The incredible capabilities of ChatGPT led to its rapid adoption after it was released to the public in November 2022. Its ability to provide coherent answers to questions, rewrite text in various styles, proofread and summarize content, and write code made it valuable across many contexts. The way it delivered well-formed answers to short prompts changed how people viewed AI's potential. However, many concerns arose during public discussions. These included worries about the accuracy and timeliness of the information it generates, its tendency to create false information, biases in its responses, and privacy issues. Allegations that copyright material was used to train models like ChatGPT without permission led to multiple lawsuits. Additionally, Perrigo (2023) highlighted how low-paid workers in some countries filtered content used for training ChatGPT. There were also concerns about the environmental impact and resource demands of AI models in general . As a result, ChatGPT and similar services like Google Gemini sparked both excitement and controversy. One significant area of debate was the role of ChatGPT in education. Many anticipated benefits of generative AI for learning included making information more accessible, personalizing education, and promoting critical thinking . Many uses for teachers were also identified . However, there were widespread fears about the potential negative effects of generative AI on education, especially regarding increased plagiarism . The difficulty in automatically detecting AI-generated text made plagiarism detection software ineffective. Furthermore, banning students from using these tools was seen as impractical. Many Universities in the UK, introduced policies, but providing clear guidance was challenging, and enforcing that guidance was even harder. Academics worried that generative AI might make students less critical or produce less authentic writing and possibly hinder their creativity . Broader concerns about generative AI, like biases, privacy, and equitable access, were also significant issues for its use in higher education . To use it safely, a certain level of generative AI literacy was necessary . Despite these challenges, some researchers believed that generative AI could be especially beneficial for students with disabilities , offering new types of support . This aligns with suggestions that generative AI could benefit people with disabilities overall. For instance, argued that generative AI creates new chances for organizations to provide tailored assistive technology solutions to employees with disabilities. found that the 'visual question answering' feature of generative AI is beneficial for blind users. However, they also noted the need to rethink the design and policies surrounding generative AI from this user's perspective to ensure equal benefits. In the learning context, there appears to be potential for generative AI to assist students with tasks such as proofreading, text summarization, and interactive support, which could greatly benefit students with learning difficulties, including dyslexia . UK schools/universities have begun offering generative AI chatbots as part of their learning management systems, including at the institution where data for this study was collected. At that institution, the Google Gemini chatbot was introduced as part of a Google-based learning platform. Given universities' legal responsibility to promote equality of opportunity and the continued under-representation of students with disabilities in UK higher education generative AI seems to hold great promise. Especially when paired with improved lecture transcription, generative AI represents a significant leap forward in technologies that support learning. However, generative AI tools were not designed with these user groups in mind, so studying their impact requires further investigation. Initial discussions about generative AI in higher education occurred with limited systematic evidence on how students were utilizing AI in various learning environments. Although research is starting to emerge regarding students' use of generative AI, a significant gap remains in studies specifically focused on how students with disabilities engage with generative AI, particularly in academic writing, which is crucial in higher education. Based on insights from previous research that developed a model of generative AI literacy for academic writing , this study aimed to explore how students with disabilities utilize generative AI.

The study aimed to explore the following research questions:

1) Which generative AI tools do students with disabilities utilize, and for what specific purposes?

2) What concerns or apprehensions do students with disabilities have regarding generative AI?

3) What types of support do students with disabilities require from their institutions to effectively use generative AI?

The early discussion about generative AI lacked a clear understanding of its actual usage. However, 18 months after ChatGPT's launch, a clearer picture is emerging about how students are adopting generative AI. We now have international surveys , national studies like those in the UK , along with institutional studies. Additionally, detailed qualitative studies by have provided valuable insights into how students from diverse backgrounds utilize generative AI in their learning. By capturing personal experiences, these studies give a more genuine and clear view of generative AI’s role in higher education.

Overall, students tend to view AI tools positively. A significant number of students, possibly even a majority, have used generative AI in their learning, as shown in research. This research summarizes students’ attitudes and behaviors in the UK, which is the focus of this study. data, from February 2024 revealed that two-thirds of UK students had used some form of AI, and a third had applied it for assessments. However, a digital divide is becoming apparent, with students from less privileged areas using AI less frequently. According to , ChatGPT is the most commonly used service. Many students use it to gain an initial understanding of a concept, though a broad range of potential uses has been identified (JISC, 2024).

Interestingly, found that most UK students believed there was a clear institutional policy on AI. Students expressed a need for better training from institutions and for more tools to be provided . Only 22% of respondents were satisfied with the support they had received . Other qualitative studies have provided a deeper insight into students’ usage patterns. For instance, , conducted interviews and observations with UK students, finding that they utilized generative AI in various ways to support different stages of writing. However, none of these surveys or interview-based studies systematically investigated students with disabilities and their use of generative AI. Malmström, mentioned that open-text comments offered anecdotal evidence of AI’s value for students with learning disabilities. This study differs from previous research in its primary focus on students with disabilities, aiming to address the gap in the literature regarding the inclusion of generative AI in higher education.

Limited research has been done on the use of generative AI in education for students with disabilities. Some researchers see its potential benefits . They have pointed out specific tasks where it might help, such as:

1) Reading: Generative AI can support students with disabilities—particularly those with dyslexia—by enhancing their ability to understand learning materials . Students with cognitive challenges often experience difficulties in processing written text, leading to slower reading speeds and a higher likelihood of errors . Generative AI tools can address these challenges by summarizing lengthy and complex texts into clear and concise formats, thereby improving comprehension . This support helps students focus on core concepts and increases their engagement with reading tasks .

2) Writing: Generative AI can assist students with disabilities—particularly those with dysgraphia—by supporting the planning and organization of writing tasks and by helping refine written output . Research indicates that these students often face challenges in goal setting, organizing ideas, developing coherence, and revising for clarity and grammatical accuracy . Generative AI can address these difficulties by clarifying writing instructions, supporting goal setting, and generating structured writing plans . Additionally, it can help improve coherence and grammatical accuracy, resulting in more transparent and more precise written work .

3) Social Engagement: Generative AI may support the development of social skills among students with disabilities ). Students with learning disabilities often face challenges in expressing themselves and interpreting others’ verbal and nonverbal cues, which can contribute to social isolation . Generative AI tools can simulate real-life communication scenarios, allowing students to practice conversational skills and improve their understanding of social interactions . This practice can enhance peer collaboration and promote greater engagement within the learning environment.

4) Teaching: Generative AI can enhance the quality of education, thereby indirectly benefiting students with disabilities. For example, it can be used to evaluate the accessibility of instructional content and may be trained to identify early indicators of learning disabilities . Such capabilities can support educators in designing personalized instructional strategies that address the diverse needs of students with disabilities . By offering both direct and indirect support, generative AI contributes to a more inclusive and supportive educational environment , improving access to educational resources and potentially enhancing learning outcomes for students with disabilities .

However, views on the impact of generative AI on students with disabilities are not entirely positive. First, students with disabilities might find it hard to use it effectively. For example, visually impaired students often struggle with text-based chatbots . Furthermore, generative AI is often trained on datasets that don’t fully consider the needs of marginalized users. This can lead to generic functions that do not meet the specific requirements of individuals with disabilities . In some cases, the outputs from generative AI can overwhelm those with learning disabilities, increasing their stress levels . Early versions of ChatGPT were limited to text-based interactions, which may have caused additional stress for students with dyslexia when trying to interpret the text outputs . AI-based facial recognition systems often lack enough training data on students with autism, making them prone to misreading emotions and failing to offer tailored support found that 13 language models associated disability-related terms with negative meanings, creating bias and possibly reducing engagement with these technologies among users with disabilities .

To understand the impact of generative AI on students with disabilities, it is essential to center their lived experiences and perspectives. However, research on how these students use generative AI remains limited and produces mixed findings. For example, reports potential adverse effects for students with ADHD, such as reduced attention due to oversimplified information processing. In contrast, highlight benefits for neurodivergent students, including improved focus on core concepts and structured learning. These conflicting findings may stem from the limited inclusion of students with disabilities as primary participants. Unlike prior studies, which often examine general student populations, this study focuses directly on students with disabilities to better understand their interactions with generative AI.

The digital revolution has led researchers to identify various literacies essential for navigating the modern world. These include computer literacy, information literacy, media literacy, and digital literacy. The rising use of AI has sparked efforts to define AI literacy. For instance, offered a key definition of AI literacy that includes 17 elements, along with guidelines for creating explainable AI. However, generative AI has altered how we perceive AI’s capabilities and its associated ethical dilemmas. In this regard, , introduced a specific model of generative AI literacy comprising five elements.

1) Pra understanding: The individual can effectively use generative AI and critically assess the information it generates.

a) The individual can choose the right tool for the task among various options, including different generative AIs like ChatGPT.

b) The individual learns to use the selected tool effectively for specific tasks, such as “prompt engineering.”

c) The individual critically interprets generative AI outputs, understanding how they function and their limitations, like accuracy, timeliness, citation, and bias.

2) Safety understanding: The individual can use generative AI safely, being aware of privacy risks.

3) Introspective comprehension: The individual can evaluate and take steps to manage the effects of AI on their experiences, including its influence on their learning and dependence on technology.

4) Socio-ethical understanding: The individual understands the societal effects of AI, including issues related to intellectual property rights, its impact on information culture, misinformation, social problems stemming from exploitative creation processes, job impacts, access equity, environmental concerns, and the influence of major tech companies.

5) Contextual understanding: The individual knows how to use generative AI appropriately in specific contexts and can make their usage clear when necessary.

This definition combines the often-discussed aspect of “prompt engineering” with the need to select and learn from a growing array of AI services . It emphasizes the importance of safety and ethics in the use of AI. Additionally, it emphasizes the importance of reflecting on how AI affects personal experiences, such as dependency or the loss of individual expression. While these dimensions capture essential components of generative AI literacy, we do not suggest simplistically measuring generative AI literacy. For example, we cannot claim that all societal effects of AI should be a primary concern, as opinions on the significance of these issues vary. Therefore, we present this definition as a broad framework to explore an emerging literacy.

Due to limited research on the use of Generative AI by students with disabilities, we used an inductive approach to explore their experiences and views through an online survey. The survey included three main sections.

The first section collected demographic information about the respondents, including their gender, level of study, academic department, perceived English language skills, perceived digital skills, and standard generative AI tools they used.

The second section had questions related to disabilities. It examined how disabilities impact academic writing and the tools used to overcome these challenges. Participants received a list of disability conditions based on terminology from an authoritative higher education source, the Higher Education Statistics Agency (HESA) (2024). Experts from the Disability and Dyslexia Support Services at the university refined these categories. Participants could choose not to disclose any disability conditions. Some participants could report more than one condition. Consequently, the total number of self-reported disability conditions was higher than the number of participants who identified as having a disability.

The third section asked questions about the general use of generative AI. This covered its application at different stages of academic writing, costs, student perceptions of generative AI, current concerns, and recommendations for universities. Open-text questions explored how reported disabilities affect students' academic writing, which specific generative AI students used to overcome barriers related to their disabilities, and their suggestions for improving university support for students with disabilities. In addition to key questions about the use of generative AI, the survey aimed to provide information related to the model of generative AI literacy discussed earlier. The complete survey design, including all questions and their order, is presented in the appendix of this paper.

The survey was sent to all students at various therapy centers across the NCR in February and March 2025, targeting a different therapy center with 5,000 registered students with disabilities. The survey received 100 valid responses from students with disabilities. The numerical data were presented descriptively, while qualitative data were analyzed using content analysis, with codes created based on the data.

Respondents were asked to describe how their disabilities affected their writing. Of the 100 respondents, 18 did not respond, and 2 reported no impact, leaving 80 valid responses. The main barriers identified are summarized in Table 1.

Note. This was an open-ended question; respondents could report multiple barriers.

The findings highlight a wide range of challenges closely associated with participants’ reported disabilities. Responses frequently conveyed emotional strain, with terms such as “struggle” and “feeling overwhelmed” appearing repeatedly. Participants described significant difficulties with writing speed, clarity, tone, and organization, emphasizing fundamental differences in cognitive and communication processes compared to peers without disabilities.

Students’ motivations for using generative AI were measured using a 5-point Likert scale (1 = strongly disagree to 5 = strongly agree). The primary driver was perceived relevance to future careers, followed by improved efficiency. Fewer students reported improvements in academic performance, and even fewer indicated increased engagement with learning.

Of the 100 respondents, 77% (n = 96) reported using generative AI, while 22% (n = 28) used it infrequently or not at all. Most respondents (91%) did not pay for AI subscriptions.

Chatbots were the most commonly used tools (66%, n = 82), followed by rewriting tools (15%, n = 18) and translation software (8%, n = 10). The most frequently reported uses were summarising reading materials, overcoming mental blocks, brainstorming ideas, rewriting assignment text, and structuring ideas. Among chatbots, ChatGPT was the most popular (52%, n = 65), followed by Gemini (8%, n = 10), Copilot (4%, n = 5), and Claude (2%, n = 2).

Respondents reported using ChatGPT across five key stages of the learning process: explaining topics, identifying resources, summarising materials, structuring writing, and refining written work.

ChatGPT was valued for explaining complex topics (15%, n = 18) by simplifying academic language and providing on-demand support. It was also used to identify learning resources (7%, n = 9), particularly by helping formulate effective search terms and reducing cognitive fatigue during information searches.

Participants frequently used ChatGPT to summarise learning materials (14%, n = 17), especially lengthy readings and lecture notes, supporting concentration and comprehension. Support for structuring writing (18%, n = 22) was highlighted as reducing anxiety and providing a clear starting point for essays. Finally, ChatGPT was commonly used for refining writing (19%, n = 24), including improving coherence, clarity, and academic tone.

Overall, respondents perceived ChatGPT as a flexible and accessible tool that supports both learning and writing processes, particularly for students with disabilities.

In addition to ChatGPT, respondents reported using Gemini (8%, n = 10), Copilot (4%, n = 5), and Claude (2%, n = 2). Their use mirrored that of ChatGPT due to similar core functionalities, although overall usage was lower. Notably, Gemini was less frequently used despite being the institutionally approved tool.

Compared with ChatGPT, these chatbots were perceived as less comprehensive: Gemini was seen as limited in identifying learning resources, Copilot in refining written work, and Claude primarily in summarising materials and refining text. Nevertheless, respondents considered these tools reliable, particularly because of their non-intrusive support that preserved students’ thinking and writing processes, as well as their ease of access (e.g., Copilot’s integration into the Bing browser).

Their lower popularity was mainly attributed to limited familiarity, with students preferring ChatGPT as the tool they were first introduced to and had learned to use confidently.

After chatbots, rewriting applications were the second most commonly used AI tools, including Grammarly (10%, n = 13), Quillbot (2%, n = 3), and Wordtune (1%, n = 1). Students used these tools primarily to correct grammatical errors (8%, n = 10), improve sentence coherence (2%, n = 3), and refine word choice and tone (2%, n = 3). These services helped students improve the overall quality of their writing.

ChatGPT was often used alongside rewriting tools, particularly Grammarly (n = 4), with students relying on Grammarly for proofreading and ChatGPT for rewording and clarifying meaning.

Translation software was the third most used category, including Google Translate (4%, n = 5), DeepL (3%, n = 4), and other tools (1%, n = 1). Students primarily used these tools for foreign language learning, understanding non-English texts, and refining written work.

The survey explored students with disabilities’ views on key controversies surrounding generative AI. The most frequently reported concerns were the inaccuracy of AI-generated outputs and the risk of unfair academic practices. Respondents also expressed concern about inequality arising from paid AI services, while concerns about broader societal impacts, such as effects on employment, were moderate. In contrast, concerns related to exploitative development practices, environmental impact, and loss of personal writing voice were minimal.

Among the 28 students who reported not using generative AI, concerns largely mirrored those of AI users, focusing on accuracy and academic integrity. A small number also cited limited expertise as a barrier to adoption.

Survey results indicate that students believe they should be permitted to use generative AI ethically and express a strong desire to be involved in policy-making. There was broad agreement that universities/Schools/therapy centers should provide training on generative AI use.

Responses to an open-ended question aligned with closed-item results, with most students requesting training on effective use of generative AI (33%, n = 41). The primary focus was on ethical and responsible use, including avoiding unfair means (22%, n = 27) and unethical practices more broadly (7%, n = 9).

A smaller proportion of students requested practical skills training, such as writing prompts (11%, n = 14), search, fact-checking, and referencing (10%, n = 12), text summarization (5%, n = 6), and exposure to a wider range of tools (3%, n = 4). Few respondents mentioned support beyond writing, such as time planning (2%, n = 3), and only a small number requested disability-specific guidance (5%, n = 6), particularly on AI tools suitable for students with learning difficulties, ADHD, or autism.

This study is among the first to collect real data on students with disabilities and their experiences using generative AI, like ChatGPT, rather than relying on general surveys or speculation about its benefits for these students. Our participants mainly included those with neurodiversity, such as ADHD, specific learning difficulties, and social or communication impairments like autism. These students remain underrepresented and need help to overcome barriers to participation and achievement . Respondents reported facing various academic writing challenges, often accompanied by anxiety and difficulties related to their disabilities. If generative AI can help break down these barriers, it could significantly enhance inclusion. However, we must be cautious about the potential loss of learning opportunities that could come from relying too heavily on generative AI. The students with disabilities who participated mainly used generative AI, particularly ChatGPT. They also used some grammar-checking applications and translation software. There was little evidence of broader use of the many generative AI tools available . Interestingly, few students were paying for a subscription. ChatGPT’s popularity seemed linked to students' familiarity with it and its ability to assist with many tasks, while other tools only supported a narrower range of activities. Generative AI chatbots were used throughout the study and academic writing process for:

1) Explaining complex topics

2) Improving search terms and conducting searches

3) Summarizing texts and other learning materials

4) Structuring ideas for writing

5) Refining text by combining ideas and improving word choice

Generative AI helps reduce anxiety around these complex tasks, which students with disabilities often find emotionally challenging. Some of this aligns with earlier speculation about generative AI’s potential to assist with reading, writing, and proofreading. However, this study offers a clearer understanding of how disabilities affect students and how they utilize generative AI to address challenges in written assignments. For instance, we gathered specific evidence on how students with disabilities use generative AI tools to find information and access learning resources.

Overall, survey responses indicated that students with disabilities view generative AI as effective assistive technology. Notably, the most helpful support often came from generative AI not recommended by their university. For example, Gemini, the approved chatbot at the institution where we collected data, was used relatively little. This gap between students' preferences and the university's recommendations raised concerns. Respondents worried that using their preferred generative AI, even if unapproved, could lead to issues around unfair means. As a result, some students hesitated or refused to use it, missing out on its benefits. This disengagement could impact fairness in higher education , particularly for students with disabilities who are already marginalized . To tackle this issue and ensure that all students, especially those with disabilities, can succeed in a more inclusive higher education setting, we need training in AI literacy .

Returning to our definition of AI literacy, we can identify both strengths and weaknesses among students with disabilities overall. The variety of uses they made to tackle writing challenges shows creative engagement with the many possibilities provided by generative AI services. Responses indicated appropriate concerns about information accuracy and academic integrity. Respondents appeared to be aware of some socio-ethical and safety issues, emphasizing those that directly affected them rather than broader societal concerns. They also reflected on the potential effects of generative AI on their learning, such as dependency. They recognized the importance of contextual relevance. We can also spot significant gaps in students' generative AI literacy from the data. They primarily used ChatGPT, seemingly without exploring alternatives. Raising awareness among students about the societal and environmental impacts of generative AI seems essential. Further research is needed to explore other aspects of their literacy, such as how effectively they create prompts.

This study is one of the first to investigate how students with disabilities utilize generative AI. Higher education has viewed generative AI as a potential threat; however, given the persistent underrepresentation of students with disabilities in universities, if generative AI can support them, it offers significant benefits. The evidence collected reveals a diverse range of uses during the writing process. Student concerns centered on the accuracy of information and its impact on academic integrity. Some worried that access to generative AI could create digital inequality, while concerns about its societal effects were less prominent. Students expressed a desire for training in using generative AI and for involvement in policy-making.

The research provides crucial findings to guide disability support units and all individuals who assist students with disabilities in their studies. Given the groups who responded, it seems appropriate to enhance support for students with ADHD, specific learning difficulties like dyslexia and dyspraxia, and social or communication impairments like autism spectrum conditions. Regarding policy, there appears to be ongoing uncertainty about the acceptable and unacceptable uses of generative AI, which hinders its widespread adoption. A policy statement clarifying allowed uses for those with disabilities could be helpful. However, the diversity of disciplinary approaches to learning and assessment makes it challenging to create general guidelines. Many students with disabilities expressed a desire for their voices to be heard in policy-making. Respondents requested training that could highlight the uses they focused on. Given their usefulness and cost, providing subscriptions for students to generative AI services could also be considered.

While our study sheds light on how students with disabilities use AI and offers practical guidance for educators and policymakers, several limitations must be acknowledged. We should consider the potential impact of non-response bias on the results, as only those who are using generative AI or opposing its use may have responded. This may lead to results that do not capture all opinions. Without further investigation, we cannot establish this. The survey remains useful as it reveals beneficial uses identified by some students. The sample size was relatively small, with only 100 out of 5000 eligible students (about 1.4%) responding. Although we explored a range of disability conditions informed by existing HESA categories, the small sample size made it hard to differentiate between the experiences of students with different disabilities. Similarly, it was challenging to identify varying experiences of students from diverse backgrounds (e.g., international vs. local; undergraduate vs. postgraduate; across different disciplines). Nonetheless, as an exploratory study focusing on this under-researched group, our findings provide valuable insights for future research. Future studies could build on these findings by investigating specific disabilities or making comparisons to understand better their impact on students' writing, which would help provide more tailored support. Upcoming research could also explore how students' backgrounds may influence their use of generative AI. The study focused on just one institution in the UK, and existing institutional support might have influenced students' views. There was also a disproportionate number of female students among respondents, although this reflects the gender distribution within the university's overall student population. Future studies could aim for a larger sample size with a more balanced gender distribution and gather data from multiple institutions.

I gratefully acknowledge my daughter, Saanvi, whose journey with autism has been a profound source of insight and inspiration. Her experiences have significantly influenced the perspective and motivation behind this research.

Vinita Dutt: Conceptualization, Data curation, Formal Analysis, Funding acquisition, Investigation, Methodology, Project administration, Resources, Software, Supervision, Validation, Visualization, Writing – original draft, Writing – review & editing

The author declares no conflicts of interest.

Which Generative AI do you use (1 - Never 2 - Rarely 3 - Sometimes 4 - Often 5 – Always)

1) ChatGPT

2) Google Gemini (formerly Bard)

3) Wordtune

4) Quillbot

5) GrammarlyGo

6) Jasper

7) Bing Chat

8) Google translate

9) Claude from Anthropic

10) DeepL

11) Otter AI

12) Elicit

13) Consensus

1) Specific learning difficulty such as dyslexia, dyspraxia

2) Neurodiversity - e.g. ADHD

3) A social/communication impairment such as Asperger's syndrome/other autistic spectrum disorder

4) A long standing illness or health condition such as cancer, HIV, diabetes, chronic heart disease, or epilepsy

5) A mental health condition, such as depression, schizophrenia or anxiety disorder

6) A physical impairment or mobility issues, such as difficulty using arms or using a wheelchair or crutches

7) Deaf or a serious hearing impairment

8) Blind or a serious visual impairment uncorrected by glasses

9) A disability, impairment or medical condition that is not listed

10) Prefer not to answer

Sometimes 4 Often 5 Aways)

1) Interpreting an assignment brief

2) Finding information about a topic

3) Summarizing reading material

4) Translation of reading material

5) Brainstorming ideas

6) Validating a draft against assessment criteria

7) Overcoming a mental block

8) Structuring ideas for my assignment

9) Rewriting phrases in my assignment

10) Proof reading of my assignment

1) Generative AI is enabling some students to gain an unfair advantage in their studies

2) Using AI has improved my working efficiency

3) Using AI has improved my academic performance

4) Using AI has made me more engaged in learning

5) Learning to use generative AI effectively is important for my future career

6) The University should offer training in how to use generative AI

7) The University should ban generative AI

8) Students should be involved in determining the School/College/Therapy center AI policy

What, if any, are your concerns about generative AI (1 Not at all concerned, 2. Slightly concerned 3 Moderately concerned 4 Very concerned 5 Extremely concerned)

1) It can be used in ways that breach academic integrity (e.g., plagiarism)

2) It can produce inaccurate answers

3) It generates prejudiced and discriminatory content

4) It creates text that does not sound like me

5) It breaches personal privacy (e.g., compromising personal information)

6) It fails to explain how the AI was trained and how it works

7) It could lead to humans losing their jobs

8) It can be used for bad purposes such as deepfakes

9) It creates unfairness if some people pay for access to better services

10) It exploits low paid workers in the creation of the service

11) It has negative environmental impacts (e.g., CO₂ emissions)