Introduction

more than 90% of the world’s adolescents live in low- and middle-income countries (LMICs) where large treatment gaps for depression and anxiety prevail [].In rural South Africa, mental health services are scarce, leaving young people with limited access to care []. Digital mental health interventions (DMHIs) have emerged as a promising opportunity to bridge this gap, offering anonymity, flexibility, and the potential to reach large numbers of adolescents at low cost [,]. Despite their promise, engagement remains a persistent challenge and an critically important determinant of intervention effectiveness [].Key factors associated with higher engagement include the relatability of situations and characters, appealing aesthetic features, and age-appropriate design [-]. Tho, much of the existing DMHI content has been developed in high-income countries [-], limiting its cultural resonance elsewhere.

To improve cultural fit and engagement, co-design approaches, whereby potential users play an active role in the intervention’s growth, have been increasingly adopted [,]. Such participatory methods help ensure that interventions reflect users’ experiences and preferences, positioning adolescents as “experts of their own experiences” []. However, co-design is resource-intensive, often requiring multiple iterative workshops with creative teams to generate narratives, images, and other materials [,]. This limits the scalability of DMHI development, notably in low-resource settings where capacity and funding are constrained.

Artificial intelligence (AI) is now being integrated into DMHIs in diverse ways, including predicting risk for mental health conditions, informing treatment selection, monitoring progress through wearable data, creating chatbots that deliver therapy, and improving care quality [,]. To date,though,most applications have focused on clinical support or content delivery rather than user cocreation. Recent advances in generative artificial intelligence (GenAI), including large language and diffusion models capable of creating text, images, and music from simple prompts, offer a new avenue for participatory design [,]. By embedding GenAI within participatory design frameworks [,], developers could make the co-design of DMHIs faster, more creative, and more scalable while maintaining the core principle of user-driven innovation. At the same time,integrating AI into global mental health design raises important ethical and cultural considerations. Most GenAI models are trained on data that overrepresent Western, educated, industrialized, rich, and democratic contexts, leading to outputs that may be biased or culturally incongruent []. Engaging adolescents from LMICs in participatory exploration of these tools offers a way to identify such biases early and ensure that future digital interventions better represent diverse cultural perspectives.

This study builds on the Digital Delivery of Behavioural Activation Therapy to Overcome Depression and Facilitate Social and Economic Transitions of Adolescents in South Africa (DoBAt) pilot trial, which evaluated the feasibility, acceptability, and initial efficacy of digitally delivered behavioral activation (BA) therapy among adolescents with depression in rural South Africa []. Adolescents in the intervention arm received BA therapy via the Kuamsha program, comprising a smartphone app (the Kuamsha app) and peer mentor support calls. The Kuamsha app was a coproduced, interactive choose-your-own-adventure narrative game that delivered the main principles of BA through a gamified story format.The app comprised 2 interactive user-led stories, one about a school song contest and one about a football match, each divided into 6 sequential episodes, engaging users through thoughtfully crafted content, images, and music. the kuamsha app and its components, including stories, images, and music, were developed through an iterative process of co-design involving a series of participatory workshops, focus group discussions, and individual interviews [] to ensure the content was appealing and relatable. However, the design process was lengthy and resource-intensive, particularly writing and visualizing the stories. The 2 stories,while engaging,had limitations in scope.Feedback from adolescents further revealed that they had a desire for more content diversity and depth.

Given these constraints and AI’s growing potential to improve participatory design, this

AI-Powered Creativity Workshop with Adolescents in South Africa: A Qualitative Study of Experiences and perspectives

Abstract

This paper details a workshop exploring the use of artificial intelligence (AI) tools – ChatGPT, MidJourney, and Soundful – with South African adolescents to generate creative content. The workshop aimed to understand adolescents’ experiences with these tools and compare AI-generated outputs with existing content from the Kuamsha app, designed to support adolescent mental wellbeing.Through a combination of hands-on activities and a focus group discussion, we gathered qualitative data on perceptions, challenges, and potential applications of AI in creative expression within this context. The study employed inductive thematic analysis of audio recordings, translated from Xitsonga and English, to identify key themes related to adolescent engagement with AI-driven creativity.

Introduction

The increasing accessibility of AI tools presents both opportunities and challenges for adolescent development,particularly in the realm of creative expression. This study investigates adolescents’ experiences using three distinct AI platforms: ChatGPT for story generation, MidJourney for image creation, and Soundful for music composition. These tools were selected to represent a range of AI capabilities relevant to creative pursuits. The workshop format allowed for direct observation of adolescent interaction with the AI, followed by a focus group to explore their perceptions and reflections. The findings contribute to a growing understanding of how AI can be leveraged to support adolescent creativity and mental wellbeing,while also acknowledging potential limitations and ethical considerations. The ability to quickly generate content, and to resolve any data, network, or more complex technical issues.

Methods

Workshop Activities

The workshop activities were structured as follows. First, participants used ChatGPT to create stories. They were asked to include a character name, location, and activity in their prompts (eg, “Please write a story about a young South African boy playing football in the village”). Participants were also shown how to modify elements of the AI-generated stories by giving prompts, such as “Can you make the story more action-packed and engaging?” or “Can you change the name of the main character to a South African name?” or “Can you change X part of the story to Y?” Second, participants used MidJourney to create images. participants were asked to be as detailed as they could with the prompts (eg, “A young African boy wearing a baseball cap in the style of a Marvel comic”). Examples of AI-generated images are mentioned in . Third, participants used Soundful to create music on their website user interface. Participants were asked to choose the music genre, subgenre, key, speed, major or minor chord, and name their song.

Focus Group Discussion

After all groups had generated their own stories, images, and music using the AI tools, participants reconvened for the second part of the workshop: the plenary focus group. The focus groups were facilitated by TN with assistance from SD, PM, and Meriam Meritze.The purpose of the focus group was 2-fold: (1) to explore the adolescents’ experience using the 3 AI tools and (2) to compare the AI-generated stories, drawings, and songs with those created by humans for the Kuamsha app. Comparison content from the Kuamsha app, and also the AI-generated media produced during the workshop, was shown on a projector screen to ensure that all participants could clearly see and hear the material.

A combination of English and Xitsonga was used throughout the workshops to facilitate the understanding and engagement of all participants. SD delivered the presentation in English, with PM providing translation into Xitsonga. Xitsonga was predominantly used during the small group breakaway sessions,and the focus group discussion was conducted in Xitsonga (even though some participants responded in English). The workshops and the focus groups were audio-recorded.

This study was designed and reported in accordance with the Consolidated Criteria for Reporting Qualitative research (COREQ) checklist for interviews and focus group discussions (). We also drew on the Guidance for Reporting Involvement of Patients and the Public to reflect the participatory co-design elements of adolescent involvement.

Analysis

The audio recordings of the workshops and focus group discussions were translated from Xitsonga to English and transcribed by TN and PM. The transcripts were uploaded to NVivo (Lumivero) [], and data were analyzed using inductive thematic analysis.SD reviewed the transcripts several times to familiarize themself with the data before creating the initial codes. SD discussed these initial codes iteratively with HO’M, AvH, and BM and then created final themes after team consensus. BM met with TN, PM, and Meriam Meritze to review the themes. given the exploratory nature and small sample, we did not aim for thematic data saturation but rather sought to capture a breadth of adolescent perspectives.

Reflexivity

The research team consisted of 7 researchers, who identified as female, and 1 researcher who identified as male, from a wide range of career stages. SD is a research assistant, TN, PM, and Meriam Meritze are experienced qualitative fieldworkers, BM (PhD and MD) is a clinician researcher, JRP (PhD

Adolescents’ Experiences Co-Creating Music and Stories with ChatGPT: A Qualitative Study

Abstract: This study explores the experiences of adolescents co-creating music and stories with ChatGPT. Through a workshop setting, participants engaged with the AI to generate creative content based on their personal interests. Findings reveal a largely positive experience, characterized by enjoyment, mood enhancement, a sense of autonomy, and personalization of outputs. While initial prompt engineering required support, participants quickly adapted and experimented with refining prompts. the realism of AI-generated media was noted, alongside awareness of cultural biases within the outputs.

Introduction: This research investigates how adolescents perceive and interact with ChatGPT as a creative tool.The workshop aimed to understand the potential of AI in fostering creativity, self-expression, and critical thinking among young people.

Methods: A qualitative study was conducted with 23 adolescent participants in a workshop setting. data was collected through observations and participant quotes, focusing on their experiences co-creating music and stories with ChatGPT.

Results:

Ease of Use and Initial Impressions: Participants generally found ChatGPT accessible and easy to use,even with limited prior experience. The AI’s ability to understand and respond to prompts was appreciated. An accurate description. Participants were also swift to recognize ChatGPT’s ability to:

Pick up that you forgot to type something in. [Participant #6]

It was difficult in the beginning but the more we did it the more it became easy. [Participant #9]

It was vrey easy. [Participant #20]

It was fun. [Participant #17]

Mood and Emotional Experience: The majority (20/23,87%) described the process as enjoyable and engaging,with most (21/23,91%) reporting that creating music improved their mood.

It’s our first time doing this and we are loving it. [Participant#4[Participant#4]

You’re even dancing to the beats.*[Facilitator[Facilitator]

When I play the music I made myself, it can make me feel good about myself, it can improve my mood or relieve the stress.* [Participant #22]

Autonomy and Agency of Creation: The freedom adolescents had to create was a prominent theme. Adolescents expressed autonomy and ownership of their creations.

I am the producer; everyone likes their own creation. [Participant #21]

We get the freedom to do what we like on the music that we make ourselves. [Participant #1]

Personalization of Outputs: Adolescents’ prompts often reflected their own characteristics, interests, hopes, and aspirations and were supported to refine prompts to make them more culturally relevant or engaging.

I chose that topic because I love playing soccer and I wanted to understand what it takes or to be a good player and what needs to be done to qualify for the tournaments. [Participant #3]

I want to be an IT specialist in future and my topic was based at that and the story that came out inspired me as it started from when the character on the story was still dreaming of becoming an IT specialist right to the end where his dream came true. [Participant #4]

I wanted it to generate Miss Universe pictures because I aspire to be a model one day when I finish school; so I wanted to see if I would be able to get their pictures using that app. [Participant #1]

Using Prompt Engineering to Change Elements of the Story: Most adolescents (18/23, 78%) required support with prompt construction; however, as the workshop progressed, adolescents and facilitators began to experiment with prompt engineering to adapt features of the story.

Let’s prompt it to change the name of the character to Admire. [Participant#7[Participant#7]

We started from bullying and we moved to supportive, let’s try to make it more exciting and more positive.*[fieldworker#3[fieldworker#3]

These images look real and I think I have seen one on TV before.*[Participant#6[Participant#6]

When you give it the prompts of what you are looking for; it was able to generate those images. For example, I was looking for images of football players and it generated exactly that.* [Participant #5]

*I wanted it to generate pictures of houses in the villages but the ones I have seen were entirely different fr

Adolescents’ Perceptions of Generative AI and its Potential for Mental Health Interventions: A Qualitative Study

Abstract

This study explores adolescents’ perceptions of generative artificial intelligence (GenAI) and its potential use in co-designing and adapting digital mental health interventions (DMHIs) in a low-resource rural South African context. Through participatory workshops and focus group discussions, we investigated adolescents’ experiences with ChatGPT, MidJourney, and Soundful – tools for generating stories, images, and songs. Findings reveal high engagement and enjoyment with GenAI, aligning with principles of behavioral activation (BA) by supporting meaningful and rewarding activities. However, challenges related to prompt engineering and the need for facilitator support were identified. Ethical considerations surrounding AI-generated content, particularly regarding artistic ownership, also emerged.

Introduction

Digital Mental Health Interventions (dmhis) offer promising avenues for addressing mental health challenges, particularly among adolescents. however, adapting these interventions to diverse cultural contexts and individual needs remains a significant hurdle. Generative AI (GenAI) presents a potential solution, enabling co-design and personalization of DMHIs. This study investigates adolescents’ initial experiences with and perceptions of GenAI tools, exploring their feasibility and potential within a low-resource setting.

Methods

A qualitative study was conducted with adolescents in rural South Africa. participatory workshops were held to introduce and explore the use of three GenAI tools: ChatGPT (text generation), midjourney (image generation), and Soundful (music generation). Participants engaged in hands-on activities, generating stories, images, and songs based on their interests. Focus group discussions followed, delving deeper into their experiences and comparing AI-generated outputs with those from Kuamsha, a human-created interactive narrative game delivering behavioral activation (BA). Thematic analysis was used to analyze the qualitative data collected.

Results

Adolescents’ Initial Experiences with GenAI

The majority of participants (all but one) had no prior experience with GenAI before the workshops. adolescents found the process of using GenAI to create content – stories,images,and music – to be active,engaging,and enjoyable. They particularly appreciated the ability to personalize outputs through prompts,aligning with their interests and aspirations.

I can use [ChatGPT] and get good scores and it saves time as it does things quickly with less errors. [Participant #17]

I would use [MidJourney] to search for pictures and I can’t print them out, I would just try to draw them myself by looking at the pictures generated using the MidJourney app. [Participant #5]

Ethical Considerations

Adolescents demonstrated nuanced ethical perspectives regarding GenAI. They generally considered it acceptable to use AI tools for content creation when they were actively involved in selecting parameters and genres. However, they expressed strong moral objections to AI being used to replicate existing artists’ voices or styles.

…they were simply selecting the genres and settings… [Participant #17]

Adolescents’ Use of GenAI in the Future

Most adolescents expressed an interest in using GenAI tools for both schoolwork and personal use.

It doesn’t help getting good grades for something that I did not do. Let’s say they ask me to tell them what I have written; I wouldn’t be able to tell them because I am not the one who wrote it. [Participant #6]

I would use it to make stories that I can read for fun; I wouldn’t use it for my schoolwork. [Participant #8]

Discussion

Principal Findings

This study reports a thematic analysis of an early-stage qualitative investigation to explore the feasibility and potential of using GenAI in the co-design and adaptation of a DMHI with adolescents in a low-resource context in rural South Africa. Participatory workshops were held to explore adolescents’ experiences using 3 GenAI tools (ChatGPT, MidJourney, and Soundful) to generate stories, images, and songs. Focus group discussions further explored adolescents’ experiences and compared the AI-generated outputs with those from the Kuamsha app, an interactive narrative game delivering BA created by humans without AI.

For all but 1 participant, the workshops were the first time they had encountered GenAI and been given the opportunity to explore how it works. adolescents found using GenAI to create stories and images with prompts and create music with an AI-based user interface to be an active, very engaging, and enjoyable process. These findings support the principles of BA, which emphasize the importance of supporting individuals with depression to undertake activities that are meaningful and rewarding. Adolescents liked aspects of the Kuamsha app that were self-relevant and, therefore, were

Limitations of AI-Generated Content for Adolescent Mental Health

This study explored how adolescents engage with AI-generated content, but several limitations deserve attention. Understanding these helps refine future research and development in this area.

Participant Bias

We intentionally used the same adolescents who previously used the Kuamsha app to both create and evaluate AI content. This allowed us to capture their direct reflections on the creative process.However, it also introduced potential bias. Participants may have favored their own AI creations due to a sense of ownership and self-expression, not necessarily because the content was better. Their prior experience with Kuamsha – and whether it improved their mood – could have also influenced their judgments. It’s unclear if their preference for AI content stemmed from genuine quality or simply stronger personal investment. Future research should involve independent groups of adolescents, or compare results across multiple samples to reduce this bias and get a more accurate assessment.

Lack of therapeutic Integration

This study focused on adolescents’ experience with GenAI, not on testing its use within a therapeutic co-design process. Unlike the Kuamsha app, the AI-generated stories, images, and music weren’t intentionally designed with behavioral activation (BA) principles or other proven psychological strategies. A guided co-design approach – where prompts specifically include therapeutic elements like goal setting or positive reinforcement – could create content more directly aligned with intervention goals. Though, our priority was to see if genai could meaningfully engage adolescents, which is a crucial first step. Thus, the absence of explicit BA elements isn’t a major limitation, as it doesn’t hinder our understanding of how adolescents emotionally and creatively interact with AI. Future work should explore systematically integrating psychological support into AI-assisted co-design to maximize both engagement and clinical impact.

Technology Constraints

We couldn’t use text-to-audio technology for song generation due to its limited availability during the study. Fortunately, multimodal AI models are rapidly improving, now offering elegant text-to-audio capabilities. Future research will benefit from these advancements.

Conclusions

Designing digital mental health interventions (DMHIs) is becoming more “humanized,” shifting from approaches led by experts to participatory co-design.