Generative AI and Creative Mediums for Youth's Emotion Regulation: An Interview Study with Clinicians

4.1.1 How Clinicians Bring Visuals and Narratives into Their Practice. Clinicians described using visuals and narratives to help youth build emotional awareness, express feelings, and practice coping skills. For instance, clinicians use feelings thermometers to support youth's emotional awareness and expression. P9 (psychologist, PsyD) explained that visual tools and scenarios can make abstract emotions more tangible and open up discussion: "I often rely on visual tools, like emotion wheels, or specific scenarios such as writing out a scenario from the week when [a] child had a hard time. We think about it together visually for the kids." Further, clinicians used stories and visuals to scaffold youth's understanding and expression of abstract emotions by representing these emotions in concrete ways. For example, P10 (art therapist) shared that she used the book Ruby's Worry [107] to teach ER skills: “I sometimes bring in a book like Ruby Finds Worries, and use it as an example. If I start with that visual image, I've planted the seed of ‘what does your worry look like if you were going to create it?’”

Clinicians also used digital images, drawing tools, and image-based worksheets, to teach and practice coping skills with youth. For example, P7 (art therapist, PhD) displayed forest-scene images to guide breathing exercises and mindfulness through visual imagery. Clinicians additionally used digital art tools to address youth's individual challenges. P18 (art therapist, PhD) shared that she uses the Jackson Pollock online tool³, which provides a blank canvas with uncontrollable color and brush behavior, to help youth with OCD or anxiety release perfectionistic tendencies: "If I want someone to practice letting go of perfectionism, I use the Jackson Pollock tool. Some kids, especially those with OCD or anxiety, fear making mistakes. These unpredictable tools help them face that discomfort." Beyond clinical settings, clinicians also recommended that parents use multi-sensory approaches at home to reinforce ER coping skills. For instance, several clinicians encouraged families to practice self-soothing techniques through the five senses, including vision, hearing, and touch. P9 (psychologist, PsyD) explained that these multi-sensory strategies can be effective for youth who have stronger visual than auditory skills, especially when emotions are heightened: "A lot of the kids I see have stronger visual than auditory skills,, relying more on visuals when kids are upset." These examples illustrate that clinicians leverage a range of visual and narrative tools to support youth's ER in concrete and accessible ways.

4.1.2 The Gaps in Practice: When Visuals and Narratives Fall Short. While clinicians widely used creative mediums, they noted two key challenges: (1) youth's varying abilities to express emotions through visuals and narratives, and (2) the limited interactivity and customization of traditional worksheets.

First, not all children and adolescents respond equally well to the same type of visual guidance or expression, since each child possesses different levels of cognitive processing and emotional understanding. For example, youth with ADHD or autism often require more concrete and engaging support than neurotypical youth. P17 (student therapist) explained: “[Neurodivergent kids] need a lot of concrete thinking, so clinicians have to be more creative in how they convey these ideas without relying on abstract comparisons.” She added that analogies, such as describing anxiety as a smoke alarm, often fail to resonate with neurodivergent children. This aligned with P1 (art therapist)’s comment that, although she tailors visual and narrative metaphors to each child's symptoms, doing so requires substantial effort: "I design my stuff and adapt it for each kid… I tend to have a lot of kids with ADHD or ADD, and they get distracted easily."

Additionally, clinicians noted that traditional worksheets lack diversity and interactivity, making it difficult to sustain engagement. P14 (psychotherapist) shared: “I think a traditional worksheet is hard for them to use because they see it so much in hospitals, doctors’ offices, therapists. They get bored or stop taking it seriously.” The repetitive use of the same worksheet over time could reduce its effectiveness. This effect might be especially strong for children with autism, who need continuously updated reinforcers. As P16 (psychologist, PhD) explained: "Children with autism have restricted interests, but those interests and what we find reinforcing can change. Someone might respond well to a sticker chart, but later it's no longer reinforcing. You need to find a new reinforcer." For self-practice at home, clinicians shared that worksheets should include more interactive features, like feedback loops for families. P20 (psychiatrist, MD) noted: "I think parents would feel more confident about some of the resources and handouts if troubleshooting was built into it." These findings illustrate that adding interactivity and personalization to worksheets holds promise for making them more engaging and effective.

AI-generated images in a pastel drawing style. (A) A gloomy-looking fox standing under an umbrella in the rain. (B) A cloud with a rainbow, a positive image.

4.2.1 Accessible Emotional Awareness and Expression. Clinicians envisioned that GenAI-generated visuals and narratives could make abstract feelings more vivid and relatable, helping children identify, express, and communicate emotions with greater autonomy and reduced psychological barriers. After using our technological probe, they emphasized that AI's creations could support youth in identifying and expressing emotions through assisted visualization. For example, P8 (student therapist) noted that GenAI could help children “see their emotions almost come to life and visualize them a bit more.” Similarly, P11 (art therapist) highlighted AI's potential to support youth in externalizing emotions through visual and narrative metaphors: "There's something powerful about a client saying, ’I'm feeling this way: I picture it as a desert with a hyena, a butterfly, and a thunderstorm.’ AI could generate that image for therapeutic use. It's different from drawing by hand, but still meaningful."

Clinicians highlighted AI's ability to generate varied forms, such as colors and weather elements, to represent different types and intensities of emotions based on youth's descriptions. They noted that it could be especially helpful for children with limited skills in identifying and expressing emotions. As P16 (psychologist, PhD) shared: “It gives children different ways to express themselves. It presents visuals in a variety of forms, which I think supports things like emotion identification.” AI's diverse content creation could represent different intensities of the same emotion. For example, rather than showing sadness as a single “rainy day,” AI could create “heavy rain,” “light rain,” or “a rainbow after the rain,” linked to children's experiences. After seeing such variations for the prompt "I'm sad, I feel like a rainy day", P19 (psychologist, PhD) shared that AI's diverse creation could help youth express a range of emotional intensities and qualities: “I think the weather one is great for helping them understand how an emotion might show up differently. When they feel angry, it doesn't always look the same way, they feel it at different intensities and with different qualities.”

Additionally, identifying and expressing emotions through GenAI could serve as a self-directed entry point for youth who struggle with emotional expression through traditional methods, such as paper and crayon. This could be due to difficulties in receiving scaffolding that supports their agency while teaching diverse ways of visualizing emotions. Clinicians envisioned that GenAI could reduce the psychological pressure of expression while providing more autonomy. For example, P12 (art therapist) sometimes gave directive starting points to children overwhelmed by a blank page, while being careful not to provide “too much directive input.” She envisioned that AI-generated visuals and narratives, grounded in children's emotional reflections, could serve as a more "autonomous starting point on their own". This autonomy could be particularly important for youth whose feelings had been suppressed by authority figures, such as parents who avoided emotional discussions. Clinicians noted that GenAI could provide emotional freedom to better identify and express authentic feelings. After using the technological probe, P11 (art therapist) appreciated the feature allowing youth to edit AI-generated illustrated stories which were based on acceptance, cognitive reappraisal, or fact-based approaches:“I love being able to restate what AI brings up. For people who are more depressive or repeatedly criticized, this rewrite narratives and drawing can help them better identify their authentic feelings.” They also emphasized that GenAI-assisted creative expression could mediate emotional communication, especially for adolescents who tend to internalize symptoms. For example, P17 (student therapist) shared: "For adolescents, especially those with more internalizing symptoms, like feeling really sad or afraid without the urge to throw things, communicating through this [the technological probe] could be helpful." These examples highlight how GenAI's visuals and narratives could lower psychological barriers and facilitate emotional communication.

4.2.2 Practicing Coping Skills in Concrete and Vivid Ways. Beyond emotional identification and expression, clinicians envisioned GenAI supporting the practice of coping strategies, such as cognitive reappraisal, a core component of CBT that involves reframing negative thoughts into more positive ones. They emphasized that visuals can help make this abstract skill more concrete and understandable for youth. For instance, P7 (art therapist, PhD) proposed:

"Imagine turning gray thoughts into sunny ones. The AI can ask, ’What makes you happy?’ It's good to acknowledge the negative emotion, but you also need to bring in positive energy with an image. If the kid says ice cream, it could generate that image."

Figure 2 illustrates this positive imagery. For example, if youth drew a fox on a rainy day representing depression (Fig. 2A), GenAI could transform it by adding a rainbow (Fig. 2B). Clinicians also described how GenAI could support clinical interventions through personalized visualizations, particularly for trauma-related symptoms. For instance, P3 (psychotherapist) suggested that for youth with obsessive-compulsive disorder (OCD), AI-generated imagery could scaffold exposure therapy: “For OCD, we use a fear hierarchy and exposure therapy. If a child fears spiders, AI-generated images could help them practice emotion regulation before progressing to real-life exposure..” Some clinicians shared that they had seen colleagues use GenAI to support the rehearsal of coping skills through imaginative storytelling and believed it could be effective. For instance, P19 (psychologist, PhD) shared:

"I've seen people use GenAI to create imaginative stories that let clients rehearse coping skills in advance. The idea is to imagine a situation and practice responding before it happens. That could be really effective."

Clinicians also valued the probe's feature that allowed youth to select from AI's creations and edit them, noting this process could help youth discover coping strategies that best fit their needs. P15 (psychiatrist, MD) highlighted that providing diverse options through AI-generated stories and visuals made coping practice more personally relevant: "I think [AI generated] stories are a really engaging way to teach coping skills. If an adolescent tells you they're feeling bored, you can say the options are either you accept it or you try to change it. Illustrating these two options (ACT / CBT) as a story can help teenagers better visualize the outcome of each route." These reflections highlight GenAI's potential to scaffold youth in finding best-fit coping strategies and to support their practice through concrete visualization.

4.2.3 Creating Personalized, Engaging Experiences. After using our probe, which generates personalized stories and drawings based on youth's conversations with the AI, clinicians shared that youth could find greater relatability and emotional connection in the AI-generated content than in static materials such as paper-based CBT handouts and existing storybooks. They noted that because the stories are rooted in the child's own emotions, the content could feel personal and relevant. As P5 (psychotherapist) shared: “I think this story-writing feels like a creative approach to introduce these skill sets, instead of just a handout. It feels personal, because each of these references ties back to why I was feeling the way I was feeling.” Clinicians saw potential for the content to serve as a shared reference point, supporting collaborative reflection and coping skill practice between youth and adults. For example, P13 (psychologist, PhD) described how an AI-generated story could facilitate parent–child dialogue while viewing an image (Figure 2):

“If the AI creates a story or image of a fox, parents can ask, ‘How does Mr. Fox feel today?’ or ‘What triggers his sadness?’ It becomes an interactive and imaginative way to teach CBT concepts. As they write the story together, they can brainstorm coping skills for Mr. Fox, making the practice more specific and engaging.”

In addition, they valued the flexibility of GenAI to adapt its content based on youth's symptoms, age, and interests. Since effective strategies and triggers vary across individuals and even across different moments, clinicians emphasized the importance of ongoing personalization. For instance, P16 (psychologist, PhD) shared: "For kids whose interests are constantly changing, they might need new ideas to stay engaged. In that sense, GenAI could be helpful, because it can create new content based on what the kid might need."Clinicians also highlighted that GenAI's adaptive visual and narrative support could be particularly helpful for neurodivergent youth, including those with ADHD or autism. P16 (psychologist, PhD) explained: “I recommend different visuals when I'm working with kids with ADHD or autism, because a lot of times they do better with a visual than with words. For neurodiverse and autistic youth, this tool gives options for expressing feelings beyond words.” P20 (psychiatrist, MD) shared: “For youth with mild autism, this tool could be very helpful because they may not naturally communicate visually or verbally. Having different ways to express what they feel and think gives them a more supportive toolkit.” Together, clinicians highlighted that GenAI's adaptable visual outputs may offer youth with diverse cognitive needs more accessible and personalized channels for emotional expression.

4.3.1 Unpredictable Outputs and Risks of Maladaptive Distress. Clinicians cautioned that the vividness of AI-generated visuals and narratives, combined with their unpredictability, could trigger trauma or reinforce maladaptive thinking in some youth. They noted that GenAI's unpredictability could produce distressing content that is hard to manage. For instance, P3 (psychotherapist) shared: "We cannot control [GenAI's] output, and it might not know all the ins and outs of patient care. So we don't know what content the child will access." They further cautioned that AI's visual and narrative creations could exacerbate youth's maladaptive thinking. As P4 (psychotherapist) noted:

"If AI's image or story creation unintentionally validates cognitive distortions, especially with paranoia, personality disorders, or long-untreated trauma, that's particularly hard."

P1 (art therapist) also warned that GenAI could intensify paranoid reactions: "I've had clients with severe paranoia, and [ER with GenAI] might not help them. They might say things like, ’This AI is going to destroy my life.’ Clients with highly distorted thoughts might not be good candidates for this."

To mitigate these risks, clinicians stressed the need for contextual and symptom-specific safeguards and active human involvement. P6 (psychologist, PhD) emphasized the risk of repeatedly prompting youth to explore emotions without adequate clinician support, noting that “Youth with depression may be at risk without appropriate support from a clinical professional. Some of the emotions they're working on could potentially trigger suicidal thoughts or even homicidal behaviors.” This highlights that using GenAI with youth experiencing depression or suicidal ideation could be unsafe. They emphasized that therapist supervision enables safe use of AI-generated art and stories. As P7 (art therapist, PhD) explained: "If you have a skilled therapist, they can guide and help reinforce the image in different ways, not just whatever this AI generated. Children need a skilled therapist that's going to take it a little further." These reflections highlight that GenAI's vivid but unpredictable outputs, when used without contextual awareness or human supervision, risk intensifying maladaptive distress.

4.3.2 Misinterpretation Between AI and Youth's Creations. Clinicians observed that the visuals and narratives generated by our probe were often metaphor-based, requiring interpretation by both the child and the AI. They noted that this openness to interpretation could introduce risks if either party misinterprets the other's intent. For example, youth may misunderstand GenAI's image- and narrative-based suggestions if they lack foundational knowledge about emotional concepts, which could lead to inappropriate responses. As P5 (psychotherapist) explained:

"Make sure that the [visual and narrative] outputs feel useful and safe, and that a patient doesn't misinterpret it. If they don't have background knowledge about ‘riding the wave’, which means sitting with an emotion and letting it pass, they might think it means acting on anger or impulsivity instead."

This observation highlights how metaphorical representations, such as describing an emotion as a “wave”, pose a risk of misinterpretation and may hinder the delivery of adaptive psychoeducation.

Conversely, GenAI's own assumptions about metaphor interpretation can also be problematic, since symbols carry different meanings for different people. P7 (art therapist, PhD) described: “The assumption that a partly cloudy day isn't positive is not correct for me. Symbols are subjective, not objective. Maybe in the AI world, a cloudy day means sad or unhappy, but for me, partly cloudy feels relaxing. Fully sunny makes me feel overwhelmed.” These insights highlight the need to balance the expressive flexibility of metaphor with clinically appropriate interpretation, ensuring that meaning is both safe and contextually accurate.

4.3.3 Failure to Detect Red Flag Symptoms. Clinicians in our study expressed concern that GenAI, particularly when generating or interpreting content through visual and narrative mediums, may fail to detect red-flag symptoms such as suicidal thoughts or self-harm intentions. For instance, P4 (psychotherapist) expressed skepticism about GenAI's reliability and emphasized that training it is like training human experts, which requires extensive data and carries inherent risks: "GenAI is trained on a set of data, influenced by what it learns, just like training new therapists. We accept some risk with trainees. How much training would an AI need before it's considered low enough risk to act without supervision?" P20 (psychiatrist, MD) similarly questioned AI's capacity to recognize distress signals embedded in youth's creations: "It's hard to know when the AI should continue or redirect. Knowing when to step in or guide is a human skill, and I don't know if AI has that. If using AI is not monitoring youth's creation like suicidal thoughts, that's risky." Clinicians were particularly skeptical of GenAI's ability to interpret metaphorical or ambiguous expressions of distress. P10 (art therapist) noted that even direct language, such as “I want to die,” may not always reflect suicidal intent and must be understood contextually: "The biggest issue that comes to mind is self-harm and suicidal ideation, and how a tool like that could handle triggers or warning signs. For example, if children or teens write, ‘Sometimes I just want to die,’ they don't always mean they want to kill themselves." Interpreting such intentions could become more complex when youth express emotions indirectly through metaphor-based visual or narrative modalities.

4.3.4 Overreliance Without Critical Thinking. Clinicians expressed concern that GenAI's personalized and engaging features may lead youth to become attached to the tool, which could ironically contribute to emotional dysregulation and impulsive behaviors. For example, clinicians noted that GenAI's flexibility and unpredictability in content generation could be risky if youth use it to reinforce maladaptive thinking. As P8 (student therapist) shared: "If there's no barrier to what they can retype or edit, I'd worry that someone could change the AI's creation so much that it ends up reflecting more negative moods." They emphasized that this risk may be greater for youth who lack the critical thinking skills to evaluate GenAI's suggestions. For these users, independent interaction could be unsafe. As P5 (psychotherapist) shared: “I'm hesitant to consider GenAI for someone vulnerable to suggestion without the critical thinking skills to evaluate it.”

GenAI's interactive features, such as conversational natural language processing and use on digital devices, raised concerns that some youth could become overly engaged or even develop addictive patterns of use. For example, P15 (psychiatrist, MD) shared concerns about youth's overattachment through ongoing conversation with Gen AI: "I've heard the horror stories like those children and teenagers who like to talk with AI robots a lot, and then they get attached to them in an unhealthy way." To address these concerns, clinicians recommended integrating control features to prevent overuse or dysregulation, such as limiting screen time. As P10 (art therapist) suggested: "Set limits, such as timers that shut AI tool down." These concerns highlight that GenAI's engaging and responsive visual and narrative content could also lead to unhealthy attachment and overreliance if not carefully managed.

4.4.1 Supporting Concrete and Personalized ER. Clinicians recommended that GenAI support personalized ER by creating diverse and descriptive visuals and narratives to foster youth's autonomous emotional expression. They also noted that these features could be especially useful in group sessions where individualized support is limited. They recommended that GenAI provide interactive and descriptive visual and narrative metaphors, helping youth co-create characters and emotional story contexts. P6 (psychologist, PhD) noted: “Having an interactive piece would be really helpful to grab kids’ attention. GenAI could ask about color, shape, or how they want to represent their emotion.” P17 (student therapist) described that descriptive content can make ER safer and more effective: "Some kids really connect with metaphors… a descriptive, realistic narrative is generally safer." Clinicians also emphasized the importance of giving youth visual choices so they can autonomously decide which coping strategies to practice. P9 (psychologist, PsyD) stressed: “GenAI should give some examples of calm thoughts visually. Giving youth the autonomy to select the calm thought or the self-affirming statement they want to use would be helpful.” Clinicians envisioned these concrete and personalized features could support group ER sessions by offering alternative ways for youth to express emotions when individual attention is limited. As P16 (psychologist, PhD) explained: “In group therapy, you usually have to go around and share, which can be hard for some kids. [Expressing with GenAI] can be another way to show what you're feeling.” Similarly, P18 (art therapist, PhD) noted its value in school-based ER sessions, where students could create their own stories and share them with the group.

4.4.2 Adapting to Individual Youth and Involving Adults. Clinicians emphasized the importance of tailoring GenAI-assisted ER to youth's individual needs while balancing autonomy and adult involvement to ensure safety. First, they noted the need for adaptability to youth's symptoms, such as using more interactive and concrete metaphors for neurodivergent youth. As P18 (art therapist, PhD) explained: “Children with autism are not going to understand symbolism or metaphors that AI creates. It's very hard because they're concrete and literal. For kids on the developmental spectrum, something more descriptive and detailed might work better.” Similarly, P14 (psychotherapist) shared: “When I work with neurodiverse youth, it's less talk therapy and more like a game I created. [GenAI] could play that role for them.”

Clinicians also emphasized the importance of involving adults to ensure the safe use of GenAI, explaining that adults can help reframe and interpret unexpected content. For instance, P3 (psychotherapist) suggested:

"If [ER with GenAI] can be done with an adult, that will be helpful. If AI says [risky things] that somehow trigger the child, the adult can work in real time to reframe it."

Adult involvement should vary by age; younger children benefit from more support, while teens value independence. P6 (psychologist, PhD) shared: “Parents should be more supportive and work alongside younger children, but it may be easier for teenagers to work independently.” P15 (psychiatrist, MD) added: “Teenagers may be reluctant to use [GenAI] if it involves parental oversight.” Accordingly, clinicians emphasized the need to balance autonomy and parental involvement. Some features can promote independence, while others should actively involve adults to ensure emotional safety and support. As P3 (psychotherapist) suggested: “Teens should answer questions independently with AI, but an adult can join during generation to process responses together.” These insights underscore the need for both personalized content and flexible adult engagement to support safe and effective GenAI-assisted ER practices.

4.4.3 Continuity of ER Between Home and Clinic. Clinicians envisioned that GenAI-assisted creations could be brought into therapy sessions to support emotional reflection and continuity between home and clinic. If youth engage with GenAI at home and bring in narratives or visuals they created, these artifacts can serve as meaningful entry points for clinical conversations. For instance, P4 (psychotherapist) noted the time-saving and relational benefits of using content created by youth at home: "If the teen accepted this narrative and made it their own, it would definitely save time because we'd have a starting point. If they endorsed a strong emotion, maybe it walks them through it." Similarly, P19 (psychologist, PhD) highlighted how such artifacts could uncover recurring emotional patterns: “Understanding a person's interpretation of a situation can lead to meaningful conversations and help intervene if a thought pattern keeps coming up.” AI-generated visuals could also support interpretation of youth's emotional states and preferred coping styles. As P20 (psychiatrist, MD) explained: “If a child says they were an orange triangle one day and a blue oval another, we can look at that together and ask, ‘What was happening then?’ That kind of specific recall is helpful in CBT, it grounds the conversation and helps us analyze what they were thinking and doing at that time.” These envisioned use cases suggest that AI-generated visual and narrative content can help youth externalize emotions, facilitate reflection, and foster engagement in ER sessions both in clinics and at home.

5.1.1 Emotional and Cognitive Support. Clinicians emphasized that AI-generated visuals and narratives have the potential to scaffold youth's emotional understanding and expression in accessible and autonomous ways, while reducing the psychological burden of initiating difficult emotional conversations (Section 4.2). Such support can help youth identify maladaptive thought patterns and articulate their emotions [25, 47, 72]. GenAI can expand youth's emotional awareness by offering diverse visualizations of both intensity and quality. For instance, as clinicians suggested in Section 4.2.1, sadness could be represented not only as a ’rainy day’ but also as ’light rain,’ ’heavy rain,’ or even ’a rainbow after the rain.’ Such variations echo traditional ER tools, such as feeling thermometers and wheels [150, 151], but offer richer scaffolding for both identification and nuanced self-expression.

Clinicians also envisioned GenAI reducing the pressure of emotional expression while preserving youth agency. Unlike paper-and-crayon tasks that may overwhelm youth with a blank page or directive prompts, AI-generated content can provide a flexible “autonomous starting point.” Youth can reflect on AI's initial creations and adapt them to represent deeper feelings. This autonomy is particularly important for youth experiencing emotional dysregulation, especially those whose emotions have been suppressed by authority figures or who internalize symptoms [13, 114].

While earlier ER technologies used predefined visuals or rule-based feedback, GenAI further enables dynamic co-creation by generating adaptive images and narratives that evolve with users’ inputs in real time. In doing so, GenAI complements and extends expressive technologies such as 3D computer games [30] and mixed reality platforms [74, 131] through a highly adaptive, responsive interaction that evolves collaboratively with the user.

5.1.2 Personal Relevance. Clinicians saw promise in GenAI's ability to make ER-focused therapy more personalized and engaging by grounding therapeutic content in youth's own creations. They emphasized the use of metaphors and stories that reflect individual youth's emotions and preferences, consistent with prior work focused on using drawing and narrative for emotional exploration [15, 47]. Moreover, clinicians noted that existing ER worksheets often lack relevance and interactivity, leading to disengagement, echoing findings that clinicians struggle to sustain youth engagement even with creative methods [72]. Many existing tools also rely on reading and lecture-style delivery, which reduces youth's and parents’ engagement and participation levels [139, 140]. To address these challenges, clinicians recommended integrating metaphors drawn from youth's everyday experiences (Section 4.2.3), supported by psychological practices showing that child-created drawings and stories can facilitate emotional expression [135, 154]. Clinicians also highlighted the value of interactivity and multisensory feedback, aligning with prior HCI research on interactive and sensory-based systems that support coping strategies through multimodal engagement [95, 104, 141].

Compared to earlier HCI systems that utilized interactivity and multisensory feedback within predefined structures, such as mobile CBT platforms and biofeedback-based ER systems [41, 82], GenAI can adaptively generate contextually resonant stories and visuals from users’ own creations , such as sketches, colors, or reflections. This generative feedback loop can transform user-made content into evolving emotional narratives, supporting collaborative meaning-making and creation of personalized metaphors, a need highlighted in Kitson et al.’s review [75].

5.1.3 Connections Between the Clinic and Home. Clinicians envisioned GenAI as a bridge between clinical treatment and home-based ER support, particularly for youth with diagnosed mental health conditions (Section 4.4.3). CBT emphasizes collaborative goal-setting and emotional monitoring among youth, therapists, and caregivers [135], which is a key factor in youth ER outcomes [110]. However, this process is often difficult in families less open to emotional dialogue [50] or with youth who are reluctant to share their therapeutic experiences at home [91]. In such contexts, AI-generated visuals and narratives may offer a non-confrontational way for youth to express emotions at home. GenAI's ability to generate metaphors and personalized stories parallels CBT techniques that use metaphor to process complex emotions [44, 97], offering an alternative to direct disclosure and supporting communication for individuals with complex needs [79]. Such conversations around AI-created artifacts could help parents and youth practice ER together at home, whether as homework between sessions or through independent ER activities [130, 140].

In addition, GenAI can bridge clinic guidance and home practice by translating clinician concepts into family-friendly, situation specific artifacts (e.g., visual cue cards or a customized reappraisal storybook) tailored to the youth's age and home caregiver. Clinicians can input their own terms or homework goals, and GenAI can draft a simplified home version with everyday examples for clinicians to refine before sharing with families. This extends prior efforts to connect clinical and home practice, such as HCI work using smartphone apps and clinician portals to link home activities with clinic care [127]. GenAI uniquely supports nuanced, artifact-based continuity by generating and iteratively refining context-specific artifacts. Such artifacts can help youth recall emotions and events in greater detail [72], which in turn can enable richer communication with clinicians.

5.2.1 Emotion Identification. The first phase of Gross's extended process model of emotion regulation is emotion identification, which is the ability to recognize and label one's emotions [49]. Emotion identification is closely tied to emotional awareness, which requires specific vocabulary to explain one's feelings [77]. Since youth are still developing these skills, their limitations can increase risk for internalizing problems [38]. As outlined in Table 2, GenAI's interactive and descriptive outputs (e.g., diverse colors and scenes) could support youth's emotional awareness and understanding. Outputs such as an illustrative story based on a user's day may help youth identify and explore their thoughts, emotions, and behaviors in open-ended ways. This design insight builds on prior HCI systems that supported emotional reflection through mobile apps and emoji-based tools for Autism Spectrum Disorders [39, 62], while extending them through more personalized and contextually responsive visual and narrative prompts for emotion labeling. For safe use, the system should gather key contextual details in case pre-questionnaires miss red-flag symptoms such as suicidal thoughts or paranoid responses. If any red flags are detected, the system should pause use and alert clinicians, as GenAI-supported ER is not recommended for youth with depression or paranoia (see Section 4.3.1). Extending prior HCI work that collects children's content preferences in AI systems [138], parents or clinicians should also provide relevant contextual information or triggers. For instance, if a child with autism is sensitive to sensory overload (e.g., loud noises, flashing lights) [106], parents could record this so the system avoids generating high-intensity or visually inappropriate scenes.

5.2.2 Strategy Selection. The second phase of Gross's ER model, selecting ER strategies, involves evaluating and choosing among options based on cognitive and physiological resources and emotional intensity [49]. Adaptive strategies such as cognitive reappraisal and mindfulness are generally preferable to emotion suppression, though their effectiveness varies by context and age [17, 20]. To support this phase, designers could leverage GenAI to present a variety of adaptive strategies (e.g., cognitive reappraisal, mindfulness) through contextually generated visuals and narratives, rather than emphasizing a single option. For safe use, the system should generate contextually-relevant scenarios based on youth's emotional and behavioral status, offering safe coping options such as positive imagery or physical activity [74, 104]. Since youth with emotion dysregulation or neurodivergent profiles often benefit from having greater autonomy and choice during ER strategy selection [106, 114], parents and clinicians should support flexibility in how they explore options. This may include prompting GenAI to regenerate strategies with more personally relevant images or narratives.

5.2.3 Strategy Implementation. The implementation phase involves translating a selected ER strategy into specific, situation-appropriate tactics [49]. Our findings suggest that GenAI could generate personalized, evidence-based examples of strategy implementation using creative and relatable metaphors (see Section 4.2). Designers could leverage youth's own creations to create interactive practices that feel meaningful to them. For example, if a child depicts sadness as a blue wave, GenAI could animate it into a breathing exercise, adjusting tempo and color to support regulation. While prior HCI work has explored interactive devices such as robots and tangible tools to support coping-skill practice [33, 153], GenAI may complement these approaches by digitally adapting youth-created metaphors into more personalized and contextually responsive regulation practices. To ensure safety, systems should use clear, easy-to-follow visual and narrative creations to guide youth and prevent confusion that could lead to maladaptive coping strategies [135]. For example, instead of vaguely saying “think positively,” the system could scaffold: acknowledge the negative feeling, ask “What does make you happy?”, take the response (e.g., ice cream), and generate a bright image to shift gray thoughts into sunny ones. Lastly, adults should help reduce misinterpretations and reinforce skill practice through co-construction. While prior HCI work has shown that adult–child scaffolding and joint meaning-making are essential for ER practice [109, 130, 140], GenAI further heightens the need for adult involvement. Because GenAI outputs can be misinterpreted or vary in stability, adult co-construction (e.g., “Which part of this suggestion could we reframe together?”) becomes especially important. When guided well, AI-generated drawings and narratives can serve as powerful mediating artifacts that deepen shared reflection between adults and youth.

5.2.4 Monitoring and Adjustment. Monitoring—an HCI-based extension of Gross's ER model—involves deciding whether to continue, stop, or adjust a chosen ER strategy after reflection [128]. Our findings suggest that GenAI could generate visual histories of youth's thoughts, emotions, and behaviors to scaffold this reflection over time. For example, youth's emotions and behaviors could be visualized in age- and symptom-appropriate ways, such as an “emotional storybook library,” aligning with prior HCI work that transformed youth's recorded emotions into storybooks for reflection [115, 159]. This approach echoes Sanches et al.’s call for mental health technologies to rely less on raw tracked data and more on supporting understanding and reflection [118]. For safe use, parents and clinicians should also co-reflect on youth's emotional and behavioral patterns, as highlighted in prior HCI work on collaborative monitoring [24, 66]. With GenAI, it is especially important to consider what youth choose to share and in what form (e.g., direct AI-chat responses or AI-generated metaphorical outputs) to balance privacy with safe ER practice.

5.2.5 When GenAI Should Not Be Used for ER. Clinicians in our study also raised contexts where GenAI-supported ER may require caution or when it should not be used at all. As described in Section 4.3.1, clinicians cautioned that GenAI-supported ER may be unsafe for youth with paranoia or long-untreated trauma, as "AI's image or story creation may unintentionally validate cognitive distortions". They also raised concerns that youth with depression may not be an appropriate fit, noting that reflecting on certain emotions without clinical support "could potentially trigger suicidal thoughts." Based on these concerns, we do not recommend GenAI-supported ER for youth presenting with these symptoms. At the same time, clinicians emphasized that youth without these symptoms may benefit from GenAI-supported ER via its visual and narrative creation, particularly youth with “internalizing symptoms during moments of sadness or fear” (Section 4.2.1) and neurodivergent youth who need “alternative ways to express feelings beyond words” (Section 4.2.3).

However, many youth and families may be unaware of underlying mental health conditions that require clinical support [72]. Future GenAI tools for ER should therefore embed validated screening questionnaires such as PHQ-9 [112], GAD-7 [134], ASQ [59], and PWQ [42]. These can be administered during onboarding and at regular intervals. If screenings flag a user as at-risk (e.g., suicidality, paranoia, trauma indicators), the system should not launch and instead redirect to clinical professionals. Users who do not screen at-risk may proceed with regular re-assessment, ensuring that at-risk youth are routed to supervised care while maintaining safer access for others. Additionally, we recommend that GenAI-based ER tools be used only with the presence of an adult for children ages 6–12, and under adult supervision for adolescents (13–17). We do not recommend use for preschool-aged children (5 years) due to risks such as false memory implantation and distorted recall [105].