Featured image source: Alan Warburton / https://betterimagesofai.org / © BBC / https://creativecommons.org/licenses/by/4.0/
DR LUCY CATON, UNIVERSITY OF GREATER MANCHESTER, UK
ZAINAB PATEL, PRINCIPAL, OLIVE TREE PRIMARY, UK
CHRIS HESKETH, KEY STAGE 2 TEACHER, HASLAM PARK PRIMARY SCHOOL, UK
Aims and context
This article reports on preliminary findings from a small-scale empirical study and outlines the implications and emerging recommendations for practitioners aiming to integrate generative AI (GenAI) to support students with ASD (autism spectrum disorder). We examine how these technologies can develop greater student autonomy and agency as they goal-set, check progress and take action on improving their work. In doing so, the research foregrounds the evolving human-centred behaviours that emerge at the cross-section of knowledge co-construction with GenAI.
The study was situated in a context where there is an accelerating adoption of GenAI within educational practice (Sharples, 2023; Luckin, 2018; Baker and Caton, 2025). Phase one of the study was undertaken during October 2024 to May 2025 across several mainstream primary schools in the north-west of England. The research presented is drawn from a wider, multi-agency investigation, which also encompasses secondary schools and is led by Dr Lucy Caton, in collaboration with specialist practitioners from the Wood Bridge SEND [special educational needs and disabilities] Service. The aim of the initial phase was to investigate how GenAI might be implemented as a supplementary tool to support ASD students in developing agency and autonomy within the context of formative feedback. For clarity, we outline three stages of the formative feedback process: the learner understanding what they are trying to achieve (goal-setting), the learner understanding how they are doing (checking progress) and the learner understanding what to do next (taking action). The overarching research questions guiding the inquiry are:
- How can GenAI enhance ASD students’ agency and autonomy within their ongoing developmental feedback?
- How can AI-enhanced feedback be integrated in ways that uphold human-centred pedagogies, including important teacher–student relationships?
Research design
The study involved a small group of primary-aged ASD students, whom we engaged as co-researchers (Bradbury-Jones and Taylor, 2015). We report insights from three participants, Student A (aged 10), Student B (aged six) and Student C (aged eight), two of whom have education, health and care plans with additional classroom support. Students were selected based on their difficulties in interpreting body language and discerning tone of voice, challenges that often complicate interactions with teaching staff. None of the children had prior experience of GenAI tools, although all expressed curiosity and willingness to engage. Parental consent was obtained for all students. Initial informal conversations with the student and their teacher explored baseline knowledge of GenAI, preferred learning styles and feedback preferences.
The research team implemented qualitative data-collection methods, including classroom observations, teacher and child initial discussions and subsequent reflections. Conversations among the multi-agency research team were recorded during collaborative team sessions as part of the data-collection process, during which recommendations were also agreed. The initial discussions aimed to explore the children’s interests and baseline awareness of GenAI. In alignment with inclusive research practices and in recognition of the diverse communication preferences among the children, they were invited to participate using a range of multimodal formats, including drawings, written reflections and recorded voice notes (Lomax, 2018). Visual prompts accompanied text-based questions to support understanding and engagement, particularly among younger children and those with higher support needs. This multimodal strategy was adopted to reduce communication-related anxiety and cognitive load and to enable the expression of nuanced experiences (Bagnoli, 2009; Sweller, 2019).
Establishing familiarity between the wider research team and the individual educational setting was a key priority, particularly as the lead researcher did not work in the selected schools. Interim visits to each school were therefore undertaken to build rapport with staff and students and to accommodate individual needs, while also being sensitive to the demands placed on participants (Goodall, 2018).
Below, we present data excerpts from three teacher–student-GenAI interactions. The activities focused on generating feedback in support of free-writing tasks aligned with the English National Curriculum for Key Stages 1 and 2.
Findings and analysis
Student A (age 10): Reflection, precision and growing autonomy
Student A began the task with teacher mediation. The teacher activated the read-aloud function, demonstrated navigation of the AI interface, uploaded the student’s work and instructed the student to compare the AI-generated feedback with their handwritten biography of Thomas Edison. The student had both a physical copy of their work and the teacher’s written feedback for reference. In response to a tense error previously highlighted by the teacher, Student A prompted the AI to generate ideas for how the sentence could be corrected and subsequently revised the sentence accurately. They also identified strengths, initiated by GenAI, within their writing, such as correct use of past tense and appropriate paragraphing.
The teacher’s role as a coach was fundamental to the success of the activity, not only guiding the student in formulating effective prompt questions but also ensuring that they remained focused and on the task. Throughout the activity, the teacher encouraged metacognitive reflection (for example, ‘Does that make sense?’, ‘What should it say?’) and emphasised the importance of maintaining independence rather than relying on the AI to complete the work. The teacher further observed that this was the student’s first experience using GenAI and anticipated that, with increased familiarisation, the student’s prompt-management skills and the overall efficacy of such tasks would continue to develop.
The teacher later noted that, although the activity was effective, the level of individualised support required would be difficult to sustain at whole-class scale, making the approach more viable in small-group settings. Building on this reflection, the teacher proposed the development of ‘GenAI Polishing Pods’, aligned with existing classroom language around independent improvements to work. The pods would provide structured physical spaces within routine pedagogy for GenAI-supported revision activities.
Student B (age six): Navigating choice, speech barriers and emotional safety
Student B engaged with GenAI to obtain feedback on a writing task about the four British seasons. A physical version of the work with written feedback was provided as a reference. The activity began with audio prompts; however, after experiencing difficulty due to a speech impediment, Student B independently chose to switch to typed prompts, a decision supported and celebrated by his teaching assistant as an expression of agency. Notably, he referred to the chat agent as ‘the robot’, indicating an emerging awareness of the distinction between human and non-human feedback sources, which staff found reassuring. The teacher reported improved focus and engagement throughout the task. When asked whether he preferred feedback from the teacher or the ‘robot’, he ultimately responded ‘both’, signalling a developing understanding that AI could complement, but not replace, teacher expertise and established classroom relationships. The principal noted that these early insights would inform the integration of AI literacy into individual student targets.
Student C (age eight): Structure, predictability and subtle indicators of engagement
Student C’s first encounter with AI-generated feedback was marked by curiosity and visible cognitive engagement. Although extended verbal communication was challenging due to ASD-related communication needs, Student C demonstrated strong engagement throughout the task, evidenced by sustained focus, positive affect and a willingness to revise their work. The responsiveness appeared to be supported by the structured and predictable nature of the AI’s time-bound prompts (for example, ‘Last week you looked at…’, ‘Now I want you to focus on…’, ‘Next time, you should read about…’). The teacher emphasised the importance of AI literacy, warning against perceptions of AI as an unquestionable authority. They stressed the need to ensure that students do not ‘just listen to the machine’; instead, learning to evaluate, question and adapt AI outputs critically was something that they would begin to build into classroom pedagogies.
Further discussions
GenAI–mediated feedback was most effective when embedded within a dual-modality approach that combined digital feedback with print-based reflection. This structure supported self-regulation, reduced cognitive load and enhanced metacognitive engagement, aligning with Sweller’s (2019) cognitive load theory and supported by digital technology. Across all cases, students worked with both a physical copy of their written work and the digitally uploaded version, with the printed text providing a grounding reference that supported sustained attention, accuracy-checking and self-regulation. For learners susceptible to digital overstimulation or dependency, particularly some ASD students, this dual-modality approach mitigated risks while maintaining human feedback as a core element of the emerging teacher–student–AI triadic model (Sorensen, 2023).
The benefits were especially evident when students independently identified and validated revisions using their physical work and written teacher feedback, and then formulated prompts asking the GenAI for improvement suggestions. The low-risk, varied and creative options generated by the AI prompted greater ownership, engagement and autonomy. Nevertheless, it remained essential for teachers to emphasise that GenAI functioned as a support for thinking and revision, rather than as a substitute for the student’s own work.
A critical factor in the effective use of GenAI for eliciting feedback was the development of initial teacher-generated starter prompts, which were prepared in advance of each AI-mediated activity. These prompts integrated anonymised summaries of the students’ specific needs, which included their developmental stage, learning preferences and relevant National Curriculum writing objectives. While this tailoring was pedagogically valuable, it also highlighted the need for caution (Selwyn, 2022); early-stage implementation of GenAI should not become excessively resource- or time-intensive for educators, underscoring the importance of scalable and sustainable approaches.
Teacher A incorporated the school’s familiar visual feedback symbols – such as coloured stars for next steps, purple hearts for student-actioned responses and green ticks for success – into the AI prompt instructions and subsequent AI-generated feedback. Mirroring the semiotic system familiar within their exercise books, this transferring of familiar signs, symbols and language helped to maintain continuity, predictability and accessibility. This approach was especially beneficial for students who rely on visual reinforcement to interpret and act on feedback.
Conclusion
The wider multi-agency research team agreed that GenAI’s value lies not in replacing the teacher nor in automating learning, but in freeing cognitive and emotional space for students to explore, revise and reflect and better investigate their own metacognition, and this was pertinent at specific points in the day. We explore this theme further in our ongoing multi-agency research. For ASD students in particular, AI can offer predictability, immediacy and low-pressure feedback, while teachers provide the human insight, emotional availability and ethical oversight that no algorithm can replicate. The study continues to explore AI’s potential across Key Stages 2 and 3, and the priority remains clear: technology should enhance the richness of human learning, not diminish it.
The examples of AI use and specific tools in this article are for context only. They do not imply endorsement or recommendation of any particular tool or approach by the Department for Education or the Chartered College of Teaching and any views stated are those of the individual. Any use of AI also needs to be carefully planned, and what is appropriate in one setting may not be elsewhere. You should always follow the DfE’s Generative AI In Education policy position and product safety standards in addition to aligning any AI use with the DfE’s latest Keeping Children Safe in Education guidance. You can also find teacher and leader toolkits on gov.uk.
