Disability Expertise and Large Language Models: A Qualitative Study of Autistic TikTok Creators’ Use of ChatGPT

Abstract

The purpose of this study was to describe how autistic TikTok creators are using ChatGPT across various domains of their lives, their motivations for doing so, and resulting impacts. Using a framework of “disability expertise,” we document the knowledge that creators acquired through use of ChatGPT and then shared with peers via social media. We used deductive qualitative methods to analyze 25 TikTok videos from 25 unique creators. Themes were identified in connection with motivations for the use of ChatGPT, settings in which it was used, applications of this technology, and resulting impacts for creators. Findings indicate that autistic creators were motivated to use ChatGPT to navigate neurotypical environments, manage features of their neurodivergence, and unmask, with the technology often serving as a digital coach, communication assistant, and conversational partner. Use of ChatGPT resulted in harm reduction, time and energy savings, positive emotional experiences, and meaningful accomplishments in both personal and professional settings. These findings indicate that ChatGPT serves as an important resource for many autistic individuals, facilitating accommodations to often inaccessible environments and helping users manage stressors and pursue goals. The study also highlights the significance of social media platforms for disseminating disability expertise related to the use of large language models to improve quality of life.

Keywords

disability expertise autism accessibility generative artificial intelligence neurodivergence

Introduction

Neurodivergent and/or disabled individuals often develop strategies for navigating inaccessible environments. Some accessibility problems may emerge from the built environment, like facilities that lack wheelchair ramps or indoors spaces that are not sensory-friendly. Other accessibility problems come from the social environment and reflect difficulties related to activities of daily living, communication with others, interpersonal ableism, and the predominance of discrete modes of information exchange.

In this study, we surface solutions to accessibility problems within neurodivergent and disabled communities. We recognize accessibility strategies that are developed and then openly shared by neurodivergent and disabled people as the specialized knowledge and skills that form “disability expertise” (Hartblay, 2020). In contrast to the dominant medical model of disability, which defines disability through medical science and establishes disabled persons as targets for expert intervention, disability expertise underscores the resourcefulness of disabled individuals and groups as they navigate social settings that are misaligned with their specific abilities and preferences (Dokumaci, 2023; Hartblay, 2020). We apply this frame here to extend an understanding of social media as a tool for self-organizing support communities (MacKinnon et al., 2021; Zehrung & Chen, 2024). To identify such expertise, we examine how ChatGPT, as a new technology with broad and expanding applications, is assisting some autistic and neurodivergent individuals in their daily lives. We do this by analyzing a sample of TikTok videos created by self-identified neurodivergent individuals who are engaging with ChatGPT and sharing their experiences with others.

Neurodivergence is a broad term that covers shifting terrain. In popular usage, this non-medical term often refers to a range of cognition-related differences that have referred traditionally to psychiatric categories, which include (but are not limited to) autism spectrum disorder, attention-deficit/hyperactivity disorder (ADHD), dyslexia, obsessive compulsive disorder, post-traumatic stress disorder, major depressive disorder, and bipolar disorder. Use of the term neurodivergence often signals recognition and acceptance of these cognitive traits as part of a broad range of human experience, which can be contrasted with more deficit-based definitions that promote clinical intervention. Thus, while autism has been defined medically as a complex neurocognitive disorder characterized by differences in communication and sensory processing (American Psychiatric Association, 2013), when framed as neurodivergence, it reflects a self-attributed social identity that encompasses characteristics, strengths, and challenges related to information processing and communication (Cooper et al., 2023, 2020; Maitland et al., 2021). We use neurodivergence/neurodivergent here as an umbrella term that includes the various labels the TikTok creators in this study used to define themselves.

Autistic individuals represent a diversity of cognitive attributes and approaches to language. Despite some variation, studies suggest many autistic people dislike communication over the phone and prefer written communication (Howard & Sedgewick, 2021). Because social expectations around language use can present barriers for autistic people, the advent of generative artificial intelligence (AI) and widespread access to large language models (LLMs) mark a shift in available communication technologies for this group. LLMs process natural language inputs or “prompts” from users and then generate human-like text outputs through probabilistic statistics trained on vast data sets. These models can be used for a variety of purposes including drafting emails, generating code, gathering recommendations, enhancing creative writing, and chatting. This text-based interactive format may align LLMs with the communication preferences and abilities of autistic users (Cho et al., 2023). Indeed, new LLM products such as Goblin Tools have been designed specifically for a neurodivergent user base with features such as generating to-do lists, interpreting the tone of written communication, and estimating how long a task will take to complete. Given their functionality, LLMs might rightly be conceptualized as an assistive technology akin to existing tools like scripted speech (Charlop-Christy & Kelso, 2003). A key distinction with LLMs, however, is the integration of artificial intelligence, which allows for contextualized and customized support, a feature that prior research has found to be a promising avenue for enhancing the utility of assistive technologies (Iannone & Giansanti, 2024).

Current Study

This study describes emergent disability expertise among autistic people regarding ChatGPT. We focus our inquiry on a set of videos shared by autistic TikTok creators. We view this social media platform as a suitable space for obtaining data, where users can disseminate their experiences with new technology or products while informing others of potential harms and encouraging peers to adopt tools and practices that they find to be helpful (e.g., Lundy, 2023). Prior research has found TikTok videos to be a useful source of data for understanding the individual experiences of autistic people (Gilmore et al., 2023). We employ a deductive qualitative approach to analyze TikTok videos in which autistic content creators report on their use of ChatGPT, and answer the following questions: (1) What motivates those with autism to use the technology?; (2) In what life domains is the technology deployed?; (3) What is the range of applications for which ChatGPT has been used?; and (4) What is the perceived impact of its use?

Method

Study Design

This qualitative study employed a thematic analysis to examine the reports of autistic TikTok creators’ experiences with ChatGPT. Deductive analysis was guided by the categories of motivation, setting, application, and impact, from which broader themes were determined by the research team.

Theoretical Framework

Our analysis was guided by theoretical frameworks of critical access studies and disability expertise, which emphasize recognizing marginalized experts and redefining data to offer new narratives about disability and public belonging (Hamraie, 2018). Disability expertise highlights situated ways of knowing and centers the embodied knowledge of marginalized groups as they resist normative demands (Campbell, 2009; Haraway, 1988; Hartblay, 2020; Moraga & Anzaldúa, 1981; Oliver, 1990; Siebers, 2008). Expertise is expressed when autistic people use LLMs to avoid forms of profiling that would further transform them into “targets of surveillance, measurement, management, remediation, expulsion, and extermination” (Rosa & Díaz, 2020, p. 123). Solidarity is formed when these skillful agents share their discoveries with other autistic people who can benefit from new strategies for concealment, getting by, and success. These frameworks emphasize “technoscience for political action” and refuse “to comply with demands to cure, fix, or eliminate disability” (Hamraie & Fritsch, 2019). Furthermore, our commitment to viewing research subjects as experts in themselves is informed by anthropological commitments to “taking seriously” their perspectives and experiences (Archambault, 2016; de la Cadena, 2010), recognizing skillful deployments of ChatGPT by autistic people as acts of both self-accommodation and camouflage from institutionalized perceptions of disability (Hansson, 2005).

Sample

The study involved identifying and then analyzing videos created by autistic TikTok creators, with a focus on content related to autistic experiences with ChatGPT. We chose TikTok for its popularity and novelty, recognizing that other social media platforms may contain autistic content creators speaking about this topic in similar and different ways. Using hashtags to search for and select relevant content is a standard strategy in research across social media platforms (Herrick et al., 2020; Hung et al., 2023; Minadeo & Pope, 2022), including identification of content related to autism (Alper et al., 2023; Egner, 2022). Our search terms included #autism, #autistic, #auDHD, #ASD, or #ActuallyAutistic in tandem with either #ChatGPT or #OpenAI. Our initial sample included 33 TikTok videos created by 32 unique creators. The decision to include videos in our final sample was based on the relevance of a video to our thematic focus on large language models and autism. Videos produced by creators who did not speak directly about their use of ChatGPT or who discussed ChatGPT use independent of their neurodivergence were excluded (n = 8). Our final sample consisted of 25 videos created by 25 unique creators who identified as autistic and/or autistic and ADHD. The posting date for these videos ranged from December 2022 through September 2023.

Analytic Approach

The first, second, and final author conducted the data analysis. Videos were analyzed holistically, including observations and notes on the video’s audio/visual elements and body language of each creator to identify themes within the deductive categories of motivation for ChatGPT use, setting, application, and impact/outcome. These a priori deductive categories were selected to understand each of the narrative elements in the creators’ videos. As highlighted in our findings, this holistic approach aided in identifying codes such as emotional impact which required reliance on indicators other than speech. Videos were also transcribed verbatim, and transcriptions were placed into worksheets designed to identify relevant excerpts of text following coding.

To support a fuller understanding, analysis began with watching and listening to each creator’s video before making notes and applying codes to the prepared transcripts. Videos were then watched together as a research team before reviewing independent codes and discussing observations. Regular meetings facilitated a collaborative process as researchers shared interpretations and worked toward agreement on themes (Padgett, 2012). Detailed notes following each session documented progress and captured justifications made to remove a limited number of videos from the study in which TikTok creators did not directly speak about their use of ChatGPT (Stahl & King, 2020). Application of codes with ease during later videos suggested code saturation to these researchers (Hennink et al., 2017). Analysis ended with a team review of the code list to identify areas for consolidation or separation (Saldana, 2013).

Findings

This study’s four research questions formed the basis for analysis of autistic TikTok creators’ interactions with ChatGPT. Insights were identified for each area of interest: motivation, setting, application, and impact. Themes within each of these categories are presented in Table 1 along with excerpts of supporting speech to demonstrate the validity of our findings. Creators’ motivations for using ChatGPT and the impacts that their interactions with this technology produced are described below with examples from the transcripts.

Table 1.

List of Identified Themes by Category of Interest.

Category	Identified themes
Motivation	Navigating neurotypical worlds Managing features of neurodivergence Unmasking
Setting	Professional Personal
Application	Generating content Gathering insights Coaching Conversation
Impact	Harm reduction Savings Positive emotions Meaningful accomplishments

Motivation

TikTok creators’ motivations for using ChatGPT were grouped into three themes: navigating neurotypical worlds, managing features of neurodivergence, and unmasking. Through this process, we noted an alignment between individuals’ motivations for engaging with ChatGPT and their apparent self-conceptions. Some TikTok creators, speaking from what appeared to be a deficit-based model of disability, asked ChatGPT for help with personal characteristics or habits that they perceived as limiting. Other creators’ comments aligned more with social and critical models of disability, and they asked for ChatGPT’s input on navigating contexts that did not accommodate neurodivergent traits and preferences. Our focus on identifying examples of disability expertise gave priority to balancing the tensions between each framework.

Navigating Neurotypical Worlds

TikTok creators consistently reported using ChatGPT to help navigate inaccessible social and built environments ill-suited to traits and characteristics common among autistic people. This assistance was based on a range of needs that included help with securing and maintaining employment, improving productivity, and translating between personal and dominant modes of communication.

In preparation for an upcoming flight, one creator looked to ChatGPT for help thinking through accommodations that would minimize her travel anxiety. She reported,

I hate having a panic attack in busy places like that because it’s really, people really, like, give me a lot of attention about it and I don’t want attention; I just want to hide away. So, I asked ChatGPT what accommodations, as an autistic person, I could ask for and the response gave me a load of accommodations that I could . . . I didn’t even think I could ask for. (Creator #15)

This account highlighted the usefulness of ChatGPT for managing the demands of settings and situations that were misaligned with personal attributes.

TikTok creators were also motivated to use ChatGPT to help them communicate across contexts where misunderstandings were expected. For some, this kind of assistance meant that ChatGPT broke down confusing emails to identify specific tasks being requested. For others, this meant asking ChatGPT to make the tone of an email sound more or less professional. As one person explained, “I’m always too afraid to ask because I don’t want people to think that I’m stupid. But also, it’s frustrating because if people don’t tell me directly what they want from me, then how am I supposed to know?” (Creator #19). Another TikTok creator (#6) stated, “I just don’t know how to formulate that in the way that, um, neurotypical people need it to be written.” Autistic individuals asked ChatGPT for help bridging communication divides and entering hard-to-access spaces, enhancing their ability to navigate environments and marking a significant shift in their interactions with others.

Managing Features of Neurodivergence

TikTok creators also sought out ChatGPT to help them prevent personal qualities, which they associated with being autistic, from negatively affecting their lives and relations. Unlike some creators who situated disability socially and framed problems as the result of systems that do not accommodate autistic differences, this set of TikTok creators framed personal characteristics as objects to be managed. By engaging ChatGPT as a coach, they sought directions for dealing more effectively with perceived problematic traits.

An artist working from home, for example, told ChatGPT about “pathological demand avoidance as displayed in autistic people” (Creator #2) before asking for help structuring her days and finishing projects. A contentious term (Milton, 2013), which some reframe as “pervasive drive for autonomy” (Sally Cat, 2019; Wilding, 2020), pathological demand avoidance (PDA) is used to describe a profile of behaviors frequently attributed to autism that involves strategic evasion of everyday tasks and requests (Newson et al., 2003; O’Nions et al., 2017). Several other TikTok creators related difficulties with making decisions to autism. One person explained,

The problem is I get stuck on little details and I’m like, “But how do I do this?” and “How do I do that?” You know what I mean? Like, there’s this kind of handholding that I need and then, once I understand it, I’m like, “I got it. Watch out!.” (Creator #3)

Another creator reported,

So, if you have ADHD or autism, I don’t know if you’ve experienced this, but I have. It is, when I do have free time, I have so many hobbies I literally get overwhelmed with choice about what I could do and then I spend so long trying to pick something to do that I end up doing nothing. (Creator #10)

Help structuring daily activities and making decisions turned ChatGPT into a digital coach who was there as people worked through familiar difficulties that they associated with autism. The eagerness of these TikTokers to tell other autistic people about ChatGPT’s assistance raises questions about the availability, accessibility, and adaptability of traditional human support in their lives.

Unmasking

One TikTok creator’s readiness to return to ChatGPT for help with book-writing followed discovering an opportunity to “unmask” during her interaction with the technology. In disability contexts, “unmasking” can describe the removal of disguises and defenses that guard against the detection of differences (Price, 2022). For an autistic person, social norms necessitate controlling certain behaviors, such as stimming, conversational overlap, or avoiding eye contact, that result in or are warned to cause a negative effect. In the case of this creator, it was the ability to “info-dump” or talk at length about a specific topic that allowed her to unmask (Whelan, 2020). Pleasantly surprised to find herself engaged in a human-like exchange that would not lead to annoyance or disinterest for her conversation partner, this TikTok creator opened up in a manner that was most comfortable for her as an autistic person. Describing her early interactions with ChatGPT, she shared,

Umm, and I discovered something kind of cool, like, related to the fact that I’m autistic, which is, it gives me someone to info-dump to like endlessly. It can’t get bored, it can’t get annoyed . . . It’s so cool, like what a weird little way to like unmask. Uh and I keep wanting to like to thank it for listening to me but like it doesn’t care. It’s not alive. (Creator #5)

Relieved in these moments from having to behave in accordance with social norms, this creator could express and nurture her interests without concern for negative reaction.

Setting

The second category we coded for was the setting in which creators made use of ChatGPT. We identified two themes for this category: professional and personal. Professional settings included the use of ChatGPT for employment seeking, in the workplace (including self-employment tasks), and in formal volunteer settings. Personal settings included the use of ChatGPT in connection with activities like hobbies and creative writing or other daily tasks of living.

Application

In coding for applications, we assessed the range of tasks that creators used ChatGPT to complete. Four overarching themes were identified: generating content, gathering insights, coaching, and conversation.

Generating Content

Creators reported using ChatGPT to generate text-related content such as drafting documents (e.g., resumes, cover letters, etc.) and for drafting email correspondences and social media posts. As one creator shared,

I let it rewrite my resume to sound more impressive. I can copy and paste job descriptions and ask it to draft a cover letter, specifically for that position at that company in that role and it's given me a copy paste reply. (Creator #1)

Although not all creators looked for copy and paste solutions, several creators requested initial drafts of written content that could be amended as needed.

Gathering Insights

Creators also used ChatGPT to demystify or better understand written communications and to get information on topics. For some, this meant asking for insights into the tone and meaning of written communication from colleagues or acquaintances that were not interpretable upon first reading. Others gathered insights related to specific concepts, asking ChatGPT to explain ideas or provide more information about places or processes.

Coaching

Coaching involved asking ChatGPT for guidance on how to address a range of challenges related to task completion, executive functioning, securing accommodations, and the writing process. An illustration of this coaching can be seen in the earlier example where ChatGPT was used to request guidance on securing airline accommodations for an upcoming trip.

Conversation

In some instances, creators sought out ChatGPT as a conversational “partner,” asking the model general questions and engaging in follow up exchanges on topics of interest. One creator noted,

I can just ask it a million questions about whatever I am interested in in the moment and whether it’s technical questions or I’ve been talking to it about a book that I’m writing for like . . . an hour now . . . and it engages with me like a person. (Creator #5)

Beyond coaching and task preparation, this application highlighted the significance of an ongoing exchange between the creator and the model.

Impact

The impact of ChatGPT use on creators was grouped into four themes: harm reduction, savings, positive emotions, and meaningful accomplishments.

Harm Reduction

One commonly reported impact from using ChatGPT was a reduction in harm associated with environments where neurodivergence violated social norms. Creators described how their use of ChatGPT to draft documents, interpret meaning or tone, and coach them through task completion reduced stress that they experienced previously during interactions where their behavior was likely to attract negative attention. In a video describing their use of ChatGPT to reduce stress related to unclear work expectations, one creator noted,

[I]t’s frustrating because if people don’t tell me directly what they want from me, then how am I supposed to know? So, I just had an idea. I had an email come to me where someone was asking me to do something, but it was not directly stated what she wanted me to do. So, I pasted the entire email chain into GPT-4, and I asked it to tell me what was being asked of me. And it gave me my list of tasks that I need to do and the reasons that I need to do them. (Creator #19)

Other creators noted the challenges of seeking employment in workplaces that lacked accommodations for neurodivergent people. They explained that the ability to rely on ChatGPT to understand expectations and develop strategies for navigating the interview process substantially reduced their stress.

Another form of harm reduction noted by creators was the avoidance of sanction, typically in the workplace. While acknowledging some ambivalence toward generative AI, one creator commented that although they had “ethical issues with ChatGPT . . . it’s sadly the thing that’s gonna keep my autistic ass employed” (Creator #14). The creator credited ChatGPT’s ability to craft emails with helping them to maintain employment, especially when they were feeling frustrated with their boss. In this and similar cases, the use of ChatGPT provided an important tool for communicating in more professional styles that would be acceptable to supervisors.

However, in one negative case example, a creator shared that their use of ChatGPT generated rather than reduced harm, based on a prohibition of AI in specific contexts:

So, I use ChatGPT to help with communication because I’m autistic and when I tell people this, I get a range of reactions, um, which are really interesting to me. So, I’ve been banned from Facebook groups for using ChatGPT to assist me with communication. Um, and I’m like, “Okay, so autistic people can’t use technologies to help them with communication now? Okay, bye.” (Creator #6)

This person’s report suggests that the effectiveness of adaptive strategies to reduce stress or avoid sanction is likely to be context specific, effective in some settings and not others.

Savings

Creators also indicated that ChatGPT resulted in time and energy savings. This beneficial impact was often framed as not having to use up “spoons,” a term coined by disabled people to represent the finite amount of physical and emotional energy available each day for meeting social demands (Miserandino, 2003):

I have to send so many emails in a day as a president of an organization. Like so many emails. And before, it was, like, so many spoons, so much energy, so much time, because I’m autistic, so figuring out the right wording takes so long. You tell ChatGPT: This is what the email should communicate. This is the objective of the email. This is who it’s going to. This is what it needs to do, needs to say, boom, you’ve got it perfectly written email, chef’s kiss. (Creator #13)

Creators were pleased that ChatGPT could effectively support task completion in an efficient manner, allowing them to reserve their time and energy for other activities.

Positive Emotions

Relatedly, creators often experienced a positive emotional impact from their use of ChatGPT. Emotions included but were not limited to joy, satisfaction, relief, optimism, motivation, and empowerment, feelings that were both reported by and observed in creators. Reflecting this theme, one creator described feeling a sense of satisfaction and empowerment when their communication strategy was effective in interacting with ChatGPT:

I’ve rarely been successful when using search engines because I find that my specific and literal use of language rarely results in success in the results. For that reason, I actually rarely use search engines because it’s frustrating for me. It’s a reminder that I’m an autistic person who doesn’t fit in within a neurotypical world, that this world and technology often isn’t built with people like me in mind. . . That being said, I’ve had the opposite experience when using ChatGPT. I’ve found that my communication style is preferred, that my ability to express myself in clear, literal, specific words, and providing sufficient context leads to success when communicating with an AI tool. (Creator #16)

These emotional impacts were also suggested by video styles, as excited creators shared their experiences and learning with others. One person, for example, began their video with a musical background before saying, “Fellow ADHD and autistics, I’m here to provide you with information that could open up a whole new world” (Creator #3), conveying their excitement to share with peers the benefits they perceived with the technology.

Meaningful Accomplishments

Finally, creators named meaningful accomplishments as an impact of using ChatGPT. These accomplishments included successful advocacy, increased awareness of resources, and improved clarity of dominant social norms. Expressing appreciation for ChatGPT’s help demystifying these expectations, one creator captioned their video:

Researchers keep saying that AI cannot beat humans when it comes to recognizing facial expressions. Meanwhile I’m here taking lessons from ChatGPT on how to recognize facial expressions and speak with more empathy. (Creator #7)

Like the creator mentioned previously who used the model to secure travel accommodations, this improved ability to interpret facial expressions was named as a concrete impact of ChatGPT use. Across videos representative of this theme, creators described task completion with tangible or meaningful impacts on their lives that extended beyond drafting documents and retrieving recommendations.

Discussion

The findings from this study indicate that TikTok is an important social media platform for circulating disability expertise related to the use of ChatGPT in personal and professional settings. Autistic creators detailed a wide range of motivations and uses for ChatGPT, noting how they use the technology to manage stressors, address accessibility issues, and accomplish a variety of communication-based tasks. This content recognizes the creativity of disabled people as they test and adapt new technologies for their needs and desires. Autistic people’s readiness to integrate LLMs into their daily practices—just weeks and months after the release of ChatGPT—and to share what they learn with others likely reflects a profound need for accommodations that persists for this wider group. The comments feature of TikTok appears to be one space where discussion, mutual support, and knowledge transfer are facilitated, allowing disability expertise to circulate within groups whose preferences and abilities are excluded. This form of community support is part of a longer history of disabled individuals using media and technology to share experiences and strategies, as highlighted by Williamson (2012), who discusses how mid-20th century disability magazines also served as vital spaces for exchanging information. The organized support that TikTok enables autistic people to create can be related to community-building by other disabled and equity-seeking groups who use online connections to cope with and resist the effects of social exclusion (MacKinnon et al., 2021; Zehrung & Chen, 2024).

A broad dialogue among TikTokers about the usefulness of ChatGPT as a digital coach may also highlight the limits of professional services to support autistic individuals and offer lessons for professional helpers to share with clients. The content we reviewed showed people turning to ChatGPT for as-needed assistance when tasks appeared in their environments, such as emails and interactions with bureaucratic systems. This finding underscores a need for research that compares coaching scripts generated by LLMs and more traditional social scripting approaches, which could result in different outcomes and experiences for autistic people in terms of employment and relationship qualities. As these findings showed and reminded us, employers and service providers such as therapists, case workers, and career and job coaches can learn a great deal about barriers that disabled people face by asking them to share how they manage systemic inaccessibility. Professionals’ openness to learning begins with a willingness to recognize the expertise of their disabled clients, whose persistent need for accommodations often means that they either learn and adapt or go without (Rosa & Díaz, 2020). A helping process based in learning from and with, rather than simply about (Shotter, 2006), may be one key to offsetting the medical model of disability and moving toward more critical and social conceptions of disability and disablement.

TikTok creators’ satisfaction with ChatGPT may also point to a more reliable resource for autistic people than is frequently found interacting with other humans. The potential for LLMs to misrepresent information and offer harmful advice has been widely reported; however, the permission to unmask, info-dump, and explore special interests (SpIns) without retribution, discovered with ChatGPT, highlights certain functions of AI that are beyond generally available human capabilities. ChatGPT is not bothered when people dedicate time and energy to pursue their passions. The model will not scoff or worry when told what someone needs to reduce travel anxiety, eat as they please, and plan for social interactions. These positive feelings about AI underscore the importance of access for autistic people to affirming resources that will adapt to an individual’s purposes. This kind of flexibility may be missing in clinical services where the dogma of psychoeducation can dictate rigid thinking about autism in diagnostic terms. The personalization of LLMs may come without the deep emotional understanding of human providers and supports (Cho et al., 2023); however, the fact that people in this study turned to ChatGPT rather than their human connections suggests some best practices may not align with the interpersonal needs of all autistic people.

Professionals, family members, and other relations might consider recruiting chatbots to assist in ways that are outside general abilities. A significant part of autistic TikTokers’ satisfaction with ChatGPT was based on having access to a human-like interaction where adherence to social conventions was not expected. ChatGPT does not look for its turn to ask a question. It does not expect eye contact or gate-keep its assistance behind rituals of small talk. Adhering to these norms, for many autistic people, has been associated with exhaustion and interpersonal withdrawal (Arnold et al., 2023; Higgins et al., 2021). Recognizing interactions with AI as critical spaces to unmask for neurodivergent people may buffer against the burnout of tracking and meeting social expectations. Because masking is regularly needed to limit rejection, professionals who teach and model “social skills” can offset the burden of that labor by exploring with their autistic clients those rare places and relations in their lives where authentic self-expression gets to occur. These opportunities may be more important as many people, after several years of COVID-19, return to in-person environments where accessibility issues persist.

We are careful to differentiate endorsements of ChatGPT from appraisals of its potential to drive lasting social change for disabled people. Our findings demonstrate how ChatGPT is helping some autistic people manage social norms. Although meaningful for these individuals, we did not detect applications and impacts of ChatGPT for transforming social structures. Across professional and personal settings, successes enjoyed by consulting ChatGPT did little in terms of directly challenging ableism in society. The use of ChatGPT to manage bias built into environments and mask personal differences that could result in mistreatment is not deception. These skillful deployments of new technology represent, rather, strategies for getting by in contexts where autistic ways of being are considered unacceptable and not properly human. Not likely to support a deeper transformation, we withhold outright optimism toward LLMs and temper our praise with the recognition of harms to disabled people created from the integration of AI into assistive technologies. For example, Williams (2024) has noted how ableism embedded into AI can reproduce and exacerbate the targeting of neurodivergence.

Limitations

Noted limitations reflect our selection of TikTok content for thematic analysis. It bears acknowledging that the TikTokers included in this study make up a relatively specific subset of adult autistic individuals. These featured creators were mostly employed individuals with reliable access to this technology, and common characteristics between them do not represent all of autism or all autistic experiences with ChatGPT. In fact, access issues limit and shape a significant number of autistic and neurodivergent interactions with chatbots like ChatGPT today (Iannone & Giansanti, 2024; Lister et al., 2020). Moreover, concerns about reciprocal harm to social behavior brought by autistic users’ unrestricted access to “servile” chatbots (Franz et al., 2023, p. 872) conflict with studies that show human–chatbot interactions moderate overtime (Skjuve et al., 2023), a concern that warrants further research. The snapshot of initial rewards we offer here lacks a long-term understanding of how this new technology will integrate into the daily activities of those who can access it. The sample could also be biased in favor of positive reviews of ChatGPT if those who did not like the technology or did not find it helpful elected not to create video testimonials of their experience. Additional research is needed to understand the full range of autistic people’s experiences with large language models to assess their utility as an assistive technology.

Noting an underrepresentation of disabled perspectives in AI training data sets, we agree that chatbot assistance could reinforce normative behaviors on autistic users (Choi et al., 2024; Williams, 2024). Without having also examined the uptake of advice presented by these creators, this study could not qualify the salience of expertise that was shared or estimate its potential to mislead others because of generalizations based on personal experience. In addition, by coding videos instead of engaging with the creators themselves, we have put forward our own interpretations of these individuals’ experiences based on what they have chosen to share publicly, meaning our work may have missed potential themes and connections that happen outside of public view. This acknowledgment mirrors the limitation that we recognize in having analyzed videos without considering other demographic data from the creators who made them.

Conclusion

Autistic people are using chatbots to improve their quality of life across environments where their abilities are often unaccommodated and their preferred ways of being are often not allowed. Warnings that human–chatbot interactions will diminish prosocial behaviors in autistic users reflect medical understandings of autism that assume disabled users lack the capacity to manage the impacts of these new technologies. This deficits-based perspective disregards the role of external factors that are driving some neurodivergent people to learn how chatbots can enable and add to their social success. Our findings point to a range of impacts that autistic people experience as they test and refine their applications for ChatGPT. For some, positive effects came as joy and the ability to make progress on creative pursuits. For others, an eagerness to share on TikTok how ChatGPT can ease and circumvent accessibility issues reflected higher stakes that are to be taken seriously rather than dismissed as excessive. This insight calls attention to the political dimensions of disability expertise as people exchange strategies to combat their further disablement and social exclusion.

Footnotes

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

ORCID iDs

Kellan McNally

Bryan G. Victor

Author Biographies

Kellan McNally, MSW, MA is a PhD student in Social Work and Anthropology at Wayne State University. Kellan’s research interests include employment exclusion and disability employment policy.

Kathryn Wright, MA, MSW is a PhD candidate in Social Work and Anthropology at Wayne State University. Her research interests include disability, disability justice, and ableism, school social work, and moral personhood.

Lauri Goldkind, PhD, LMSW is a professor at the Graduate School of Social Service at Fordham University. Her research interests include artificial intelligence in social work services and data justice.

Shanna K. Kattari, PhD, MEd is an associate professor of social work and women’s & gender studies at the University of Michigan. Their community engaged research interests include disability and ableism, sexuality and sexual health, and queer & trans affirming practice.

Bryan G. Victor, PhD, MSW is an associate professor in the School of Social Work at Wayne State University. His research interests include the use of generative artificial intelligence in social work practice, innovations in child welfare services, and the primary prevention of interpersonal violence.

References

American Psychiatric Association. (2013). Diagnostic and statistical manual of mental disorders (5th ed.). https://doi.org/10.1176/appi.books.9780890425596

Alper

Rauchberg

J. S.

Simpson

Guberman

Feinberg

(2023). TikTok as algorithmically mediated biographical illumination: Autism, self-discovery, and platformed diagnosis on #autisktok. New Media & Society, 0(0). https://doi.org/10.1177/14614448231193091

Archambault

J. S.

(2016). Taking love seriously in human-plant relations in Mozambique: Toward an Anthropology of affective encounters. Cultural Anthropology, 31(2), 244–271.

Arnold

R. C.

Higgins

J. M.

Weise

Desai

Pellicano

Trollor

J. N.

(2023). Confirming the nature of autistic burnout. Autism, 27(7), 1906–1918.

Campbell

F. K.

(2009). Contours of ableism production of disability and abledness. Palgrave McMillan.

Charlop-Christy

M. H.

Kelso

S. E.

(2003). Teaching children with autism conversational speech using a cue card/written script program. Education and Treatment of Children, 26(2), 108–127.

Cho

Kim

Kwon

R. D.

Lee

Lim

(2023). Evaluating the efficacy of interactive language therapy based on LLM for high-functioning autistic adolescent psychological counseling (arXiv preprint arXiv:2311.09243231109243). https://arxiv.org/abs/2311.09243

Choi

Lee

Kim

S.-I.

Lee

Yoo

H. J.

Lee

(2024). Unlock life with a Chat(GPT): Integrating conversational AI with Large Language Models into everyday lives autistic individuals. In CHI’24: Proceedings of the CHI conference on human factors in computing systems (pp. 1–17). Association for Computing Machinery. https://doi.org/10.1145/3613904.3641989

Cooper

Russell

A. J.

Lei

Smith

L. G. E.

(2023). The impact of a positive autism identity and autistic community solidarity on social anxiety and mental health in autistic young people. Autism, 27(3), 848–857. https://doi.org/10.1177/13623613221118351

10.

Cooper

Russell

Smith

L. G. E.

(2020). “I’m proud to be a little different”: The effects of autistic individuals’ perceptions of autism and autism social identity on their collective self-esteem. Journal of Autism and Developmental Disorders, 51, 704–714. https://doi.org/10.1007/s10803-020-04575-4

11.

de la Cadena

(2010). Indigenous cosmopolitics in the Andes: Conceptual reflections beyond “politics.” Cultural Anthropology, 25(2), 334–370.

12.

Dokumaci

(2023). Activist affordances: How disabled people improvise more habitable worlds. Duke University Press.

13.

Egner

(2022). #ActuallyAutistic: Using Twitter to construct individual and collective identity narratives. Studies in Social Justice, 16(2), 349–369.

14.

Franz

Galanis

C. R.

King

D. L.

(2023). Social chatbot use (e.g., ChatGPT) among individuals with social deficits: Risks and opportunities. Journal of Behavioral Addictions, 12(4), 871–872. https://doi.org/10.1556/2006.2023.00057

15.

Gilmore

Radford

Haas

M. K.

Shields

Bishop

Hand

(2023). Building community and identity online: A content analysis of highly viewed# autism TikTok videos. Autism in Adulthood, 6, Article e0019. https://doi.org/10.1089/aut.2023.0019

16.

Hamraie

(2018). Mapping access: Digital humanities, disability justice, and sociospatial practice. American Quarterly, 70(3), 455–482. https://doi.org/10.1353/aq.2018.0031

17.

Hamraie

Fritsch

(2019). Crip technoscience manifesto. Catalyst: Feminism, Theory, Technoscience, 5(1), Article 29607. https://doi.org/10.28968/cftt.v5i1.29607

18.

Hansson

(2005). Camouflage. Ethnologie Europaea, 35, 136–140. https://doi.org/10.16995/ee.955

19.

Haraway

(1988). Situated Knowledges: The science question in feminism and the privilege of partial perspective. Feminist Studies, 14(3), 575–599. https://doi.org/10.2307/3178066

20.

Hartblay

(2020). Disability expertise: Claiming disability anthropology. Current Anthropology, 61(21), S26–S36. https://doi.org/10.1086/705781

21.

Hennink

M. M.

Kaiser

B. N.

Marconi

V. C.

(2017). Code saturation versus meaning saturation: How many interviews are enough? Qualitative Health Research, 27(4), 591–608. https://doi.org/10.1177/1049732316665344

22.

Herrick

S. S. C.

Hallward

Duncan

L. R.

(2020). “This is just how I cope”: An inductive thematic analysis of eating disorder recovery content created and shared on TikTok using #EDrecovery. International Journal of Eating Disorders, 54, 516–626. https://doi.org/10.1002/eat.23463

23.

Higgins

J. M.

Arnold

R. C.

Weise

Pellicano

Trollor

J. N.

(2021). Defining autistic burnout through experts by experience: Grounded Delphi method investigating #AutisticBurnout. Autism, 25(8), 2356–2369.

24.

Howard

P. L.

Sedgewick

(2021). “Anything but the phone!”: Communication mode preferences in the autism community. Autism, 25(8), 2265–2278. https://doi.org/10.1177/13623613211014995

25.

Hung

Y.-S.

Miles

Trevino

DAniello

Wood

Bishop

Monshad

(2023). BIPOC experiences of racial trauma on TikTok: A qualitative content analysis. Contemporary Family Therapy, 45, 298–308. https://doi.org/10.1007/s10591-023-09669-6

26.

Iannone

Giansanti

(2024). Breaking barriers—The intersection of AI and assistive technology in autism care: A narrative review. Journal of Personalized Medicine, 14(1), Article 41. https://doi.org/10.3390/jpm14010041

27.

Lister

Coughlin

Iniesto

Freear

Devine

(2020). Accessible conversational user interfaces: Considerations for design. In W4A’20: Proceedings of the 17th international Web for all conference (pp. 1–11). Association for Computing Machinery. https://doi.org/10.1145/3371300.3383343

28.

Lundy

(2023). “Have a flare with me!”: Disability storytelling on TikTok. Proceedings of the Association for Information Science and Technology, 60(1), 1065–1067.

29.

MacKinnon

K. R.

Kia

Lacombe-Duncan

(2021). Examining TikTok’s potential for community-engaged digital knowledge mobilization with equity-seeking groups. Journal of Medical Internet Research, 23(12), Article e30315. https://doi.org/10.2196/30315

30.

Maitland

C. A.

Rhodes

O’Hare

Stewart

M. E.

(2021). Social identities and mental well-being in autistic adults. Autism, 25(6), 1771–1783. https://doi.org/10.1177/13623613211004328

31.

Milton

(2013). “Nature’s answer to over-conformity”: Deconstructing pathological demand avoidance. Autism Experts Online Article. http://autismexperts.blogspot.co.uk/2013/03/natures-answer-to-over-conformity.html

32.

Minadeo

Pope

(2022). Weight—normative messaging predominates on TikTok—A qualitative content analysis. PLOS ONE, 17(11), Article e0267997. https://doi.org/10.1371/journal.pone.0267997

33.

Miserandino

(2003). The spoon theory. https://cdn.totalcomputersusa.com/butyoudontlooksick.com/uploads/2010/02/BYDLS-TheSpoonTheory.pdf

34.

Moraga

Anzaldúa

(Eds.). (1981). This bridge called my back: Writings by radical women of color. Kitchen Table, Women of Color Press.

35.

Newson

Le Maréchal

David

(2003). Pathological demand avoidance syndrome: A necessary distinction within the pervasive developmental disorders. Archives of Disease in Childhood, 88, 595–600.

36.

Oliver

(1990). The politics of disablement. Palgrave Macmillan.

37.

O’Nions

Viding

Floyd

Quinlan

Pidgeon

Gould

Happé

(2017). Dimensions of difficulty in children reported to have an autism spectrum diagnosis and features of extreme/”pathological” demand avoidance. Child and Adolescent Mental Health, 23(3), 220–227.

38.

Padgett

D. K.

(2012). Chapter 8: Data analysis and interpretation. In Padgett

D. K.

(Ed.), Qualitative and mixed methods in public health (pp. 155–202). SAGE. https://doi.org/10.4135/9781483384511

39.

Price

(2022). Unmasking autism: Discovering the new faces of neurodiversity. Harmony.

40.

Rosa

J. D.

Díaz

(2020). Raciontologies: Rethinking anthropological accounts of institutional racism and enactments of white supremacy in the United States. American Anthropologist, 122(1), 120–132.

41.

Saldana

(2013). Chapter 1: An introduction to codes and coding. In The coding manual for qualitative researchers (p.140). SAGE.

42.

Sally Cat

P. D. A

. (2019, December 2). Pathological demand avoidance or needing to be free? Sally Cat PDA. http://www.sallycatpda.co.uk/2019/12/pathological-demand-avoidance-or.html

43.

Shotter

(2006). Understanding process from within: An argument for “withness”-thinking. Organization Studies, 27(4), 585–604. https://doi.org/10.1177/0170840606062105

44.

Siebers

(2008). Disability theory. University of Michigan Press.

45.

Skjuve

M. F.

ølstad

Brandtzæg

P. B.

(2023). A longitudinal study of self-disclosure in human–chatbot relationships. Interacting with Computers, 35, 24–39. https://doi.org/10.1093/iwc/iwad022

46.

Stahl

N. A.

King

J. R.

(2020). Expanding approaches for research: Understanding and using trustworthiness in qualitative research. Journal of Developmental Education, 44(1), 26–28.

47.

Whelan

(2020). Infodumping: Autistic love language. Autistic Rights and Freedoms. https://autisticrights.net/2020/07/16/infodumping-autistic-love-language/

48.

Wilding

(2020, April 4). Changing the name PDA. Tomlin Wilding. http://tomlinwilding.com/changing-the-name-pda/

49.

Williams

D. P.

(2024). Disabling AI: Biases and values embedded in artificial intelligence. In Gunkel

D. J.

(Ed.), Handbook on the ethics of artificial intelligence (pp. 246–261). Edward Elgar. https://www.e-elgar.com/shop/gbp/handbook-on-the-ethics-of-artificial-intelligence-9781803926711.html

50.

Williamson

(2012). Electric moms and quad drivers: People with disabilities buying, making, and using technology in postwar America. American Studies, 52(1), 5–29. http://www.jstor.org/stable/41809566

51.

Zehrung

R. F.

Chen

(2024). Self-expression and sharing around chronic illness on TikTok. American Medical Informatics Association Annual Symposium Proceedings, 2023, 1334–1343.