Abstract

This article presents the Immersive Online Reading for Mental Wellbeing project, developed through Northern Ireland's Connected 5 knowledge exchange fund. The project explored how a storytelling-based bibliotherapy programme delivered by the Verbal Arts Centre might be remediated for individual online use through AI-supported systems within a hypertext frame of reference. Working collaboratively across higher and further education, we examined how humanities scholarship, digital humanities methods, and computing expertise might contribute to accessible and scalable mental wellbeing interventions. We outline the challenges of translating a group-based storytelling model into a hypertextual individual digital experience. We describe the development of a multi-agent architecture and focus in particular on the annotation of short stories for AI retrieval and moderation. The project functioned as a proof of concept rather than a full prototype. Our findings indicate both the potential and the limits of AI in narrative-based wellbeing contexts. Human curation, ethical safeguards, and interdisciplinary collaboration remain central, underpinned by the hypertext goal of augmenting human cognition.

1 Introduction

Chatbots are widely adopted hypertext applications, text-to-voice based interactive interfaces that offer adaptive explorations of documents or system features. Chatbots find, however, limited interest in the Hypertext research community 1. A reason may be that they have only recently had a resurgence in popularity (during the 2019 Covid-19 pandemic). Another reason could be due to their apparently limited application to long-standing hypertext community interests in narrative and creative works. Furthermore, it could be argued that chatbots are already well explored and that they represent a straightforward classic hypertext system. They use a layer of natural language processing techniques to capture and map user intent into a traditional workflow. Any variation is just superficial and concerns the modality, not the mechanics or their capabilities. However, a new generation of chatbots are powered by Large-Language models (LLMs) replacing the imperative, pre-designed workflow with a new type of engine. This actually provides new capabilities for fluid navigation and for producing genuinely new experiences.

During the current resurgence of AI, a strong effort has been underway to address critical bottlenecks in provisions such as healthcare, for example. Indeed, medicine is among the most invested fields of AI, characterised by groundbreaking results in specific fields such as drugs development and image diagnostics. However, mental health and wellbeing remains a huge, untapped potential area for which LLMs, with a human-like ability to adapt and converse, seem to be a perfect fit. Hence, they represent a potential application worth exploring.

According to the Department of Health, 38% of people in Northern Ireland report concerns for their mental health in 2024/25, an increase compared to previous years. In addition, 25% of respondents reported experiencing loneliness [7]. There is a need for cost effective interventions that can support the mental wellbeing of the population and foster community cohesion.

The Verbal Arts Centre in Derry/Londonderry has long experience of immersive reading programmes grounded in discussion and shared reflection. These have been proven to increase wellbeing and resilience in groups such as schoolchildren [9]

The Immersive Online Reading for Mental Wellbeing project emerged from Northern Ireland Department for the Economy's Connected 5 fund. The Connected programme is designed to strengthen knowledge exchange between academia and industry, including the creative and cultural sectors. The fund supports collaboration across higher and further education and encourages exploratory, proof-of-concept work.

Our guiding questions were straightforward but demanding. What might arts and humanities academics contribute to a digital wellbeing intervention? Could AI systems support accessibility without eroding the human qualities that define storytelling? What forms would a proof of concept in using stories and interactive discussions about those stories take when moved from in-person group settings to individual online use? How would an overall hypertextual structure unfold in this context?

We approached these questions in partnership with the Verbal Arts Centre, a charity based in Derry/Londonderry with long experience of immersive reading programmes grounded in discussion and shared reflection. Technical collaboration was provided by Belfast Met, whose expertise in computing and multimedia informed interface design and system thinking.

This was exploratory research with in-built uncertainties. We hoped that the open-ended approaches to options might provide ways forward that would arise precisely because of this co-creative process of exploration.

2 Literature Review

The activities of Verbal are grounded in bibliotherapy approaches. The concept of bibliotherapy has expanded into a wide-ranging category that includes diverse aims and methods. Liz Brewster notes that there are many interpretations of what defines the practice, with contested approaches as to who can or should deliver it and how it should be administered [5]. In her co-edited volume (with Sarah McNicol), Brewster provides useful summaries of varied critical and practice-based approaches that show reading as used variously for ‘information, guidance, and solace’ (2018, p.2) [14]. Although bibliotherapy developed largely in the US and UK as a therapeutic treatment, particularly post-World War I, initiatives using literature for wellbeing now also appear in conflict-recovery settings, in community-building initiatives, as part of public library offerings and embedded in staff psycho-social support schemes [12]. Edmund G. C. King in a section ‘Towards a history of bibliotherapies’ in [6] draws on the recent critical survey by Gill Partington of the developing research literature on bibliotherapy, noting how she suggests that thinking in terms of ‘bibliotherapies’ rather than ‘bibliotherapy’ can be one way of acknowledging the ‘variety of disparate activities and ideas’ now bundled within the term [18].

Though generally understood then as the use of literature to promote mental health, bibliotherapy takes several forms depending on intention, context, audience and method. Nicholas Mazza's seminal text, Poetry Therapy. Theory and Practice [13] is often seen as a key instigator of the field and the journal edited by Mazza, Poetry Therapy: The Interdisciplinary Journal of Practice, Theory, Research, and Education continues to theorise the various approaches, whether institutional, clinical or developmental. Institutional bibliotherapy takes place in hospitals, prisons or recovery centres like those run for military veterans. This type of intervention has also been called ‘Literary Care-giving’ [10]. Clinical bibliotherapy, usually led by a medical practitioner, aims to encourage behavioural change. Developmental bibliotherapy centres on self-development and sustaining mental health in people without a mental health diagnosis. McCarthy Hynes and Hynes-Berry distinguish between ‘reading bibliotherapy’ where readings are prescribed to be read by an individual and ‘interactive bibliotherapy’, where group exchange and the process of developing a relationship between participant, text and facilitator are key [11].

In this context, the Verbal approach (subject of this paper) can be understood as a form of developmental, interactive bibliotherapy. It involves shared reading and storytelling emerging from that experience of reading to support mental wellbeing in a non-clinical setting. Reading in a social group has also been identified as a central component of the twenty-first century reading sphere [9]. The Covid-19 pandemic lead to experimentations with digitally-mediated social reading, with promising results but also perceived dangers [3].

The Verbal Arts Centre in Derry/Londonderry has successfully experimented with digital support for its pioneering “reading rooms” programmes, but this activity has so far focused on the use of conversational media to explain the working of its programmes, rather than remediating the programmes themselves [16]. The approach in this previous collaboration was informed by a “hypertext as method” [1] framework that focused on enabling the augmentation of human intellect, building upon Ted Nelson's seminal theorisations [17] and Doug Engelbart's conceptualisation of design frameworking [8].

The use of AI in storytelling is not new. AI and other forms of adaptive or generative mechanisms have been part of experiments to, for example, enrich narrative with dynamic animations since the early 2000s. More recent approaches have focused on applications for children such as, for example, parent-child bonding using interactive, conversational books that would help parents co-develop narrative skills with their children, or support the development of reading skills [7]. However, the use of AI in sensitive areas, such as children or health, has been limited due to legitimate challenges in accounting for safety and privacy. There is currently a lack of coherent proven approaches to the design of such systems, clearly still in an early maturity stage and needing a more significant effort in reconciling practice with theory.

3 Background

3.1 Verbal's Storytelling Model

Verbal delivers outreach programmes built around storytelling, immersive reading, and structured discussion. Its work often involves marginalised communities, cross-community groups, older adults, and educational settings. The ethos is participatory and dialogic.

A typical in-person programme lasts six to eight weeks. At each weekly session, a trained facilitator reads a short story in segments. After each segment, participants are invited to discuss characters, dilemmas, and emotional resonances. The session takes approximately 45 to 50 minutes.

Behind each session lies careful design (see Figure 1). Stories are either commissioned from contemporary writers or selected from out-of-copyright material. Commissioned stories are preferred because they are more likely to reflect contemporary experience. Each story is scaffolded by discussion prompts. Psychologists review materials to ensure that the intervention aligns with wellbeing goals and avoids unintended harm, and is suitable for its intended audience, such as schoolchildren or older adults.

Figure 1: Verbal's in-person programme for mental health and wellbeing.

The stories are not simplistic or overtly therapeutic. They are literary. They contain ambiguity, moral tension, and recognisable human conflict. For example, ‘Shooting an Elephant’ by George Orwell which, though more of an essay than a story, is a text in which the narrator faces a moral dilemma shaped by authority, colonial politics, and public expectation. Such texts invite reflection rather than prescribe comfort.

3.2 The Challenge of Online Remediation

The power of the bibliotherapy programme lies in dialogue, where meaning emerges collectively. Participants test interpretations against one another and stories become animated through discussion. During the COVID-19 pandemic, Verbal was forced to move some programmes online. This shift revealed both possibilities and constraints. Online delivery expanded reach but introduced new costs, with moderators requiring additional training and participants facing technological barriers, such as poor internet connections or lack of suitable devices. Coordination and software management also demanded resources.

The Verbal Arts Centre therefore contacted long-term collaborators at The Open University posing a new question. Could their bibliotherapy model be redesigned specifically for individual online delivery? Hypertextual frameworks have been used successfully in the collection and study of reading experiences [2]. Rather than replicating a group session through video conferencing, could a hypertext-based, mobile-optimised platform support solo users who might not participate in group calls?

Three principles were agreed early on in the one-year project. Passive content libraries would not work. Generic mindfulness tools would not reflect Verbal's distinctive storytelling ethos. Rigid, one-size-fits-all pathways would undermine user agency. The system had to feel hypertextual: responsive and personal.

At the same time, ethical concerns were foregrounded. In imagined user journeys, the team considered cases in which an individual experiencing anxiety or depression might use the platform alongside clinical support. Any digital intervention would require clear boundaries and signposting to human-provided support services if serious distress were detected. The confines of the funding scheme meant that we could not build such a system, but the need for human oversight remained central to our design thinking.

3.3 Reading Online: Design Questions

To remediate immersive reading for digital delivery, we had to reconsider what “reading online” means [15]. Numerous platforms support online reading, including book groups and participatory writing sites such as Wattpad. These platforms assume relatively high levels of digital literacy and active participation.

Verbal's target users for this project were different. The aim was to provide structured storytelling for wellbeing support, potentially offered to employees through partnerships with businesses and organisations. The software therefore needed to be multi-platform, mobile-optimised, accessible, and oriented toward listening as much as reading. These considerations generated practical questions. Should stories be delivered through human voice recordings or synthetic AI-generated speech? Should images accompany the narrative? If so, should they be static or dynamic, realistic or stylised? Should each character in a story have a distinct voice? Should there be background sound, music, or ambient effects?

The project therefore shifted from producing a polished prototype to constructing proofs of concept. We aimed to understand what would be required for planning rather than to complete the full build.

4 Methodology

In this work, we focused on an existing specific mental health and wellbeing intervention using a combination of narrative and guided group discussions. The object of the study was not the efficacy of the programme, but the ability to use an LLM to remediate its core essence into a chatbot. This would result in an alternative form of provision, not based on in-person, in-group within a controlled session with fixed duration but focused on single individuals, online and fitting the intervention to the short time intervals they have available.

To this end, we engaged in a system design process aimed at identifying the characteristics of the medium: (1)emerging behaviour, (2.1) internal mechanisms and (2.2) practices that not only focus on (2.2.a) the interactions of the final user but aim at (2.2.b) deep integration within the experience, assets and way of working of the organisation and team providing the mental health interventions. The outcome of this process was an architecture that clarifies and integrates design decisions informed by the mental health and narrative expertise of Verbal and their vision for organisation service provision. On the latter, we addressed challenges new and external to the existing programme, related to the need to create conditions to verify and monitor the efficacy of the intervention, support its improvement and significantly scale up and manage the number of participants.

The design was articulated as an iterative process that moved its focus from identifying the core characteristics of Verbal's programme to the definition of the system and practices around it, combining requirements collection and analysis, experiments and validation of design choices with experts.

5 System Design

5.1 Co-design Process

The process combined pre-design inquiry, co-design practice, and structured knowledge exchange between partners with different forms of expertise [4]. We began from the existing intervention protocol rather than from technical affordances. This pre-design phase focused on understanding what should be preserved when moving from facilitated group settings to individual online use. It clarified that the aim was not to replicate surface features such as discussion prompts, but to retain the underlying purpose of guided reflection through narrative. Co-design workshops then translated these insights into system requirements. These sessions exposed differences in language and assumptions across disciplines. Literary scholars described narrative in terms of tone, ambiguity, and moral tension, while technologists required explicit categories and parameters. Considerable effort was spent aligning these perspectives into shared representations that could inform system behaviour. For example, early ideas such as adding bespoke artwork or rich media for each story were explored but quickly found to be impractical at scale. Recording hundreds of stories with professional voice actors, creating bespoke soundtracks, and producing images for each narrative would exceed the scope of a small charity. AI-generating media tools exist and are increasingly sophisticated, but their use raises questions of quality, coherence, and sustainability, not to mention the risk of dwindling revenues for creators [21], including the writers Verbal relies upon to create new stories. This led to a prioritisation of narrative structure and annotation over audiovisual enhancement.

Knowledge exchange was continuous throughout. Insights from experiments fed back into design discussions, while domain expertise reshaped technical expectations. This iterative loop allowed the team to progressively refine both the conception of the system and the understanding of what AI could realistically support. Rather than converging on a fixed solution, the process clarified boundaries, surfaced trade-offs, and established a foundation for future development.

5.2 From Programme to System Architecture

Understanding Verbal's in-person programme was the foundation for technical design. The first in-person session of a Verbal programme includes enrolment, expectation setting, story delivery in segments, moderated discussion, and reflective closure. It also includes ongoing evaluation across multiple weeks.

Figure 2: AI roles for user and programme management.

To translate this into a digital context, we conceptualised a multi-agent architecture. At its core, the design included four agent roles about user and programme management (see Figure 2):

An onboarding agent to introduce the platform, clarify its scope, and collect light-touch information about user expectations and current mood.\

A triage agent to recommend suitable stories based on user input.\

A moderator agent to guide the user through the story, posing questions at key moments.\

An insight or monitor agent to analyse session patterns and support iterative improvement.\

Surrounding these were user knowledge bases and curated narrative assets. The architecture also incorporated engagement functions to encourage return visits while avoiding intrusive prompts. The aim was not to replace human facilitators entirely but to model certain aspects of their work in a structured, accountable way. Scaling required control and evaluation to be embedded in the architecture rather than added later.

The most crucial moment to plan was the hypertextual branching, the moment when users would need to search the available stories and choose the one most suitable to their present conditions and goals. The system would require the ability to categorise and recommend stories based on user prompting.

5.3 Proofs of Concept

We divided the work into four interrelated proofs of concept (PoCs), critical to demonstrate the feasibility of the overall system architecture (see Figure 3).

PoC1 addressed session moderation. Could an AI agent approximate the pacing and questioning style of a trained facilitator? Early experiments suggested that generic chatbots were inadequate. They tended either to overproduce text or to respond superficially.

PoC2 focused on user triage and narrative recommendation. The challenge was to gather minimal but meaningful information. Questions such as “How are you feeling today?” or “Would you prefer a calm or an energising story?” needed to be carefully framed to avoid clinical overreach while still enabling useful matching.

Figure 3: System architecture and the four proof of concepts.

PoC3, narrative description and retrieval, became the most demanding. AI systems cannot intuit the qualities of a story unless these are explicitly encoded. If an agent is to retrieve a story about moral conflict in a contemporary urban setting with a cautiously hopeful resolution, those features must be described in machine-readable form.

PoC4 concerned session analysis and user characterisation. Could behavioural indicators such as completion rates or response length inform adaptive design without compromising privacy or leading to data over-interpretation? This question remains open.

6 Discussion

6.1 Teaching AI to Use Narrative

While experiments have been conducted to use AI to identify narrative structures [19], the narrative description task in the project required sustained collaboration between creative writing and digital humanities perspectives.

We experimented with tagging narrative elements such as theme, character type, and emotional trajectory, using a tabular format. These tags were intended as signals to guide retrieval and recommendation. Stories were labelled for emotional valence, setting, character age, plot type, and resolution. This format was efficient for humans but proved insufficiently expressive for reliable AI retrieval. We then moved to narrative-style annotations. For a 500-word story, we produced descriptive summaries exceeding 1,000 words. These summaries explained setting, character motivation, central conflict, tonal shifts, and thematic implications. In this richer format, AI agents were better able to select appropriate stories and justify their choices. This shift revealed a key tension. The more we tried to specify meaning for the system, the more labour-intensive and less scalable the process became. Annotating at this level is feasible for a small corpus but difficult to scale to hundreds of stories without significant resource investment.

Parallel to this, we conducted iterative experiments with LLM-based agents acting as moderators and “readers.” These experiments were structured as probes rather than full implementations. We tested how agents responded to prompts, how they handled constraints such as sticking to a source text, and how they formulated reflective questions. The outcomes informed successive refinements of prompts, annotations, and interaction design. They also highlighted persistent limitations, such as the tendency of the system to extend or alter source material and its inability to reliably infer narrative purpose without explicit instruction.

Parallel experiments confirmed another insight. AI systems can generate plausible short stories, but these lacked the textured ambiguity and lived resonance that characterise the commissioned work used by Verbal. They recombined patterns without experiential grounding [20]. In wellbeing contexts that rely on moral nuance and emotional recognition, this limitation matters.

6.2 Ethical and Environmental Considerations

Discussions during and after the project also addressed broader concerns. Large-scale AI deployment carries environmental costs. Training and maintaining models require substantial computational resources and energy-intensive data centres. Any proposal to scale AI-supported storytelling must consider sustainability alongside functionality.

Ethical safeguards are equally critical. Automated systems are not equipped to manage acute mental health crises. Any deployment in wellbeing contexts would require clear escalation pathways and continued human oversight. The proof-of-concept nature of our project allowed us to bring these issues into focus, for instance discussion of how triage might be important at the onboarding moment. It allowed us to surface these issues without prematurely implementing inadequate solutions.

7 Conclusions and Lessons Learned

The findings of this project point to a fundamental shift in how narrative systems for wellbeing must be conceived when AI is involved. The introduction of an LLM does not simply automate existing practices. It creates a new design space in which narrative, interaction, and interpretation must be explicitly constructed rather than assumed. A key insight concerns the gap between narrative form and narrative purpose. While AI systems can process and reproduce textual patterns, they do not grasp why a story is being used in a given context, as highlighted in the Proofs of Concept. In the case of this collaboration with Verbal, the purpose is not entertainment or information alone, but the facilitation of reflection linked to mental wellbeing. Our experiments show that this purpose must be externalised through scaffolding. Without detailed guidance, the system cannot distinguish between a story as a sequence of events and a story as a vehicle for reflection. This limitation reframes design as a process of encoding intent, not just content.

This leads to a second insight about the role of annotation and representation. The attempt to make stories machine-readable exposed the difficulty of translating literary knowledge into operational terms. Simple tagging systems proved insufficient, while richer narrative descriptions introduced issues of scale and sustainability. This suggests that current approaches to knowledge representation are not well aligned with the complexity of narrative as used in humanities-informed interventions. Future work may need to explore hybrid models that combine structured metadata with more flexible forms of interpretive guidance.

The project also highlights a paradox in the use of AI as a “reader.” On one hand, the system exhibits behaviours that resemble human reading, such as filling in gaps, making assumptions, and moving non-linearly through material. On the other hand, these behaviours are not grounded in understanding. They are generated through statistical association rather than lived experience or interpretive intent. This creates a tension. The system can mimic certain surface aspects of reading while lacking the depth that makes reading meaningful in a wellbeing context. Designing with this tension requires careful control of outputs and clear boundaries around the system's role.

Co-design and knowledge exchange were essential in revealing these issues. The collaboration made visible the implicit knowledge embedded in existing practices. These practices are literary in the broadest sense as they depend on multi-layered and interlocking understandings of how narrative is a form of knowledge-in-process. The collaboration also showed that interdisciplinary work is not only about combining skills but about negotiating different epistemologies. What counts as a meaningful feature in a story differs across domains, and aligning these views is a non-trivial task. This has implications for future projects, which will need to allocate time and resources to this alignment rather than treating it as a preliminary step.

7.1 Next Steps

The work raises broader questions about scale and sustainability since the level of human input required to guide AI behaviour remains high including in ongoing monitoring and ethical oversight. While AI offers possibilities for extending access to wellbeing interventions, it creates a demand for different forms of human labour, such as annotation, prompt design, and system supervision. This all suggests that AI-supported bibliotherapy could be understood as an emergent socio-technical system as its effectiveness depends upon the careful integration of narrative theory, design practice, and technical expertise. Our project helps to make these (inter) dependencies explicit, pointing out some of the challenges that need to be addressed for such systems to move beyond proof of concept.

A further challenge lies in the need to provide a feedback loop. The system cannot independently assess whether a story has had a positive effect on a reader, nor can it refine its future recommendations based on such outcomes. Building this kind of adaptive system would require substantial additional resources and infrastructure. As it stands, the project offers only a glimpse of what AI-assisted, hypertextual reading experiences might become, rather than a fully realised model.

Interestingly, the experiment proved especially valuable from a critical and reflective perspective. Engaging with a non-human “reader” prompted deeper questions about the nature of narrative and the act of reading itself. It highlighted the extent to which human readers bring context, emotion, and interpretive frameworks to texts—elements that go far beyond decoding words or recognising patterns. While AI can identify syntactical relationships and common associations (such as words that frequently appear together), it cannot understand metaphor, symbolism, or the emotional resonance behind descriptions. A phrase like “the sky was blue and without clouds” might suggest calmness or contrast with a character's inner turmoil, but such layers of meaning remain inaccessible to AI.

Current developments in AI are addressing such limitations by, on the one hand, improving the models but, on the other hand, developing practices around correct integration. Of this second type, the current focus has moved beyond prompting engineering, focused on how to optimise requests, toward “harnessing”. Harnessing is about building an architectural fence around a model. By carefully designing the operational context, curating data assets, and defining the overall system architecture, developers provide the exact constraints required for a specific task. While AI cannot overcome its fundamental lack of understanding of Human phenomena, it can be steered to activate the correct statistical pathways so it can tap into the appropriate patterns. Our multi-agent architecture breaks down AI roles, clarifies expectations, goals and provides curated narratives and insights on past interactions. Our approach is compatible and coherent with harnessing, making our design still valid and in line with current trends in AI.

This absence of deeper understanding ultimately underscores what human readers contribute. Language acquisition and interpretation in humans are embedded in lived experience, metaphor, and shared cultural meaning. By contrast, the AI lacks any sense of purpose or embodied perspective. Efforts to “teach” it resemble, in some ways, guiding a child through language—pointing out objects and naming them—but without the underlying cognitive and emotional development that gives those words significance. In this way, the limitations of AI make the richness of human reading practices more visible.

Despite these challenges, the work points toward practical next steps. Expanding hypertextual connections throughout the material, rather than concentrating them at the beginning, could create a more dynamic reading experience. Incorporating visual diagrams may also help illustrate complex relationships within the text, though such representations require careful design to remain accessible. Overall, even without a fully developed system, the process has clarified key ideas and opened up productive directions for further exploration.

The Immersive Online Reading for Mental Wellbeing project demonstrates both the promise and the complexity of remediating bibliotherapy through AI-supported systems. We succeeded in articulating requirements, modelling a multi-agent architecture, and testing narrative annotation strategies. We clarified what would not work and identified where human expertise remains indispensable.

AI can assist with accessibility, retrieval, and adaptive structuring. It cannot replace the relational depth of facilitated discussion or the craft of literary storytelling.

For digital humanities, the project affirms that hypertextual and networked thinking are not only technical constructs but literary and epistemic ones. Designing such systems requires attention to narrative structure, affect, ethics, and scale.

The proof of concept now awaits further development. Its future depends not only on technological capability but on careful, interdisciplinary stewardship of story, system, and reader.

Notes

1We counted four papers explicitly mentioning bots in the title, from the curated references of ACM Hypertext Conference from Mark Anderson.

Source

ACM full-text HTML

ACM Digital Library

Reconciled against the supplied 7-page PDF using coordinate-aware layout extraction (SHA-256: 7b4d17112530aa7b36ce72cb300c8713d6bed6ca06d553cb160533fcae4bd36f). ACM Reference Format: Francesca Benatti, Siobhan Campbell, and Alessio Antonini. 2026. Connected Collaborations: Remediating Immersive Reading for Mental Wellbeing through AI. In 37th ACM Conference on Hypertext (HT '26), September 14--18, 2026, London, United Kingdom. ACM, New York, NY, USA 7 Pages. https://doi.org/10.1145/3800935.3830847

References

[1] Claus Atzenbeck and Peter J Nürnberg. 2019. Hypertext as method. In Proceedings of the 30th ACM Conference on Hypertext and Social Media. 29–38.

[2] Francesca Benatti, Shafquat Towheed, Sally Blackburn-Daniels, and Alessio Antonini. 2024. @ TellMeWhatUReadingbot: the Multi-modal Strategy of the READ-IT Project for Collecting Experiences of Reading. In Proceedings of the 35th ACM Conference on Hypertext and Social Media. 217–222.

[3] Judith Blundell and Simon Poole. 2023. Poetry in a pandemic. Digital shared reading for wellbeing. Journal of Poetry Therapy 36, 3 (2023), 197–209.

[4] Keld Bodker, Finn Kensing, and Jesper Simonsen. 2009. Participatory IT design: designing for business and workplace realities. MIT press.

[5] Liz Brewster. 2018. Bibliotherapy: a critical history. Bibliotherapy (2018). https://api.semanticscholar.org/CorpusID:149890330

[6] Siobhán Campbell, Sara Haslam, and Edmund GC King. 2025. A Hundred Years of Bibliotherapy: Healing Through Books. Taylor & Francis.

[7] Marc Cavazza, Fred Charles, and Steven J Mead. 2001. AI-based animation for interactive storytelling. In Proceedings Computer Animation 2001. Fourteenth Conference on Computer Animation (Cat. No. 01TH8596). IEEE, 113–120.

[8] Douglas C. Engelbart. 1962. Augmenting Human Intellect: A Conceptual Framework. Technical Report. Air Force Office of Scientific Research.

[9] Danielle Fuller and DeNel Rehberg Sedo. 2013. Reading beyond the book: The social practices of contemporary literary culture. Routledge.

[10] Sara Haslam. 2020. Reading, trauma and literary caregiving 1914-1918: Helen Mary Gaskell and the War Library. Journal of Medical Humanities 41, 3 (2020), 305–321.

[11] Arleen McCarty Hynes and Mary Hynes-Berry. 2019. Bibliotherapy: The interactive process a handbook. Routledge.

[12] Edmund King, Sara Haslam, and Siobhan Campbell. 2018. Bibliotherapy: How Reading and Writing Have Been Healing Trauma since World War I. The Conversation 15 (2018).

[13] Nicholas Mazza. 2016. Poetry therapy: Theory and practice. Routledge.

[14] Sarah McNicol and Liz Brewster. 2018. Bibliotherapy. Facet.

[15] Simone Murray. 2018. Reading online: Updating the state of the discipline. Book History 21, 1 (2018), 370–396.

[16] Iman Naja, Natalie Divin, Sarah Coward, Siobhan Campbell, Francesca Benatti, and Alessio Antonini. 2024. Conversational Media for Inclusive Access to Mental Health Interventions for Schoolchildren. In Proceedings of the 35th ACM Conference on Hypertext and Social Media. 29–34.

[17] Theodor Holm Nelson. 1965. Complex information processing: a file structure for the complex, the changing and the indeterminate. In Proceedings of the 1965 20th national conference. 84–100.

[18] Gill Partington. 2023. Bibliotherapy: A survey of literature. (2023).

[19] R Rajakumari et al. 2025. AI Powered Fabula and Syuzhet Modeling for Narrative Structure Analysis in Contemporary Cinema. In 2025 2nd International Conference on Artificial Intelligence and Knowledge Discovery in Concurrent Engineering (ICECONF). IEEE, 1–9.

[20] P Fathima Shifa and T Shameema. 2025. The Rendezvous of Literature and Artificial Intelligence: A Fortune or a Fiasco?New Literaria 6, 1 (2025), 1–9.

[21] Florent Vinchon, Todd Lubart, Sabrina Bartolotta, Valentin Gironnay, Marion Botella, Samira Bourgeois-Bougrine, Jean-Marie Burkhardt, Nathalie Bonnardel, Giovanni Emanuele Corazza, Vlad Glăveanu, et al. 2023. Artificial intelligence & creativity: A manifesto for collaboration. The Journal of Creative Behavior 57, 4 (2023), 472–484.