AI as a guiding storyteller in care—uncovering stories and understanding

This is part of a series exploring care-experienced young people's perspectives on AI in UK children's social care. Read the introduction and overview here.

This article examines how AI is increasingly used to document children's lives in care—and what care-experienced young people fear about being "reduced to AI summaries." In this article, you'll explore:

Why stories matter — a care-experienced young person's testimony reveals what's at stake when we discuss AI in care records
How children's stories are documented and heard in care — How children are written about in records and why relationships with social workers matter
AI as a storyteller — How AI tools are being used to transcribe and write assessments
What care-experienced young people and social workers told me — Six key findings from my research on AI's potential and risks
What this means for AI product development and social care practitioners — Practical steps to resist reduction and expand understanding

Why stories matter: Lamar's testimony

In February 2025, Lamar, a 19-year-old care-experienced young person, testified before the House of Commons Education Committee. During their years in care, Lamar struggled to maintain relationships with siblings, community, and local areas due to frequent placement changes. Calling for change, Lamar highlights that they cannot remember most of their childhood, and rewrites their story, reminding themselves of "doing well for my age".

I open this article with Lamar's testimony because it reveals what's at stake when we discuss AI in children's social care. Ownership of story and identity is critical for children in care, with research showing that when care-experienced individuals later access their care records, it can support trauma healing, identity building, and memory recovery. Lamar's inability to remember their childhood—despite having care records documenting it—highlights fundamental questions: whose voices shape these records, how are children represented, and with AI increasingly used to document cases, what are the implications for how children's stories are told?

How children's stories are documented and heard in care

Just-a-child: unheard and filtered voices

How we think of children fundamentally shapes how we relate to them. As childhood studies scholar Rachel Rosen describes, dominant constructions depict childhood as "a time of happy innocence, but also an essentially vulnerable phase of life that required special protections", alongside perceptions of children as "wild or antisocial, requiring the firm regulation of an adult". These views profoundly influence 'adult-child power relations' within children's social care.

Research analysing case record practices reveals practitioners describe children through adult lenses, with assumptions often taking precedence over authentic representation. Practitioners dismiss children's voices—researchers documented them saying 'they're just children'—and limit involvement to child-appropriate areas. Children often lack unique identities in reports, becoming indistinguishable from siblings or portrayed solely through vulnerability.

Care records document family history, education, health, and views, serving dual purposes: evidencing decisions while 'constructing a social reality' about families. Research shows written evaluations often position children as objects of inspection rather than co-creators of their stories, missing children's own words and containing gaps that distort experiences.

Even when children are formally included, social workers have significant editorial discretion over how children's voices are interpreted. Roy and colleagues found in a 2025 study that 60% of children weren't consulted about discharge procedures, and only 1/3 participated in court proceedings. Children's words are consistently filtered through professional interpretation, raising concerns about accuracy and reproducing power imbalances. Misinterpretations negatively impact trust in social workers.

Why relationships with social workers matter for feeling heard

The Munro Review of Child Protection (2011) highlighted that trusting relationships with practitioners are central to enabling children's participation—a finding consistently replicated in subsequent research. Munro found that social workers felt unprepared to communicate with children and were constrained by administrative tasks. Over a decade later, the MacAlister Review (2022) and the House of Commons Education Committee (2025) continue to document these same challenges.

Frequent changes in social workers due to high staff turnover and heavy workloads create infrequent, inconsistent encounters, significantly undermining relationship-building and contributing to young people's lack of trust. Children engage more meaningfully when they have stable, consistent relationships with professionals who demonstrate genuine care and interest.

Studies spanning over a decade show children identify listening capacity as key to 'good' social workers: stopping when children speak, being present rather than distracted by paperwork, and showing genuine interest rather than assuming they know better. These relationships operate within what sociologist Berry Mayall calls generational order—systematic power structures where adults maintain authority over children. While complete equality remains unrealistic, reducing unnecessary inequalities becomes essential for children's participation.

AI as a storyteller

Increasingly, social workers use AI tools like Magic Notes and Microsoft Copilot to transcribe meetings, summarise, and write assessments using Large Language Models (LLMs). LLMs "predict and generate plausible language" based on calculating the probability of words appearing next to each other. They can misuse words and reproduce embedded biases around race, gender, and religion, whilst being particularly prone to what AI researchers call 'hallucination' and 'sycophancy'—producing unfounded information and over-agreeing to avoid conflict.

These tools have demonstrated practical benefits—social workers report that automated transcribing allows them to focus more fully on conversations with children and families rather than note-taking. However, these technologies cannot recognise the non-verbal communication that practitioners are encouraged to interpret when working with children. If AI learns predominantly from historical social care records, it risks perpetuating the filtered, adult representations of children, where structured and probabilistic ways of organising text may overshadow complex narratives and individuality.

What care-experienced young people and social workers told me

About this research and its participants

Through qualitative research — one-to-one interviews and a focus group — I engaged seven care-experienced individuals (aged 19-39), two social workers, and three advocates to understand their experiences, perceptions, and hopes regarding AI tools in care workflows. These participants brought diverse stances to our conversations about AI in children's social care — from pragmatic optimism about administrative efficiency to deep scepticism about whether technology would genuinely benefit young people.

I showed them real screenshots of AI tools being used or trialled in children's social care — from transcription and assessment writing to case record exploration and automated risk scoring:

Magic Notes — AI that transcribes, summarises & helps write assessments;
Reimagining Care Management Demo — AI for a better view into case records and the life of a child in care;
Policy Buddy — AI for a better view into local and personal care policy and procedures;
Allegheny Family Screening Tool — AI for automated decision-making and risk assessments.

Then I asked: What do you think? What are your fears? What are your hopes?

All participant names have been changed for privacy. Given the sensitivity of participants' experiences, this research was conducted under ethical approval from University College London, with particular care taken around informed consent, participant wellbeing and the writing of findings.

Individual participant stances:

Sam (care leaver, one-to-one interview) — Saw AI's potential for reducing admin, but strongly emphasised children should not be reduced to "AI summaries". AI should help social workers think long-term and empower young people to understand their rights.

Jessie (care leaver, one-to-one interview) — Viewed AI as useful but emphasised "it's only as good as the person using it" and should enable "better tailored help". Their key concern was young people feeling processed rather than cared for.

Safeguarding SW (social worker, one-to-one interview) — Initially worried AI felt like "an intrusion" on relationship-based practice, but became enthusiastic about its admin potential, emphasising the crucial need for human analysis—the "So what?"

Independent SW (social worker, one-to-one interview) — Thought AI could handle everything, but "shouldn't be the be-all, and end-all", cautioning that social workers still need to source check and apply their own analysis.

Sceptically Suspicious group (care-experienced young people and advocates, focus group) — Were deeply sceptical AI would benefit young people, calling it "completely pointless" and worried it would make young people feel terrible and undermine relationships with social workers.

Cautiously Curious group (care-experienced young people and advocates, focus group) — Were open to AI improving care processes, but highlighted the gap between professionals' focus on time-saving and young people's lived reality of needing individual understanding: "It's about someone's life."

What participants understood about AI:

All participants had used ChatGPT, with some also mentioning CoPilot, MidJourney, and transcription tools like Otter. They found AI helpful for 'quick answers' and research. Participants had varying levels of understanding about how AI works — Sam explained Large Language Models and how they mimic language, while others thought of AI as a search tool that synthesises information and talks like a human. All participants recognised AI's intelligence through its ability to make choices and construct responses to human prompts.

Importantly, all participants were well aware AI makes mistakes. They showed strong caution, mainly because they'd witnessed errors themselves — transcription mishearing words, producing nonsensical text, and being particularly prone to number errors. Participants didn't view mistakes as reason to abandon the technology but recognised the need to check outputs and edit them.

Participants revealed that care records and the way children's stories are captured profoundly shape young people's experiences of care through what gets written—and how it gets written.

Manual note-taking creates inaccuracy

Sam described how time-constrained meetings led to vague notes with "really egregious issues sometimes from the really inaccurate note-taking." Information gets missed, misheard, or recorded without the nuance that shapes its meaning. All participants valued AI's ability to record precisely what was said—"help make sure that you get every detail" [Jessie], with potential to "stop fights or arguments with young people" [Cautiously Curious]. Sam warned against AI "chang[ing] the meaning inadvertently."

However, all participants recognised that AI makes mistakes—transcription errors, hallucinations, and misconstrued summaries, especially in the types of unstructured, natural conversations that emerge from direct work with families. The Sceptically Suspicious worried this "won't really save time in the long run."

The 'bad' things written about young people affect their care

Care-experienced participants described discovering awful things in their records only after leaving care—information they didn't have an opportunity to see or correct whilst it was actively shaping how they were treated. Sam explained a friend's particularly damaging experience—foster carers' complaints were recorded as fact, leading to progressively worse placements as each new carer read the file before meeting them.

The Independent SW immediately identified how AI could disadvantage Black families: "who trains the AI around cultural nuance?" They gave examples of culturally respectful forms of address being misinterpreted as inaccurate names. Sam saw AI's potential to be "less biased than a human being" because it "seems more mathematical", but worried it could give higher risk scores to marginalised communities with more involvement with services. The Independent SW worried: "My understanding is that AI learns as it goes along and we're learning then from a flawed system."

Social workers emphasised they don't want to rely solely on others' interpretations in care records. The Independent SW explained: "I would still want to ascertain information myself (...) because what one person describes as aggressive, I might not"—emphasising that "things are relationship dependent."

AI presence may affect how people feel and open up

The Safeguarding SW worried they'd become "conscious about what I say, because I know I'm being recorded" and might focus on assessment needs rather than relationship building. They shared that not all conversations would be appropriate to record "verbatim", especially when people are visibly distraught or during first meetings.

Care-experienced participants worried about feeling processed rather than cared for. Jessie explained: "the moment you start kind of introducing things that are almost going to help the social worker do their job you kind of might introduce a feeling of 'Oh I'm just their job to my social worker, I'm not actually being cared about.'"

Information in care records can be overwhelming and disconnected

Social workers explained having to manually piece together chronologies across multiple systems. The Independent SW was blunt: "What more likely happens in a lot of cases is we don't search for information because we don't have time for that. So we base our judgments and recommendations on the limited information that we do have easy access to."

All participants saw potential in AI creating more complete pictures. Sam highlighted the lack of multi-agency working in their own care—police were involved from age 8, but social workers seemed to have "no clue" when they finally entered care at 15. As the Independent SW explained: "if I want to know a list of all the episodes where alcohol's been involved, and what was the impact of those episodes, and AI then gives me that back, that information was already in existence in the child's file anyway, it's just saved me from hours of trawling entries to find it."

AI can generate richer records—yet records are not the whole person

Jessie and Sam were clear: AI can generate richer case records with authentic children's views, but records are not the whole person. Jessie wanted AI to capture "everything" for complete understanding, but recognised it "is not going to understand like dreams, aspirations, goals (...) whereas a human will."

Even with diverse professionals making notes—police, medical, care—they're patchy fragments in relation to the child's involvement with services. "They don't know that child. None of them do", as Sam explained, "when you're in care, you're not really seen as a young person with passions and dreams. You're seen as a case."

All participants supported AI for admin work—automated transcripts, initial summaries, and documentation. However, there was strong consensus that humans are essential because situations constantly change and context matters. Jessie worried AIs would hallucinate around human topics, whilst "humans understand humans" and didn't want "computers assessing how at risk a child is" because that can be construed as fact. Focus group participants emphasised AI won't pick up on body language or tone and "doesn't know what's actually going on".

The fear: being reduced to an AI summary, like they've been reduced to notes

"I don't think children should be reduced to AI summaries. (...) I don't think even humans can really ever do that fully." [Sam]

One Cautiously Curious participant emphasised: "everyone here has different experiences of care... we all have different trauma. (...) without being care experienced yourself, it would be difficult to even design" an AI that captures this complexity.

Focus group participants agreed wholeheartedly that summaries constructed for specific purposes mean "what you lose in that process is more than what you would gain" [Cautiously Curious]—the humane sense-making is lost. Additionally, Sam worried Policy Buddy creating guides for young people, would be "impersonal", missing the translation between policy written "for adults" and real lived struggles. They contrasted this with a charity that hired care-experienced people to create resources: "[they] could write and create with the nuance of knowing what could go wrong."

What this means

Care-experienced participants and social workers shared remarkably similar concerns: both worry about accuracy, lost nuance, and AI summaries replacing genuine understanding. Young people specifically fear intensifying what they've already experienced—being reduced to incomplete, judgmental case notes.

These findings align with research describing case records as 'time-capsules' where children are primarily objects of inspection, only sometimes co-creators of their stories. My research reveals care-experienced young people are acutely aware of how they've been written about and its consequences.

The findings caution against viewing AI solely as a means to increase accuracy. Whilst exact transcripts capture what was said, they don't capture why, the context, or the non-verbal communication that shapes meaning. The Safeguarding SW's concern about becoming self-conscious reveals how recording technology may change the quality of conversation—a shift from relationship-building to assessment-gathering that young people already fear.

In summary making and text writing, AI makes decisions about what to include, interpret, and emphasise. Starting from AI-generated summaries risks limiting the exploratory thinking and holistic understanding participants identified as essential. Young people know the difference between being documented and being known.

What this means for AI product development and care practitioners

The challenge isn't to make AI transcribe perfectly or summarise efficiently. The challenge is to resist AI's inherent tendency toward reduction and instead design systems that expand practitioners' capacity to see children as whole people.

For AI designers and developers:

Build features that surface complexity, not just summaries — Instead of condensing to a single narrative, highlight contradictions and missing information. When children's views conflict with professional assessments, flag it. When wishes aren't recorded or strengths aren't documented, make that visible. The goal is to expand understanding, not provide quick answers.
Make bias visible and questionable — Flag emotionally charged language in care records in real-time. Prompt reflection: "This uses 'aggressive' or 'difficult'—is this observation or interpretation? What might this look like from the child's perspective?" Build in friction requiring practitioners to actively choose subjective language rather than defaulting to it. This matters because AI trained on historical records risks amplifying existing patterns of deficit-focused, adult-filtered documentation.
Enable verification, not just generation — Give young people access to AI-generated transcripts and summaries before they're finalised. Let them flag inaccuracies, add context, or dispute interpretations. This is about co-creating the narrative that will shape their care—not just producing cleaner documentation.

For children's social care practitioners:

Treat AI outputs as starting points for exploration — When AI surfaces information from records, use it to generate questions rather than conclusions. If AI flags repeated conflict in a child's history, ask: what's underlying this? What support might help? What is this child trying to tell us? Actively document strengths, interests, and what makes this child laugh—not just the crises that brought them into care.
Use AI to create space for relationship-building—and be aware of what changes — If AI handles administrative tasks, redirect that time to unhurried conversations and genuine presence. But be alert to how recording technology may change the quality of the conversation itself. The Safeguarding SW's concern about becoming "conscious about what I say, because I know I'm being recorded" matters. The goal is to be more present, not to perform for a transcript.
Question bias in documentation, including your own — When reviewing AI-generated summaries or drafting records, ask: am I describing behaviour or interpretation? What might this look like from the child's perspective? Resist the pull toward deficit-focused records that children will one day read and that will shape every future placement and professional relationship.
Explain AI use honestly — Tell young people when AI is being used in their meetings, how it works, and what they can do if uncomfortable. Transparency about technological processes is part of treating young people with dignity—not an administrative formality.

For local authorities and policymakers:

Measure quality of care, not just documentation efficiency — Don't assess AI success solely through time saved on recording. Ask: do young people feel heard? Do they see their wishes reflected in plans? Do they have stable relationships with practitioners who know them as whole people? The detailed exploration of these questions is taken up in AI within children's social care systemic constraints—from 'case' to care.
Build mechanisms for young people to access and shape their records — Create systems where children can review what is written about them, flag inaccuracies, and add their own perspectives. Give young people the same access to AI tools that social workers use to navigate their records and understand their rights. This isn't about creating extra processes—it's about treating young people as co-creators of their own stories, not objects of inspection. The full case for young people's direct AI access is made in AI within children's social care systemic constraints—from 'case' to care.

AI in children's social care must help practitioners see children as whole people with complex lives, evolving identities, and futures beyond case files. The technology should never reduce children to summaries—it should expand understanding of their full humanity. As Sam said: "I don't think children should be reduced to AI summaries. I don't think even humans can really ever do that fully." That's not a limitation to overcome—it's a reality to design around.

Continue exploring this series on AI in UK children's social care, what care-experienced young people actually want: AI as a facilitator of information in care—encouraging better support conversations | AI within children's social care systemic constraints—from 'case' to care

References & Further Reading

Major Reviews & Reports:

Munro, E. (2011). The Munro Review of Child Protection: Final Report - A child-centred system. Department for Education.
MacAlister, J. (2022). The independent review of children's social care -- Final report. The Independent Review of Children's Social Care.
Education Committee. (2025). Children's social care: Seventh Report of Session 2024–25. House of Commons.
UK Parliament. (2025, February 11). Children's social care / Education Committee [Video]. Parliament Live TV.

Research on Children's Participation & Documentation:

O'Keefe, R., Geddes, E., Vincent, S., & Davies, P. (2025). Enabling child-centred case recording in children's social work: The voice of practitioners. Child & Family Social Work.
Ogle, J., Vincent, S., & Hawkes, M. (2022). Authenticity, power and the case record: A textual analysis of the participation of children and young people in their child protection conference. Child & Family Social Work, 27(2), 278-286.
Roy, J., Staines, J., & Stone, B. (2025). Is it a positive or a negative? Children's participation in discharge of care order proceedings. European Journal of Social Work, 28(4), 743-756.
Race, T., & Frost, N. (2022). Hearing the voice of the child in safeguarding processes: Exploring different voices and competing narratives. Child Abuse Review, 31(6), e2779.

Research on Relationships & Trust:

Friel, C., Symonds, J., & Cartwright, M. (2025). Views and experiences of children and social workers about communication in social work exchanges: A mixed methods systematic review. Child & Family Social Work, 1-21.
Cossar, J., Brandon, M., & Jordan, P. (2011). Don't make assumptions: Children's and young people's views of the child protection system and messages for change. Office of the Children's Commissioner.
Juul, R., Husby, I., Kaalvik, H., & Salkauskiene, I. (2025). Characteristics of the qualitative research on inclusion of the youngest children in child welfare and protection work processes: A qualitative systematic review. Child & Family Social Work.
Diaz, C., Pert, H., & Thomas, N. (2018). 'Just another person in the room': Young people's views on their participation in Child in Care Reviews. Adoption & Fostering, 42(4), 369-382.
Dillon, J. (2021). 'Wishes and feelings': Misunderstandings and missed opportunities for participation in child protection proceedings. Child & Family Social Work, 26(4), 664-676.
Pösö, T. (2022). Mundane participation: Moments and possibilities within child welfare everyday life. Children and Youth Services Review, 135, 106395.
Mitchell, F. (2022). Care-experienced children and young people's perspectives on what helps with staying well. Adoption & Fostering, 46(1), 33-50.

Research on AI in Social Work:

AI Tools & Technical Resources:

Theoretical Frameworks:

Mayall, B. (2001). Introduction. In L. Alanen & B. Mayall (Eds.), Conceptualizing child-adult relations (pp. 1-10). RoutledgeFalmer.
Rosen, R. (2020). Childhood studies. In The sage encyclopedia of children and childhood studies (Vol. 4, pp. 344-350). SAGE Publications, Inc.
Taylor, C., & White, S. (2006). Knowledge and reasoning in social work: Educating for humane judgement. British Journal of Social Work, 36(6), 937-954.

On AI Limitations:

Üveges, I. (2025, May 27). Conditioned for servility -- Why do AI models flatter us at the expense of reality? Constitutional Discourse.

Composed with the help of AI, drawing on my dissertation in Sociology of Childhood and Children's Rights (UCL, 2025).