-focused techniques, and assigning weekly reflection-based homework. The intervention is designed to promote emotional wellbeing and alleviate mild anxiety and depressive symptoms. |
|
| Inactive control (neutral writing tasks) | Placebo Comparator | Participants in this arm will complete four 20-minute sessions over four weeks involving neutral writing tasks unrelated to mental health. Writing prompts cover non-emotive topics (e.g., daily routines, neutral observations) and do not include emotional support, feedback, or guidance. This arm serves as an inactive control condition to isolate the effect of the chatbot intervention on emotional wellbeing and perceived empathy. |
|
| Name | Type | Description | Arm Group Labels | Other Names |
|---|---|---|---|---|
| AI-delivered emotional wellbeing support | Behavioral | Participants in this arm will complete four 20-minute sessions with an AI-powered chatbot designed to deliver CBT-based emotional wellbeing support. The chatbot provides conversational guidance using techniques such as cognitive restructuring, behavioural activation, and emotion validation. Sessions are delivered via Zoom, and participants receive structured homework after each session to reinforce learning. The chatbot was not custom-built for this study but is an existing tool provided to NUS students as part of their digital mental health benefits. |
| Measure | Description | Time Frame |
|---|---|---|
| Perceived empathy (Robot's Perceived Empathy Scale - RoPE) | Assesses participants' perception of the AI chatbot's empathy (cognitive and affective components), using the validated Robot's Perceived Empathy (RoPE) scale. The RoPE consists of 20 items rated on a 5-point Likert scale (1 = Strongly Disagree to 5 = Strongly Agree). Score range: 20-100. Higher scores indicate: Greater perceived empathy. | Immediately after Session 1 and Session 4 (baseline and post-intervention for chatbot group only). |
| Change in emotional wellbeing (WHO-5 Wellbeing Index) | Assesses overall emotional wellbeing and life satisfaction using the WHO-5 Wellbeing Index, a 5-item self-report scale. Each item is rated on a 6-point Likert scale (0 = At no time to 5 = All of the time). Score range: 0-25. Higher scores indicate: Greater wellbeing. | Day 1 (Baseline), Week 4 (End of Session 4), and Week 8 (1-month follow-up after last session). |
| Change in anxiety symptoms (Generalised Anxiety Disorder 7-item - GAD-7) | Assesses anxiety symptoms using the GAD-7, a 7-item self-report scale. Each item is scored from 0 ("Not at all") to 3 ("Nearly every day"). Score range: 0-21. Higher scores indicate: Greater severity of anxiety symptoms. | Day 1 (Baseline), Week 4 (End of Session 4), and Week 8 (1-month follow-up after last session). |
| Change in depressive symptoms (Patient Health Questionnaire-9 - PHQ-9) | Assesses depressive symptoms using the PHQ-9, a 9-item self-report scale. Each item is scored from 0 ("Not at all") to 3 ("Nearly every day"). Score range: 0-27. Higher scores indicate: Greater severity of depressive symptoms. | Day 1 (Baseline), Week 4 (End of Session 4), and Week 8 (1-month follow-up after last session). |
| Measure | Description | Time Frame |
|---|---|---|
| Change in resilience (Brief Resilience Scale - BRS) | Assesses participants' ability to bounce back or recover from stress. The BRS includes 6 items scored on a 5-point Likert scale (1 = "Strongly disagree" to 5 = "Strongly agree"). Score range: Average score between 1 and 5. Higher scores indicate: Greater resilience. | Day 1 (Baseline) , Week 2 (Post-Session 2), Week 4 (Post-Session 4) , Week 8 (1-month follow-up after last session) |
Not provided
Inclusion Criteria:
Current undergraduate student at the National University of Singapore (NUS)
Aged 21 years and above
Proficient in English
Not currently undergoing psychological treatment or taking psychiatric medication
No history of self-harm or suicide
Scores below 10 on both the GAD-7 and PHQ-9 (i.e., subclinical levels of anxiety and depression)
Exclusion Criteria:
Currently undergoing psychological treatment or counselling
Currently taking psychiatric medication
History of self-harm or suicidal ideation
Scores ≥10 on the GAD-7 and/or PHQ-9, indicating moderate to severe anxiety or depression
Inability to provide informed consent or comply with study procedures
Not provided
Not provided
Not provided
Not provided
Not provided
| Name | Role | Phone | Extension | |
|---|---|---|---|---|
| Pak (Martin) Hei Ng, Trainee Clinical Psychologist | Contact | +65 9616 0572 | e1350419@u.nus.edu |
Not provided
Not provided
This study involves personal mental health disclosures, even if anonymised. Sharing individual participant data (IPD), even coded, increases the risk of re-identification and may breach the trust established during consent.
Not provided
Not provided
Not provided
Not provided
Not provided
This study uses a parallel assignment model in which participants are randomly allocated to one of two groups: (1) an experimental group receiving a four-session intervention with a mental health chatbot designed to deliver structured conversational support using CBT and emotional validation techniques, or (2) an inactive control group completing neutral writing tasks unrelated to mental health. Both groups complete sessions over four weeks, with follow-up assessment one month post-intervention to evaluate sustained effects.
Not provided
Not provided
Participants are partially masked to the study's purpose. They are not informed of the specific aim to evaluate the chatbot's impact on mental health outcomes, nor are they aware of the control versus intervention group distinctions. This partial masking helps reduce demand characteristics and expectancy effects. No other parties are blinded.
Not provided
|
| Neutral writing tasks | Other | Participants in this arm will complete four 20-minute sessions of neutral writing tasks unrelated to mental health. Prompts include writing factually about daily routines or describing an object in detail. The content is emotionally neutral to act as a comparison for the chatbot intervention. Sessions are conducted online via Zoom. |
|