Comments:
Critical digital literacies, generative AI, and the negotiation of agency in human-AI interactions
Author: Kyu Yun Lim and Ron Darvin
Kyu Yun Lim, Ron Darvin, Critical digital literacies, generative AI, and the negotiation of agency in human-AI interactions, System, Volume 136, 2026, 103904, ISSN 0346-251X, https://doi.org/10.1016/j.system.2025.103904. (https://www.sciencedirect.com/science/article/pii/S0346251X25003148)
0 General Document comments
0 Sentence and Paragraph comments
0 Image and Video comments
Abstract
Drawing on the notion of critical digital literacies (CDL), this qualitative case study investigates the extent to which multilingual secondary school students in Canada exercise agency when they interact with generative artificial intelligence (GenAI) technologies.
Recognizing how learners have unequal access to resources that involve the use of GenAI, it poses the question: To what extent do learners enact CDL, and what enables or constrains their enactment?
Data were collected through interviews, observations of learners' GenAI usage, and digital artifacts, including screenshots of prompts and AI-generated texts.
Thematic analysis was employed to understand learners’ dispositions towards these technologies, and their GenAI practices, while multimodal analysis provided insights into how learners negotiated their agency as they interacted with GenAI platforms.
Represented by two exemplar cases, findings show how learners who used GenAI primarily through apps in mobile phones or chatbots embedded in social media platforms, tended to adopt a passive stance towards GenAI, accepting generated output without much critical evaluation.
In contrast, learners with access to a laptop and multiple GenAI platforms, demonstrated a more critical disposition towards this technology, comparing outputs across platforms, questioning their reliability, and verifying information through lateral reading.
By drawing attention to how material conditions of access and platform design can constitute dispositions towards GenAI and CDL, this study calls on educators to support learners in identifying and challenging inequalities and ideological biases embedded in these technologies, to foster more equitable and inclusive GenAI literacy practices.
May 7
Paul A's Question
Paul A and/or Thinking Partner
:
I'm wondering how they will define -- so that it can be measured -- "exercise agency". I'll be looking for more explanation as I read.
more
Thinking Partner with Paul A's edits:
- That note sets up a strong reading goal: you are tracking whether the study turns a big idea into something observable. In research writing, that often means watching for how the authors connect a term to behaviors, decisions, or patterns in the data.
- You might keep an eye on whether “exercise agency” gets shown through things learners actually do, like comparing outputs, questioning answers, revising prompts, or checking other sources. If the term stays broad, that may be worth noting in your response.
- Another useful test is this: does the article give a clear difference between agency and simple use? That distinction may help you decide whether the concept is being measured carefully or just named.
Try this in Your Thoughts: write one sentence that begins with “I’m looking for…” and one sentence that begins with “I’ll know the authors define it clearly if…”. Then either click Continue with AI to think through the next paragraph, or go to Create Document to upload another source to read and annotate or to start writing.
A classmate or teacher might spot a different clue about how the article handles key terms, so it could be useful to compare notes once you gather a little more evidence.
Keywords
1. Introduction
The rapid advancement of generative artificial intelligence (GenAI) tools such as ChatGPT has brought remarkable changes to how we live, work, and communicate.
While AI technologies such as adaptive tutoring systems and chatbots have been present in educational contexts for years, the emergence of GenAI has granted language learners access to powerful AI tools through an intuitive, user-friendly interface, representing an "access revolution" (Mintz et al., 2023, p. 325).
The ability to generate human-like texts, voices, and images in response to prompts has created new possibilities for language education (Kaplan-Rakowski et al., 2023).
GenAI can serve as a conversational partner for simulating real-world language use (Han, 2024), assist students with writing tasks (Song & Song, 2023), and provide instant, individualized feedback (Chen et al., 2023).
However, alongside these affordances, GenAI has also raised significant concerns regarding mis/disinformation, data privacy, the violation of intellectual property rights, and its impact on the economic, environmental, and sociopolitical landscape (Zhang & Tur, 2024).
Given these implications, it is crucial to critically examine not only how these tools function but also how they shape educational practices and relations of power.
A systematic review of studies on ChatGPT in language education since its release in 2022 (Li et al., 2024) has shown that that the primary research foci have been on exploring ChatGPT's role in enhancing language skills, assessing its affordances, and identifying strategies for its integration.
A growing body of GenAI research has explored its pedagogical potential in language and literacy education.
This includes studies on enhancing language learning and teaching (Bonner et al., 2023; Guan et al., 2024; Kohnke et al., 2023; Lee & Lee, 2024; Pack & Maloney, 2023), examining learner perceptions and practices (Liu et al., 2024; Wang, 2024) and teacher readiness (Kaplan-Rakowski et al., 2023; Moorhouse, 2024).
Other areas of focus address the development of AI literacies (Kang & Yi, 2023; Warschauer et al., 2023), comparison of linguistic features of human versus AI-generated texts (Jiang & Hyland, 2025), the impact of AI tools on writing assessment (Woo et al., 2024) and the role of GenAI in multimodal composition (Jiang & Lai, 2025; Smith et al., 2025).
While these studies provide valuable insights, they largely emphasize the functional digital literacies that enable learners to operate GenAI tools and to achieve their intentions by engaging strategically with existing features and functionalities.
Critical digital literacies (CDL) on the other hand raise questions about the logics embedded in the design of these tools, the output that they generate, and the unequal ways in which features and functionalities are made available to learners.
For Darvin (2025a), CDL are “the practices of interrogating and resisting how power operates in the reproduction of ideologies, inequalities and modes of exclusion in digitally mediated spaces” (p.3).
CDL recognizes that while AI tools provide learners with new agentive possibilities in terms of language learning, the designs and algorithms of these tools can also steer interactions in ways that are often invisible to users (Godwin-Jones, 2024; Jones, 2021).
By providing singular, definitive responses to prompts, GenAI platforms have the power to shape ways of thinking and to assert particular linguistic norms and algorithmic biases (Shieh et al., 2024).
CDL highlights how GenAI is entangled with issues of power, ownership, and inequality in language education.
GenAI systems reflect and reinforce existing dominant power structures as “a registry of power” (Crawford, 2021, p.8).
Kelly-Holmes (2024) draws attention to the ways AI tools might reshape linguistic practices, particularly through the privatization of digital spaces and invisibilization of technology.
Technology companies design GenAI tools that are deployed into educational contexts, but that run on proprietary algorithms hidden from users.
This lack of transparency has significant implications for learner agency, as it constrains users from understanding how they operate and how specific ideologies are embedded in their design.
These technologies can “simultaneously enhance and obscure elements of control for individuals” (Kessler & Casal, 2024, p.31), implying that while learners may gain new capabilities, they also become subject to hidden forms of control.
GenAI platforms could concentrate power in tech companies while constraining how teachers and learners interact with language and learning.
Operating as a black box, GenAI reinforces dominant linguistic ideologies by privileging standardized, monolingual forms of communication (Wang et al., 2021).
For example, AI-powered voice recognition systems frequently struggle with regional variations and accents (Ramesh et al., 2023), marginalizing non-standard language users and flattening linguistic diversity.
This issue presents challenges for language educators striving to promote pluralistic thinking and intercultural competence.
Without transparency regarding how training data is selected or how algorithms are configured, language learners may find it difficult to challenge the biases embedded in AI-generated content, and to recognize whose voices are amplified or silenced.
Leander and Burriss (2020) posit that addressing this issue involves redefining critical literacy to account for the complexities of interacting with intelligent technologies.
AI-mediated literacies need to involve questioning the origins, biases, and implications of AI-generated content, and building “more ethical assemblage with computational agents” (Leander & Burriss, 2020, p.1263).
Such considerations underscore the importance of examining learners’ critical digital literacy practices to navigate, critique, and leverage GenAI tools in ways that help them achieve their intentions while negotiating inequalities and modes of exclusion.
While teachers and school leaders have expressed the need to guide students in interrogating AI-generated information (Chiu, 2023), there remains a gap in research on how learners critically engage with these tools in practice.
Scholars have begun to explore how GenAI technologies reshape our culture, knowledge, and identities (Jones, 2025) through algorithmic decision-making and platform designs, but studies that examine the extent to which learners demonstrate a critical awareness of GenAI remain scarce.
To address this gap, this paper aims to explore the critical digital literacy practices of Canadian multilingual youth in the context of generative AI technologies.
Specifically, it seeks to address the following research questions:
By examining how learners interact with AI tools and analyzing the broader sociocultural and educational implications, this study contributes to a deeper understanding of critical digital literacies as a pedagogical imperative in the age of AI.
Additionally, it sheds light on the ways in which power, agency, and identity are negotiated within human-AI interactions through a lens of critical digital literacies (Darvin, 2017, 2025a).
2.1.
Critical digital literacies (CDL)
To address these research questions, this paper draws on the framework of critical digital literacies (CDL) informed by the critical sociomateriality (Fenwick, 2015).
A sociomaterial view foregrounds how meaning-making and learning with digital technologies are shaped by the interplay of human and non-human actors, including users, devices, platforms, and algorithms i.e. human-AI interaction.
From a critical perspective, these interactions are always “a site of struggle” (Darvin, 2025b, p.4) because learner agency is always negotiated with platform designs, algorithms and outputs that index various ideologies and have the power to assert ways of thinking and doing.
Digital spaces are not neutral; rather, they are sites that can privilege some users while excluding others.
CDL recognizes how structures of power within digital contexts shape knowledge production, the formation of identities, and the negotiation of social relations (Darvin, 2017).
This framing considers how entire ecosystems of digital practices and platforms configure who gains access to what kinds of knowledge and influence (Darvin, 2023).
In this conceptualization, criticality involves seeing beyond the functionalities of digital tools to examine how platform architectures and ideologies shape agency.
Within this view, agency is not possessed by individuals alone but emerges relationally, distributed across these interconnected entities (Godwin-Jones, 2024).
The notion of distributed agency highlights how human intention can be both enabled and constrained by dynamic interactions among social, cultural and technological systems.
Through the use of digital tools and the influence of ideologies surrounding them, learners develop different dispositions towards particular technologies, recognizing their affordances, constraints, and potential pitfalls.
These dispositions in turn shape their digital literacy practices.
The conceptualization of CDL has evolved through the foundational work of earlier scholars.
Avila and Pandya (2013) establish an initial understanding of CDL that is closely aligned with traditional notions of critical literacy.
Drawing on Luke's (2012) notion of critical literacy and its intersection with multiliteracies, their approach positions CDL as a socially situated praxis.
They focus on how digital texts not only mirror but also reinforce existing power structures and emphasize the importance of redesigning these texts to serve the interests of marginalized communities.
In doing so, they call for empowering learners to move beyond consumption toward active production, critique, and the reconfiguration of digital texts and contexts.
Building on the work of Avila and Pandya (2013), subsequent scholars have continued to expand and refine the concept of CDL in response to shifting technological and social terrains.
Pangrazio (2016) argues that the rapidly evolving digital landscape requires a rethinking of what it means to engage critically with digital media.
She identifies three primary orientations within existing conceptualizations of CDL: an ethical focus on power and ideology, a personal lens emphasizing identity and affect, and a design-based approach that encourages production and creativity.
Pangrazio (2016) proposes an integrative framework called “critical digital design” (p.170), which encourages learners to understand digital platforms as “architectures” (p.160) shaped by political, economic, and cultural forces, while also recognizing how personal investments and creative endeavors can become sites of critical reflection and action.
Recognizing the significance of sociotechnical systems in the way digital practices evolve, (Pötzsch, 2021) insists on including the material conditions of digital infrastructures as central to CDL.
He brings environmental degradation, resource extraction, energy consumption, and labor exploitation into the discussion.
In this way, CDL transcends textual analysis and includes the socio-political and ecological sustainability of digital networks.
In a similar vein, Garcia and de Roock (2021) argue that “critical literacy must be reexamined and redefined under the current yoke of platform capitalism and racial capitalism more broadly” (p. 193).
They advocate for an “abolitionist” vision within CDL, urging us to “interrogate civic platforms: classroom, political sphere, and digital platforms” (p.196).
Their emphasis on social movements adds civic dimensions to CDL.
Jones (2022) adds further complexity by arguing that CDL must attend to the social, affective, and relational dimensions of online engagement.
He calls for expanding beyond a cognitive frame of literacy to account for how emotions, affinity, and action shape what we do with digital tools and content.
Leander and Burriss (2020) push the conversation further into posthumanist territory.
They ask how CDL must adapt when AI, algorithms, and nonhuman agents now play integral roles in generating and interpreting digital texts and CDL must now consider the assemblages of human and computational agents that co-create meaning.
Their perspective urges literacy educators and researchers to confront how nonhuman actors reshape identity, ethics, and social relations and to “build more ethical assemblage with computational agents” (p.1263).
Shared across these perspectives of what constitutes CDL is an understanding of how digital practices do not just emerge as a matter of choice, but that they are constantly negotiated with material tools, contexts, and varied resources.
They recognize that to exercise agency in the use of these tools, learners do not just need to know how to use these tools, but also how these tools operate with a particular logic embedded in their design and their output, and serve the intentions and interests of their developers.
By understanding digital literacy practices as a negotiation of power, CDL is driven by the imperative to move from mere interpretation toward transformative action.
From a CDL perspective, the recent turn toward generative AI tools such as ChatGPT adds both urgency and complexity to the longstanding call to interrogate power, ideology, and epistemology in digital contexts.
While initial studies, as reviewed by Li et al.
(2024), have focused on practical integration, examining how ChatGPT can enhance language skills, provide feedback, or improve instructional design, they often overlook the deeper critical implications.
CDL scholars emphasize that digital literacies should not be limited to a set of skills or the capacity to identify technological affordances.
Instead, they must situate digital literacy practices in relation to structures of power, political economy, and sociotechnical design.
Within these critical framings, GenAI must be approached not only as a tool that can produce texts, but as a sociotechnical system that encodes particular ideologies, prioritizes certain epistemologies, and operates through specific economic priorities.
Therefore, it is imperative to transform learners into critical agents capable of interpreting, contesting, and reshaping digital ecosystems rather than merely consuming them.
Focusing on these issues of power, materiality and access, this study employs CDL as a lens to investigate how multilingual youth engage with GenAI technologies.
By examining how power circulates through digital platforms and how learners negotiate agency within these infrastructures, this framework situates digital literacy as an active, socially situated practice that extends beyond technical proficiency to encompass critical engagement and negotiation of agency.
2.2.
Critical perspectives on GenAI and digital literacies of young learners
While research on GenAI use among language learners is still emerging, empirical studies have begun to examine learners' practices and perceptions across diverse populations and educational contexts.
These studies reveal a recurring tension in learners’ interactions with GenAI between developing functional digital literacies (e.g., prompt engineering) and cultivating critical digital literacies, such as interrogating bias, evaluating reliability, and recognizing issues of power.
Although some research has focused on how learners leverage GenAI to support their writing, idea exploration and content creation (Baek et al., 2024; Picton & Clark, 2024), educators frequently note that learners need to develop critical awareness, necessary to engage with these tools (Higgs & Stornaiuolo, 2024; Mah et al., 2024; Moorhouse, 2024).
Yan (2023) and Higgs and Stornaiuolo (2024) explore how students engage with GenAI inside and outside formal educational settings.
Yan's study of Chinese EFL undergraduates highlights their evaluation of the tools: they were impressed by the speed, structural coherence, idiomaticity, flexibility, and above-average writing quality of ChatGPT's outputs.
However, the study also addresses the ethical and pedagogical difficulties that surfaced as students engaged with ChatGPT.
Many participants voiced serious concerns about fairness and academic honesty, viewing the tool as a disruptive presence that could offer undue advantages to those who rely on it to complete assignments.
Participants feared that, if left unchecked, widespread use of ChatGPT might devalue genuine learning, reduce motivation, and undermine the equitable principles that should guide educational assessment.
Meanwhile, Higgs and Stornaiuolo's research (2024) documents how secondary students critically reflected on GenAI's affordances and constraints, including its impact on creativity, originality, and bias.
Students demonstrated a keen awareness that GenAI output is not neutral.
Echoing sociotechnical critiques, they expressed concerns about embedded biases and data sets reflecting societal inequities.
They also recognized that widespread AI adoption could shift what counts as authentic knowledge production and might even threaten future opportunities to learn and practice human-centered writing.
Despite this recognition, students also expressed an overall lack of learning opportunities in school “contributed to their uncertainties about the social impacts of technology” (Higgs & Stornaiuolo, 2024, p.644).
A key insight from these studies is that students frequently use GenAI as a thought partner but express concerns about its influence on authenticity and creativity.
Mah et al.
(2024) introduce the concept of “cognitive lift” (p.8), emphasizing the need to differentiate between AI as a scaffold that supports learning versus a shortcut that bypasses cognitive effort.
Similarly, Hellmich et al.
(2024) examine students learning non-English languages at a large U.S. university and find that 74% of learners opted not to use ChatGPT due to concerns about over-dependency, cheating, intellectual theft, and distrust in AI's accuracy.
Some also pointed out ChatGPT's limitations in capturing cultural and linguistic nuance.
Donley (2024) investigates how teachers use ChatGPT for multilingual learners in U.S. public schools, documenting how ChatGPT can be used for translation, differentiated instruction, and writing support within a translanguaging literacy pedagogies framework.
While students who received guidance in prompt construction demonstrated improved agency and content access, the study found minimal efforts to foster critical reflection on AI's biases, limitations, or sociopolitical dimensions.
This highlights a broader pedagogical gap in cultivating critical digital literacy practices that move beyond functional AI use.
These studies collectively indicate that while GenAI tools are widely accessible, critical engagement remains inconsistent.
Students are capable of reflecting on these tools' ethical and academic implications (Donley, 2024; Hellmich et al., 2024; Mah et al., 2024; Yan, 2023), but students’ critical engagement often depends on explicit scaffolding and instructional interventions that foreground ethical and sociotechnical considerations (Higgs & Stornaiuolo, 2024).
While earlier research on GenAI has primarily focused on its affordances, more studies adopting critical perspectives are needed to foreground how AI systems reproduce power hierarchies, exacerbate inequality, and mediate knowledge production in ways that privilege certain epistemologies while marginalizing others (Vassel et al., 2024).
To this end, this study aims to contribute the understanding of the extent to which learners negotiate agency in contesting, reinterpreting, and shaping their GenAI practices.
This study employs a case study methodology that involves an in-depth examination of a phenomenon through particularization and thick descriptions (Duff, 2014; Yin, 2014).
A case study design enables a comprehensive understanding of complex phenomena through a case as an exemplar (Duff, 2014).
In this study, we recruited six multilingual learners from a secondary school in Metro Vancouver, BC.
The researchers first introduced the study's aims and procedures in each class.
Students who expressed interest were asked to complete a brief survey about their linguistic backgrounds and their technology use.
Among the interested students, we contacted those who were multilingual and reported using GenAI for personal and/or academic purposes.
Students who provided both assent and parental consent were then formally recruited for the study.
Participants were in Grades 8 through 12 and spoke languages other than English at home, including Farsi, Malayalam, Vietnamese, Punjabi, Hindi, and Sinhala.
Data was collected through two 1-hour sessions with each participant.
All sessions were conducted in person in a private office space within the school, with the exception of one participant who completed his second session online.
Each session involved semi-structured interviews and observation of the participants’ use of GenAI.
Data includes the audio-recording of each session and digital artifacts such as screenshots of their interaction on GenAI platforms.
In the first session, participants discussed backgrounds, their access to and use of digital devices, their motivations for using GenAI, and their practices and beliefs regarding these platforms.
They were also encouraged to share their prior interactions if they felt comfortable.
Participants verbalized their thoughts and intentions through think-aloud protocols while showing how they use GenAI platforms.
Between sessions, participants were asked to document their GenAI use by collecting screenshots of interactions.
These artifacts served as prompts for the second session.
In the second session, participants explained how they used GenAI for their personal and academic purposes, and how their use of GenAI evolved over time.
Some of them also performed tasks live such as multilingual interactions or walked through their workflow using GenAI tools, and researchers observed and asked follow-up questions.
These interactions were also captured and shared with researchers after the interviews.
Audio recordings of the interviews were transcribed and analyzed using thematic analysis.
As the transcript included the multilingual interactions, we asked participants if the transcript of the interview reflects their interaction correctly.
Two participants agreed to check their transcriptions voluntarily.
Data collection took place between February and March 2024.
During this period, participants who used the free version of ChatGPT accessed the GPT-3.5 model, which only provided information available up until September 2021 and did not offer image generation functions.
After transcription, screenshots were linked to specific sections of the transcript, allowing for direct cross-referencing of participants’ verbal description with their visual representations.
To analyze data, reflexive thematic analysis (Braun & Clarke, 2006) was employed using NVivo, a computer-assisted qualitative data analysis software.
In the first round, we conducted an open coding process, assigning initial codes such as “purpose of using ChatGPT” or “comparison between platforms.”
In the second round, the codes were categorized and organized in relation to research questions and conceptual lens, generating major themes.
Two researchers independently coded the data and then compared and discussed the themes.
Additionally, a multimodal discourse analysis (Jewitt, 2017) was applied to analyze the screenshots of GenAI interactions and to explore how participants engage with each platform's design.
The research team comprised two researchers with academic and professional backgrounds in digital literacy and language education.
Both had previously worked as secondary language teachers, equipping them with pedagogical knowledge in language and literacy.
This expertise enabled them to engage meaningfully with school contexts and to interpret learners’ digital literacy practices.
Findings from the study have shown that while participants generally demonstrate functional digital literacies in their capacity to use GenAI tools to achieve their own intentions, they enact varying critical digital literacies.
Some learners have a more critical disposition towards these tools, recognizing the limitations of different large language models and the need to verify output.
Such differences can be traced to certain material inequalities, access to a wider range of tools i.e. GenAI platforms and versions, laptops and phones, and the negotiation of agency.
Across six participants, a diverse range of GenAI platforms were used, including ChatGPT, Copilot, Designer, CharacterAI, ChatAI, and Snapchat My AI—each distinguished by unique design features and interaction styles.
These platforms mediated learning activities that spanned academic support (e.g., brainstorming ideas for essay writing, checking marking schemes), language learning (e.g., looking up unfamiliar words, refining sentence structures), coding and web design, image generation, and personal usage such as creating nicknames or generating yearbook quotes.
Learners demonstrated varying dispositions toward GenAI.
Some viewed it as a highly reliable source, with one stating, “I think ChatGPT will give me 100 % correct answer,” while others approached it with skepticism, saying “I don't know if that's the correct answer.”
Concerns about over-reliance also surfaced, with one learner acknowledging “I'm relying on it way more than I should in general.”
Additionally, privacy concerns emerged, particularly regarding Snapchat My AI's potential access to personal data and photos.
To better understand how GenAI engagement is shaped by sociotechnical design, material conditions and learner agency, this study closely examined two contrasting cases: Jennie and Ranjith.
Jennie, who primarily used Snapchat My AI, and Ranjith, who utilized multiple platforms for different purposes, built distinct dispositions and experiences with GenAI technologies as they navigated diverse platforms.
Their cases highlight how learners negotiate their agency in interaction with GenAI, and how access to digital resources and platform affordances interact with learners’ dispositions towards GenAI.
Jennie is a Grade 8 student who speaks English at home and learns French as an additional language at school.
She has her own phone but shares two desktops at home with other family members.
She uses Google and Microsoft Teams for research and other schoolwork.
While she can use the desktop when she needs it, she mainly uses her phone to access various digital tools.
When she learned about the emerging popularity of GenAI tools, she decided to search for apps on her phone rather than to look at sites from a browser on the desktops at home.
She believed that ChatGPT costs money, so she searched for “free AI” in Apple's App Store and downloaded six platforms and tried them, but only kept Snapchat's My AI when she realized that the other platforms were not actually free.
Because My AI is embedded in a free social media platform that Jennie already uses to message her friends, this tool easily became integrated into her practices.
In Excerpt 1, Jennie mentions why she chose My AI as her primary tool.
| Excerpt 1 | |
|---|---|
| Researcher: | What would make you choose to use Snapchat AI over your other AI apps? |
| Jennie: | Because it's free and unlimited. You can just keep asking questions. You don't only get three and you don't pay monthly and (…) |
| Researcher: | And like in terms of so far, the answers that you get for those questions, are you generally happy about the answers? |
| Jennie: | Yeah, they give pretty similar answers, but this one gives sometimes more detailed answers, I guess. |
| Researcher: | More details. |
| Jennie: | Maybe, but I think it's just because this one's free and the other one is like, it's like not a free trial, but you only get three questions, so you can't ask it to have too much detail … |
No paragraph-level conversations.
Start one.
No paragraph-level conversations.
Start one.
No paragraph-level conversations.
Start one.
No paragraph-level conversations.
Start one.
In Excerpt 1, Jennie states that she prefers My AI because it is “free and unlimited”, in contrast to other AI platforms that restrict access through paywalls or limited trials.
Because GenAI platforms are monetized, access to models that provide more reliable information require paid subscription.
ChatGPT Plus, for instance, costs US$20 a month and provides access to GPT-4 that is significantly more advanced than the free model GPT-3.5 in terms of reasoning, multilingual capabilities, safety and contextual understanding (OpenAI, 2023).
My AI runs on GPT-3.5, and because she prioritizes cost, Jennie does not explore the features of Snapchat + which costs C$6.99 a month and provides access to GPT-4.
Apart from My AI being free, Jennie also expresses satisfaction because it “gives sometimes more detailed answers” than the others, and one can keep asking questions to “ask it to have too much detail.”
Jennie appears to value “level of detail” as a significant feature of a GenAI platform, and does not mention accuracy.
Because My AI is located within the messaging interface of Snapchat, conversations with the chatbot appear alongside chats with other friends, suggesting that engaging with My AI can be treated like a conversation with a friend.
As shown in Fig.
1, Jennie's initial conversation with My AI begins with the chatbot being positioned as a friend (“You are friends with Smart One”).
Smart One is the name that Jennie chose for this “friend” which initiates the conversation by sending her a “snap” (an image that appears only once and is reflected here as having been “Opened” by Jennie).
Represented by a humanlike avatar, Smart One also uses emojis and addresses Jennie by name (“Sure, Jennie, I'd be happy to write you a poem”, mimicking human-like interaction.
Implicated in these features, Jennie often draws parallels between her interactions with My AI and the way she would interact with a human teacher or friend.
For instance, she describes using My AI to help brainstorm ideas for a poem as similar to what she would do in class with a teacher.
She also perceives a degree of personalization in the chatbot's responses, stating “I feel like it has a little bit, builds onto your personality more.”
She draws comparison between Google and Snapchat My AI, highlighting the different interaction patterns.
| Excerpt 2 | |
|---|---|
| Jennie: | I mean Google can like help sometimes but it's pretty like a robot. Like there's no … you can't really ask it to write you a poem kind of thing? Like it won't have emotion or anything? Or like an essay. Like Google can't write the literature things, but it can help you with like the scientific and mathematical ones more than on the AI. |
No paragraph-level conversations.
Start one.
No paragraph-level conversations.
Start one.
No paragraph-level conversations.
Start one.
No paragraph-level conversations.
Start one.
In Excerpt 2, Jennie contrasts searching for information using Google and asking AI to generate creative texts, such as poems and essays.
She believes Google is like a “robot” that lacks “emotion,” implying that she perceives My AI is more than a machine, and can express emotion.
This anthropomorphization of AI is tied to its capacity to create literary texts like poems that express emotion, and is affirmed by the positioning of the chatbot as a friend.
In this case, the chatbot's avatar, use of emojis, and pattern of referring to Jennie by her name appear to shape the way she engages with the tool i.e. an interaction that resembles human conversation.
| Excerpt 3 (Looking at the poem in Fig. 2) | |
|---|---|
| Researcher: | So are you satisfied with their … |
| Jennie: | It's better than something I could write. |
| Researcher: | Why do you think so? |
| Jennie: | Well, because, like, I would just write something super simple. I may be writing about eating pizza for lunch. |
| I wouldn't think of gentle raindrops fall quenching. With each drop life flows nature's call. That would never cross my mind to write down. It wrote it down. | |
No paragraph-level conversations.
Start one.
No paragraph-level conversations.
Start one.
No paragraph-level conversations.
Start one.
No paragraph-level conversations.
Start one.
No paragraph-level conversations.
Start one.
Excerpt 3 demonstrates how Jennie perceives AI-generated text as superior.
Jennie passively accepts the authority of the tool as writer and does not explore other ways in which this poem could be written.
The AI output is perceived not only as more sophisticated, but as something she would not be capable of (“would never cross my mind to write down”) and she does not acknowledge her own agency to create something of good quality on her own.
This perception raises critical questions about how learners like Jennie decide what counts as good writing and how they can default to AI-generated output as “better” (Bartsch et al., 2025).
Without an awareness of the way AI systems are trained on vast corpora of human writing, Jennie does not recognize that such texts are based on statistical predictability rather than creativity or originality.
Particularly, the fact that Jennie uses her phone and the platform Snapchat My AI to approach GenAI shapes the patterns of her interaction with AI.
Jennie attributes social conventions to Snapchat My AI, treating it similarly to texting a friend.
Snapchat My AI uses the same interface Jennie uses for texting friends.
She points out, “You can just text them [Snapchat My AI]” and called the prompts to My AI “chat.”
On Snapchat, a user is notified if someone takes a screenshot of their chat.
When she takes screenshots of My AI's responses, she feels the same social awkwardness she would if she were screenshotting a human friend's message.
She says, “if you're texting your friends and then you take a screenshot, it will tell them,” underscoring how the platform's features prompt her to treat My AI as a human interlocutor.
Even though she insists “[she] would never consider it a friend,” the overlapping features (sending and receiving emojis, receiving screenshot notifications) contribute to the sense that My AI is a participant in a familiar, human-style interaction through its cultures-of-use (Thorne, 2016).
When Jennie used My AI for her Social Studies assignment on the Fall of Rome, she recognized that the tool's conversational tone did not align with academic conventions.
Fig.
3 is a screenshot of her interaction with My AI.
| Excerpt 4 | |
|---|---|
| Researcher: | What did you think about the answer? [looking at Fig. 3] |
| Jennie: | I didn't like this answer, and that's why I said, ‘can you write it like a teacher would and then put a little bit more information in’? And then I was like, okay, that's pretty good. And then this [Fig. 3] was like the best one because it has numbers and like … Like real facts in it, not just like … rough idea of what it was. |
| Researcher: | What was the thing that you didn't like? |
| Jennie: | Because no one says things like, ‘it was a real bummer’. You know, like kids, so if you were giving that to a teacher, it would know. And too big for its britches, I feel like that's kind of British. Britches is like pants in British, I think. |
| […] | |
| Researcher: | Can I ask why you said a teacher would? [looking at Fig. 3] |
| Jennie: | Because teachers would like their own writing, I feel. They would have to like their own writing. And like, it has more of like, because this has more slang that even no one uses, really. But this has more straight to you |
| Researcher: | And which part you were not satisfied with? Because you did another one [Fig. 3]? |
| Jennie: | I did another one [Fig. 3] because it just says, you know, Rome, this mighty empire once ruled over vast territories. Yeah, who ruled it, though, you know? And, like, you know, and just says, it doesn't explain the conflicts, it just says that there were conflicts. |
No paragraph-level conversations.
Start one.
No paragraph-level conversations.
Start one.
No paragraph-level conversations.
Start one.
No paragraph-level conversations.
Start one.
No paragraph-level conversations.
Start one.
As demonstrated in Fig.
3, because My AI is positioned as a “friend”, it responds to Jennie's request using a casual and conversational tone (“sure thing, mate” or “you see”) that does not align with Jennie's intention to get an answer that would meet the requirements of her school assignment.
She did not accept the initial response, and through iterative prompting (“write it like a teacher would”, “write it with more facts”) she was able to get a response that met her satisfaction.
Jennie tried to evaluate GenAI responses based on the inclusion of specific, factual information such as “numbers” or “real facts”, using “teacher's language.”
Through iterative prompting, Jennie was finally able to get an output that maintained an academic register.
However, her idea of the “best” answer was based on the style of the response, and the inclusion of “numbers”, without verifying these “real facts” that were presented to her.
Instead, she immediately assumes the accuracy of these answers based on the linguistic conventions that she recognizes as legitimate.
Jennie enacts agency by controlling the direction of the conversation through iterative prompting, and by achieving her intention.
However, she cedes agency by immediately accepting the output of the tool as valid, and positioning the tool as an unquestionable authority (“Smart One”).
Her preoccupation with register and tone rather than accuracy of content, together with her disposition towards GenAI as “smarter,” seems shape her practices.
Interacting with My AI within the confines of Snapchat that occupies the entire screen of her phone, Jennie does not move to other spaces like Google to engage in lateral reading (Darvin, 2023) because this process requires exiting the app.
Users are not able to compare My AI's answers with other sources side by side on the phone.
In this case, the material differences of how a platform is accessed shapes the extent to which this learner is able to engage critically with generated texts.
Ranjith is a Grade 9 student who moved from Colombo, Sri Lanka six months ago.
He speaks English and Sinhala at home, but since moving to Canada, he has gradually shifted to using English more frequently with his family.
He found the language transition relatively smooth.
Ranjith possesses both a personal HP laptop and a phone and uses them for different purposes.
Additionally, he has access to his father's Microsoft subscription account for certain software and resources.
He relies on his phone for simple tasks such as searching for information, watching short videos or translating words, while his laptop is reserved for more complex activities, such as web design.
In terms of his engagement with GenAI technologies, Ranjith uses ChatGPT or Copilot on his phone for quick searches while at school.
However, when working on personal projects, he would use multiple platforms including Dall.E, Bing Chat, Copilot, or ChatGPT to generate multimodal texts, edit code, or design logos for websites using his personal laptop.
He described how ChatGPT and Copilot were particularly useful for his academic needs in subjects like English or Social Studies.
He mentioned how he used ChatGPT to figure out how to write a good essay for a class assignment by entering the assessment rubric provided by the teacher and asking the tool to provide specific writing tips.
This illustrates that Ranjith turned to GenAI as an intermediary for knowledge translation rather than seeking clarification from a teacher.
However, at the same time, he explicitly said that he did not ask for revising or improving essays because “Chat GPT doesn't improve your knowledge.
It just gives you a shortcut that's not good.
You don't learn anything.”
This statement reflects his disposition toward GenAI as a tool that provides information or interprets a text (“a shortcut”), but is still limited.
Unlike some participants who passively accept AI-generated texts as accurate, Ranjith acknowledges that AI does not inherently lead to learning unless the user actively engages with the content.
Ranjith also described moments when ChatGPT 3.5 produced random or irrelevant prompts and “messed up” the continuity of his story pages, causing frustration: “Sometimes when I asked it to continue, it just started generating random prompts.”
Recognizing these shortcomings led him to seek more advanced models (GPT-4) or alternative platforms (Copilot).
In so doing, Ranjith gained awareness that GenAI is not infallible and learned to approach it with caution, particularly for more complex tasks.
This pattern of behavior suggests that a key driver for this learner in developing critical dispositions toward GenAI technologies is encountering GenAI limitations firsthand.
How learners recognize these limitations however appears to be dependent on their own knowledge of legitimate texts.
He also used Copilot for an in-class Social Studies activity that involved researching about the cause of conflict between the French and English in the Ohio River Valley in the U.S. As shown in Fig.
4, Copilot provided direct website references, allowing Ranjith to trace sources of information more easily.
In contrast to Jennie, who accepted output without verification, Ranjith actively accessed embedded links to assess the accuracy of Copilot's response.
He found Copilot's GPT-4 integration provided “better answers,” while ChatGPT 3.5 still necessitated users to explicitly request for sources that it referred to.
In this case, the hyperlinks that appear at the end of Copilot's response framed as a way to “Learn more” (See Fig.
4) and the hover link preview i.e. the pop-up that appears when Ranjith mouses over keywords serve as sociotechnical structures (Darvin, 2023) that encourage lateral reading.
Because he accesses Copilot on a browser on his laptop, he is able to move across different tabs to verify information, an affordance that is not available when accessing GenAI through an app on a phone (as in the case of Jennie with My AI).
Throughout the interview, he would continually compare ChatGPT and Copilot in terms of their affordances and constraints as in Excerpt 5.
| Excerpt 5 | |
|---|---|
| Researcher: | Okay. Why did you decide to use Copilot particularly in this situation? |
| Ranjith: | It has GPT-4, ChatGPT, you need to pay for GPT-4, but in Copilot, you get it for free. […] |
| Researcher: | Okay. But in this case, you decided to use this because you wanted GPT-4. |
| Ranjith: | Yeah. |
| Researcher: | Because what for you is different about GPT-4 compared to ChatGPT? |
| Ranjith: | So it's supposed to be better and faster and like work better, give better answers, and it should know more than GPT-3.5, which is what we mainly use in OpenAI. |
| Researcher: | And you've seen that? Have you been able to compare answers? |
| Ranjith: | Yeah, so like, I made a bot kind of thing from Arduino. And so like when I pasted the code from OpenAI, the code was like a bit off and it didn't have all the functionality as like BingAI. |
| Researcher: | So the code that Copilot provides was 100 % useful? |
| Ranjith: | Not 100 %, like 80 %. |
| Researcher: | Whereas the one from GPT 3.5 was completely off? |
| Ranjith: | 40 %? |
No paragraph-level conversations.
Start one.
No paragraph-level conversations.
Start one.
No paragraph-level conversations.
Start one.
No paragraph-level conversations.
Start one.
In Excerpt 5, Ranjith compares AI-generated code across platforms, suggesting an awareness of AI's varying reliability.
Rather than treating AI as a solution-provider, Ranjith tested its limitations, verified output and actively shaped how AI fits into his learning process.
At the same time, Ranjith's choice to use Microsoft Copilot over OpenAI's ChatGPT was driven by his belief that newer models are inherently more accurate as he said, “so it's supposed to be better and faster (…) it should know more than GPT-3.5” Instead of passively consuming AI-generated output, he actively evaluated information provided by the tool and recognized inaccuracies.
However, despite recognizing AI's inconsistency, he still placed considerable trust in its functionality, particularly for coding tasks.
Ranjith's interactions with these tools suggest a developing critical awareness of both the affordances and the constraints of these tools.
Ranjith also actively used GenAI technologies to build his “Choose Your Own Adventure” website involving a narrative that he had written about a person lost in a mansion and that included various possible story arches.
To generate images using Dall.E and Microsoft Designer to help tell the story (Fig.
5), he inputted parts of his narrative to ChatGPT so that it could generate prompts for image creation in the other platforms (Fig.
6).
Explaining why he chose this two-step process, he says, “I used OpenAI to generate a prompt because I didn't know what a good prompt would be because all the prompts I tried, they either get blocked or they don't work like they're supposed to.”
When Ranjith realized that Copilot would not generate images from prompts that included words that suggested violence like “blood” or “dying,” which were parts of his stories, he used ChatGPT to generate prompts for image generation to avoid such words.
In this process, he became aware that GPT-3.5 did not filter violent words while GPT-4.0 filtered such words.
Similarly, he also found that GPT-4.0 did not generate what he referred to as “mean” phrases while GPT-3.5 could generate such phrases.
This awareness of the differences between platforms in terms of what is considered permissible language use indicates Ranjith's critical digital literacy, recognizing how platform designs and algorithms filter input and generate output.
Ranjith devised his own strategies to compare and contrast various platforms, tailoring their use to meet his specific goals.
When Ranjith engaged with more challenging and complex tasks, he demonstrated a critical disposition by identifying errors and limitations in responses and evaluating the strengths and weaknesses of different platforms.
While earlier studies on GenAI have primarily focused on its learning affordances and pedagogical potential (Li et al., 2024; Lo, 2023), this study adopts a critical digital literacy (CDL) lens to examine how power operates in human-AI interactions, and the extent to which learners demonstrated a critical awareness of the inequalities and ideologies that shape these interactions.
The comparative analysis of the cases of Jennie and Ranjith reveals that learners demonstrate varying degrees of CDL when interacting with GenAI tools.
Such differences are shaped by their dispositions towards these technologies and their sense of agency, which in turn appear to be constituted by their access to material resources and their capacity to negotiate platform designs.
Access to material resources.
Examining the GenAI practices of the learners through a critical sociomaterial lens enabled an understanding of how material resources such as devices (laptop vs. phone) and platform subscriptions mediated learners’ practices and dispositions toward GenAI (Darvin, 2023) in unequal ways.
Because of limited resources, Jennie primarily used My AI, embedded in a free social media platform and accessible through a mobile phone.
The use of My AI which ran on GPT-3.5 limited her access to affordances offered only by advanced GenAI models.
Accessing My AI through a mobile app appears to discourage her from accessing other online spaces like Google to verify information.
In contrast, Ranjith who had his own laptop and access to advanced GPT models engaged with multiple platforms, leveraging their diverse functionalities to support his learning and creative projects.
His access to multiple platforms enabled him to compare their outputs, explore a wider range of GenAI functionalities, and engage in more autonomous language learning through digital multimodal composition (Jiang & Hyland, 2025).
This broader digital repertoire enables him to navigate GenAI technologies with greater agency and a critical awareness of the affordances and constraints of these tools.
Negotiation of platform design.
By recognizing the capacity of platform design to steer human-AI interactions in particular ways, the findings have demonstrated how sociotechnical structures like suggested prompts or conversation starters (e.g. “You are friends with Smart One” vs. “Ask anything”) further mediate GenAI practices (Bucher, 2018; Darvin, 2023).
From a sociotechnical lens, architecture of platforms can afford and constrain various online actions through linguistic and semiotic forms as “sites of display” (Jones, 2017, p.139).
These structures appear to shape dispositions towards GenAI, whether as a tool that provides immediate answers to one's queries, or an anthropomorphized collaborative partner, or even a helpful friend (Darvin, 2025b).
Jennie's platform, Snapchat My AI is embedded within a social messaging app, and its design features an avatar and a welcoming statement that position the learner as a friend.
The chat interface, enriched with emojis, stickers and conversational prompts, fosters casual and personalized interactions and stimulate friendship.
These design features shape her dispositions toward the AI as human-like and helpful, while she characterizes Google as a “robot” devoid of emotional connection.
The conversational interaction also influences how she responds to the output.
While she engages in iterative prompting for the tool to generate satisfactory outputs, she positions GenAI as an authority, naming the chatbot “Smart One” and assuming that the output is accurate.
Unlike Jennie, who relies primarily on one platform, Ranjith navigates diverse platforms that emphasize productivity and utility, such as ChatGPT, Copilot, Midjourney, or DALL-E, and tests their affordances and constraints.
These platforms have minimalist designs and prompts like “How can I help?,” implying their task-driven functionalities.
Rather than interacting with the tool in a way that simulates human conversation, Ranjith uses prompts as commands and views the interaction as just a series of input and output.
He demonstrates his dispositions toward GenAI as a tool and machine to facilitate his learning or design process.
The platforms' functionalities such as attaching photos or files and generating multimodal outcomes also enable him to generate images and logos and improve his coding skills to complete his web-designing project.
Additionally, features that some platforms offered such as Copilot's linked sources nudged him toward critical evaluation and verification of generated output.
Learner dispositions and agency.
By recognizing how unequal access to resources shape learners' negotiation of platform design, the findings also reveal how these material differences are mutually constitutive of their dispositions, and the ways they exercise agency in human-AI interactions.
Jennie's limited access to devices and a variety of platforms, together with her positioning of My AI as the Smart One seemed to contribute to an acceptance of the information that the tool provided.
This tendency to adopt a passive stance as a receiver of information rather than that of a critical reader is also demonstrated in her recognition of AI-generated poems as superior to her own writing.
By often accepting AI-generated outputs without questioning underlying assumptions, verifying information, or modifying generated texts to maintain her own voice, Jennie tends to cede agency to GenAI.
By contrast, Ranjith demonstrates more developed critical digital literacies.
His goal-oriented disposition towards GenAI, supported by access to a laptop and paid subscriptions, encourages more agentive engagement.
He navigates multiple platforms and leverages their diverse functionalities to support his learning and creative projects.
The task-driven interfaces and extended functionalities, such as file uploads or multimodal generation, fosters his digital multimodal composition.
Recognizing how GenAI supports a wide range of functionalities that serve a learner's different intentions (Darvin, 2025a), Ranjith leverages GenAI for multiple purposes that spans the compositional (producing images, logos, and codes), informational (retrieving information about historical events), evaluative (producing evaluative rubrics), and modificative (revising codes).
His diverse engagement with GenAI tools created opportunities to test its boundaries and critically interrogate his practices.
As he tries to command more complex tasks, he builds his awareness that these tools make mistakes, and their answers are controlled by algorithms.
In doing so, he exercised his agency in aligning GenAI's capacities with his intended goals.
He identifies discrepancies between GPT-3.5 and GPT-4, questions the reliability, and recognizes the algorithmic nature of AI responses.
His capacity to verify, compare, and evaluate reflects a stronger enactment of CDL.
Moreover, while both learners engaged in informational searches, Ranjith's use of Copilot, which included embedded source links, encouraged him to verify information, an evaluative step Jennie did not take.
However, while he demonstrates a more critical approach, his trust in AI remains tied to model sophistication; he assumes GPT-4 is inherently superior to GPT-3.5 without fully interrogating the biases embedded in these systems.
Maintaining such assumptions demonstrates how even with access to a wider range of material resources, youth can enact CDL practices that are still limited in some way.
The findings also suggest that the nature of the tasks learners engage in can shape the extent to which they invest in agentive GenAI literacy practices.
Ranjith pursues creative, self-initiated personal projects such as website development and fiction writing, and this participation in the digital wilds appears to encourage sustained exploration of GenAI affordances.
In contrast, Jennie's engagement is largely for schoolwork, with tasks focused on retrieving information or producing texts.
The differences in the nature and purpose of the task appear to shape the desire to explore this “AI wilderness” (Alm, 2024), echoing research that asserts how the interest-driven nature of the digital wilds translates into investment in a wider range of literacy practices (Liu & Darvin, 2024).
By negotiating their intentions with platform designs and algorithmic processes, the two cases in the study illustrate how agency in interactions with GenAI is distributed and emerges from a complex assemblage of human, technological, and contextual elements (Godwin-Jones, 2024).
To foster and enact CDL, however, learners need to resist ceding agency to the tool, to recognize its limitations, and to assert their own intentions.
This critical stance aligns with Liu et al.’s (2025) view that agency and CDL are interactive and mutually constitutive.
Learners do not simply use GenAI.
They can also be positioned by this tool and steered towards particular interactions.
At the same time, learners can resist such positioning, and how they are able to recognize and navigate these material and ideological conditions shape their development of CDL.
Pedagogical implications These findings underscore the necessity of integrating CDL into education in support of learners’ GenAI use.
To cultivate such literacies, educators must move beyond teaching functional aspects and explicitly address the inequalities and ideological underpinnings of GenAI use.
This critical stance includes:
-
●
Encouraging learners to compare different GenAI platforms to observe variations in responses and to test reliability.
-
●
Encouraging learners to observe and interrogate algorithmic patterns, verify generated output, and identify and challenge biases embedded in GenAI technologies.
-
●
Supporting lateral reading and source verification practices through external sources, especially for informational tasks (Darvin, 2025a).
-
●
Inviting students to reflect on the differences between human- and AI-generated outputs, fostering awareness of authenticity and voice.
-
●
Highlighting the importance of learners' own “imperfect but authentic” expressions over machine-generated fluency (Du et al., 2025), thereby supporting learner agency.
-
●
Creating opportunities to compose various forms of meaning-making, including digital multimodal composition, to expand learners' creative and critical engagement (Jiang & Lai, 2025)
Ultimately, CDL involves equipping learners with critical awareness to interrogate how AI systems reproduce dominant ideologies, marginalize voices, and influence what counts as legitimate knowledge (Darvin, 2023; Vassel et al., 2024).
By examining the interplay of device constraints, platform design, and user agency, we can better understand how digital interactions shape and potentially limit the ability to engage critically with AI technologies.
This aligns with Darvin's (2023) argument that CDL should empower learners to recognize and resist ideological structures embedded in digital tools and enable more agentive, equitable and inclusive GenAI practices.
This study contributes to ongoing discussions on how learners' CDL in their use of GenAI is constructed by examining the ways in which access, platform design and learner agency influence learners' engagement with AI-generated content.
However, this study has several limitations.
First, most of the observations on learners’ use of GenAI occurred outside authentic use environments and were limited to interviews and self-reported screenshots.
Because of this limitation, we suggest future research can incorporate ethnographic studies of how learners interact with AI in their everyday digital practices, particularly in educational settings.
Additionally, a larger or more diverse sample would provide insights into how different sociocultural and linguistic backgrounds shape GenAI engagement.
As Gen AI interactions are contextual and situated, shaped by devices, platforms, and evolving cultures-of-use, this study underscores the need for integrating critical digital literacies in GenAI-mediated language learning.
Such an approach can help learners become active agents in human-AI interactions, while also encouraging teachers to reflect on how these platforms shape knowledge production and educational practices.
CRediT authorship contribution statement
Kyu Yun Lim: Writing – original draft, Formal analysis, Data curation, Conceptualization.
Ron Darvin: Writing – review & editing, Supervision, Project administration, Methodology, Investigation, Funding acquisition, Data curation, Conceptualization.
This research was supported by the UBC Hampton Research Endowment Fund [Grant ID: GR025496].
References
- Alm, 2024A. AlmExploring autonomy in the AI wilderness: Learner challenges and choicesEducation Sciences, 14 (12) (2024), p. 1369https://doi.org/10.3390/educsci14121369CrossrefView in ScopusGoogle Scholar
- Avila and Pandya, 2013J. Avila, J. PandyaTraveling, textual authority, and transformation: An introduction to critical digital literaciesJ. Avila, J. Pandya (Eds.), Critical digital literacies as social praxis: Intersections and challenges, Peter Lang (2013), pp. 1-14View at publisherCrossrefGoogle Scholar
- Baek et al., 2024C. Baek, T. Tate, M. WarschauerChatGPT seems too good to be true”: College students’ use and perceptions of generative AIComputers and Education: Artificial Intelligence, 7 (2024), p. 100294https://doi.org/10.1016/j.caeai.2024.100294View PDFView articleView in ScopusGoogle Scholar
- Bartsch et al., 2025A. Bartsch, C. Neuberger, B. Stark, V. Karnowski, M. Maurer, C. Pentzold, C. SchemerEpistemic authority in the digital public sphere: An integrative conceptual framework and research agendaCommunication Theory, 35 (1) (2025), pp. 37-50https://doi.org/10.1093/ct/qtae020View at publisherCrossrefView in ScopusGoogle Scholar
- Bonner et al., 2023E. Bonner, R. Lege, E. FrazierLarge language model-based artificial intelligence in the language classroom: Practical ideas for teachingTeaching English with Technology, 23 (1) (2023), pp. 23-41https://doi.org/10.56297/BKAM1691/WIEO1749Google Scholar
- Braun & Clarke, 2006V. Braun, V. ClarkeUsing thematic analysis in psychologyQualitative Research in Psychology, 3 (2) (2006), pp. 77-101https://doi.org/10.1191/1478088706qp063oaView at publisherCrossrefView in ScopusGoogle Scholar
- Bucher, 2018T. BucherIf... then: Algorithmic power and politics, Oxford University Press (2018)https://doi.org/10.1093/oso/9780190493028.001.0001Google Scholar
- Chen et al., 2023Y. Chen, S. Jensen, L.J. Albert, S. Gupta, T. LeeArtificial intelligence (AI) student assistants in the classroom: Designing chatbots to support student successInformation Systems Frontiers, 25 (1) (2023), pp. 161-182, 10.1007/s10796-022-10291-4Google Scholar
- Chiu, 2023T.K. ChiuThe impact of generative AI (GenAI) on practices, policies and research direction in education: A case of ChatGPT and MidjourneyInteractive Learning Environments, 32 (10) (2023), pp. 1-17https://doi.org/10.1080/10494820.2023.2172044View in ScopusGoogle Scholar
- Crawford, 2021K. CrawfordAtlas of AI: Power, politics, and the planetary costs of artificial intelligenceYale University Press (2021), 10.2307/j.ctv1ghv45tGoogle Scholar
- Darvin, 2017R. DarvinLanguage, ideology, and critical digital literacyS. Thorne, S. May (Eds.), Language, education and technology (encyclopedia of language and education, vol. 9, Springer (2017), pp. 17-30CrossrefGoogle Scholar
- Darvin, 2023R. DarvinSociotechnical structures, materialist semiotics, and online language learningLanguage, Learning and Technology, 27 (2) (2023), pp. 28-45https://doi.org/10.64152/10125/73502CrossrefView in ScopusGoogle Scholar
- Darvin, 2025aR. DarvinThe need for critical digital literacies in generative AI-mediated L2 writingJournal of Second Language Writing, 67 (2025), Article 101186View PDFView articleView in ScopusGoogle Scholar
- Darvin, 2025bR. DarvinIdentity and investment in the age of generative AIAnnual Review of Applied Linguistics, 45 (2025), pp. 10-27CrossrefGoogle Scholar
- Donley, 2024K. DonleyTeaching with ChatGPT as a linguistically responsive tool for multilingual learnersTechnology in Language Teaching & Learning, 6 (3) (2024)1719-1719https://doi.org/10.29140/tltl.v6n3.1719Google Scholar
- Du et al., 2025Q. Du, D.M. Gross, P. HongVoice, agency and style: What goes missing when AI chats backTimes Higher Education (2025)https://www.timeshighereducation.com/campus/voice-agency-and-style-what-goes-missing-when-ai-chats-backGoogle Scholar
- Duff, 2014P. DuffCase study research on language learning and useAnnual Review of Applied Linguistics, 34 (2014), pp. 233-255, 10.1017/S0267190514000051View in ScopusGoogle Scholar
- Fenwick, 2015T. FenwickSociomateriality and learning: A critical approachD. Scott, E. Hargreaves (Eds.), The SAGE handbook of learning, SAGE (2015), pp. 83-93CrossrefGoogle Scholar
- Garcia and de Roock, 2021A. Garcia, R.S. de RoockCivic dimensions of critical digital literacies: Towards an abolitionist lensPedagogies: An International Journal, 16 (2) (2021), pp. 187-201, 10.1080/1554480X.2021.1914058View in ScopusGoogle Scholar
- Godwin-Jones, 2024R. Godwin-JonesDistributed agency in language learning and teaching through generative AILanguage, Learning and Technology, 28 (2) (2024), pp. 5-31https://doi.org/10.64152/10125/73570CrossrefView in ScopusGoogle Scholar
- Guan et al., 2024L. Guan, E.Y. Zhang, M.M. GuExamining generative AI–mediated informal digital learning of English practices with social cognitive theory: A mixed-methods studyReCALL (2024), pp. 1-17, 10.1017/S0958344024000259Google Scholar
- Han, 2024Z. HanChatGPT in and for second language acquisition: A call for systematic research. Studies in Second LanguageAcquisition, 46 (2) (2024), pp. 301-306https://doi.org/10.1017/S0272263124000111View in ScopusGoogle Scholar
- Hellmich et al., 2024E.A. Hellmich, K. Vinall, Z.M. Brandt, S. Chen, M.M. SparksChatGPT in language education: Centering learner voicesTechnology in Language Teaching & Learning, 6 (3) (2024)1741-1741https://doi.org/10.29140/tltl.v6n3.1741Google Scholar
- Higgs & Stornaiuolo, 2024J.M. Higgs, A. StornaiuoloBeing human in the age of generative AI: Young people’s ethical concerns about writing and living with machinesReading Research Quarterly, 59 (4) (2024), pp. 632-650https://doi.org/10.1002/rrq.552CrossrefView in ScopusGoogle Scholar
- Jewitt, 2017C. JewittMultimodal analysisA. Georgakopoulou, T. Spilioti (Eds.), The Routledge handbook of language and digital communication, Routledge (2017), pp. 69-84https://doi.org/10.4324/9781315694344Google Scholar
- Jiang and Hyland, 2025F. Jiang, K. HylandDoes ChatGPT argue like students? Bundles in argumentative essaysApplied Linguistics, 46 (3) (2025), pp. 375-391, 10.1093/applin/amae052View in ScopusGoogle Scholar
- Jiang and Lai, 2025L. Jiang, C. LaiHow did the generative artificial intelligence‐assisted digital multimodal composing process facilitate the Production of quality digital multimodal compositions: Toward a process‐genre integrated modelTesol Quarterly (2025), 10.1002/tesq.3390Google Scholar
- Jones, 2017R. JonesTechnology and sites of displayC. Jewitt (Ed.), The Routledge handbook of multimodal analysis, Routledge (2017), pp. 139-151Google Scholar
- Jones, 2021R.H. JonesThe text is reading you: Teaching language in the age of the algorithmLinguistics and Education, 62 (2021), p. 100750https://doi.org/10.1016/j.linged.2019.100750View PDFView articleView in ScopusGoogle Scholar
- Jones, 2022R.H. JonesCommentary: Critical digital literacies as action, affinity, and affectTesol Quarterly, 56 (3) (2022), pp. 1074-1080https://doi.org/10.1002/tesq.3153CrossrefView in ScopusGoogle Scholar
- Jones, 2025R.H. JonesCulture machinesApplied Linguistics Review, 16 (2) (2025), pp. 753-762https://doi.org/10.1515/applirev-2024-0188CrossrefView in ScopusGoogle Scholar
- Kang & Yi, 2023J. Kang, Y. YiBeyond ChatGPT: Multimodal generative AI for L2 writersJournal of Second Language Writing, 62 (2023), p. 101070https://doi.org/10.1016/j.jslw.2023.101070View PDFView articleView in ScopusGoogle Scholar
- Kaplan-Rakowski et al., 2023R. Kaplan-Rakowski, K. Grotewold, P. Hartwick, K. PapinGenerative AI and teachers’ perspectives on its implementation in educationJournal of Interactive Learning Research, 34 (2) (2023), pp. 313-338https://doi.org/10.70725/815246mfssgpCrossrefView in ScopusGoogle Scholar
- Kelly-Holmes, 2024H. Kelly-HolmesArtificial intelligence and the future of our sociolinguistic workJournal of SocioLinguistics, 28 (2024), pp. 3-10, 10.1111/josl.12678View in ScopusGoogle Scholar
- Kessler & Casal, 2024M. Kessler, J.E. Casal(Socio)linguistics and generative AI: Taking the reins as researchers and steering its use toward ethical outcomesJournal of SocioLinguistics, 28 (2024), pp. 31-34https://doi.org/10.1111/josl.12682CrossrefView in ScopusGoogle Scholar
- Kohnke et al., 2023L. Kohnke, B.L. Moorhouse, D. ZouChatGPT for language teaching and learningRELC Journal, 54 (2) (2023), pp. 537-550https://doi.org/10.1177/00336882231162868CrossrefView in ScopusGoogle Scholar
- Leander & Burriss, 2020K.M. Leander, S.K. BurrissCritical literacy for a posthuman world: When people read, and become, with machinesBritish Journal of Educational Technology, 51 (4) (2020), pp. 1262-1276https://doi.org/10.1111/bjet.12924CrossrefView in ScopusGoogle Scholar
- Lee & Lee, 2024H. Lee, J.H. LeeThe effects of AI-guided individualized language learning: A meta-analysisLanguage, Learning and Technology, 28 (2) (2024), pp. 134-162https://doi.org/10.64152/10125/73575CrossrefView in ScopusGoogle Scholar
- Li et al., 2024B. Li, V.L. Lowell, C. Wang, X. LiA systematic review of the first year of publications on ChatGPT and language education: Examining research on chatgpt's use in language learning and teachingComputers and Education: Artificial Intelligence, 7 (2024), Article 100266http://www.sciencedirect.com/science/article/pii/S2666920X24000699View PDFView articleView in ScopusGoogle Scholar
- Liu & Darvin, 2024G. Liu, R. DarvinFrom rural China to the digital wilds: Negotiating digital repertoires to claim the right to speakTesol Quarterly, 58 (1) (2024), pp. 334-362https://doi.org/10.1002/tesq.3233CrossrefGoogle Scholar
- Liu et al., 2024G.L. Liu, R. Darvin, C. MaExploring AI-mediated informal digital learning of English (AI-IDLE): A mixed-method investigation of Chinese EFL learners’ AI adoption and experiencesComputer Assisted Language Learning (2024), pp. 1-29https://doi.org/10.1080/09588221.2024.2310288View in ScopusGoogle Scholar
- Liu et al., 2025G.L. Liu, J.S. Lee, X. ZhaoCritical digital literacies, agentic practices, and AI-mediated informal digital learning of EnglishSystem, 103797 (2025)https://doi.org/10.1016/j.system.2025.103797Google Scholar
- Lo, 2023L.S. LoThe CLEAR path: A framework for enhancing information literacy through prompt engineeringThe Journal of Academic Librarianship, 49 (4) (2023), p. 102720https://doi.org/10.1016/j.acalib.2023.102720View PDFView articleView in ScopusGoogle Scholar
- Luke, 2012A. LukeCritical literacy: Foundational notesTheory into Practice, 51 (1) (2012), pp. 4-11https://doi.org/10.1080/00405841.2012.636324View at publisherCrossrefView in ScopusGoogle Scholar
- Mah et al., 2024C. Mah, H. Walker, L. Phalen, S. Levine, S.W. Beck, J. PittmanBeyond cheatbots: Examining tensions in teachers’ and students’ perceptions of cheating and learning with ChatGPTEducation Sciences, 14 (5) (2024), p. 500https://doi.org/10.3390/educsci14050500View at publisherCrossrefView in ScopusGoogle Scholar
- Mintz et al., 2023J. Mintz, W. Holmes, L. Liu, M. Perez-OrtizArtificial intelligence and K–12 education: Possibilities, pedagogies and risksComputers in the Schools, 40 (4) (2023), pp. 325-333https://doi.org/10.1080/07380569.2023.2279870View at publisherCrossrefView in ScopusGoogle Scholar
- Moorhouse, 2024B.L. MoorhouseBeginning and first-year language teachers’ readiness for the generative AI ageComputers and Education: Artificial Intelligence, 6 (2024), p. 100201https://doi.org/10.1016/j.caeai.2024.100201View PDFView articleView in ScopusGoogle Scholar
- Pack & Maloney, 2023A. Pack, J. MaloneyPotential affordances of generative AI in language education: Demonstrations and an evaluative frameworkTeaching English with Technology, 23 (2) (2023), pp. 4-24, 10.56297/BUKA4060/VRRO1747View in ScopusGoogle Scholar
- Pangrazio, 2016L. PangrazioReconceptualising critical digital literacyDiscourse: Studies in the Cultural Politics of Education, 37 (2) (2016), pp. 163-174https://doi.org/10.1080/01596306.2014.942836CrossrefView in ScopusGoogle Scholar
- Picton & Clark, 2024I. Picton, C. ClarkChildren and young people’s use of generative AI to support literacy in 2024(2024)National Literacy Trusthttps://literacytrust.org.uk/research-services/research-reports/children-young-people-and-teachers-use-of-generative-ai-to-support-literacy-in-2024/Google Scholar
- Pötzsch, 2021H. PötzschBringing materiality into thinking about digital literacy: Theories and practices of critical education in a digital age., Brill (2021), pp. 52-73Critical digital literacies: Boundary-crossing practiceshttps://doi.org/10.1163/9789004467040_004CrossrefView in ScopusGoogle Scholar
- Ramesh et al., 2023K. Ramesh, S. Sitaram, M. ChoudhuryFairness in language models beyond English: Gaps and challenges(2023)arXiv preprint arXiv:2302.12578Google Scholar
- Shieh et al., 2024E. Shieh, F.M. Vassel, C. Sugimoto, T. Monroe-WhiteLaissez-faire harms: Algorithmic biases in generative language models(2024)arXiv preprint arXiv:2404.07475Google Scholar
- Smith et al., 2025B.E. Smith, A.Y. Shimizu, S.K. Burriss, M. Hundley, E. PendergrassMultimodal composing with generative AI: Examining preservice teachers' processes and perspectivesComputers and Composition, 75 (2025), Article 102896, 10.1016/j.compcom.2024.102896View PDFView articleView in ScopusGoogle Scholar
- Song & Song, 2023C. Song, Y. SongEnhancing academic writing skills and motivation: Assessing the efficacy of ChatGPT in AI-assisted language learning for EFL studentsFrontiers in Psychology, 14 (2023), p. 1260843, 10.3389/fpsyg.2023.1260843View in ScopusGoogle Scholar
- Thorne, 2016S.L. ThorneCultures-of-use and morphologies of communicative actionLanguage, Learning and Technology, 20 (2) (2016), pp. 185-191https://www.lltjournal.org/item/10125-44473/CrossrefView in ScopusGoogle Scholar
- Vassel et al., 2024F.- M. Vassel, E. Shieh, C.R. Sugimoto, T. Monroe-WhiteThe psychosocial impacts of generative AI harmsProceedings of the AAAI Symposium Series, 3 (1) (2024), pp. 440-447, 10.1609/aaaiss.v3i1.31251View in ScopusGoogle Scholar
- Wang, 2024C. WangExploring students’ generative AI-Assisted writing processes: Perceptions and experiences from native and nonnative English speakersTechnology, Knowledge and Learning (2024), pp. 1-22https://doi.org/10.1007/s10758-024-09744-3Google Scholar
- Wang et al., 2021J. Wang, Y. Liu, X.E. WangAssessing multilingual fairness in pre-trained multimodal representations(2021)arXiv preprint arXiv: 2106.06683Google Scholar
- Warschauer et al., 2023M. Warschauer, W. Tseng, S. Yim, T. Webster, S. Jacob, Q. Du, T. TateThe affordances and contradictions of AI-generated text for writers of english as a second or foreign languageJournal of Second Language Writing, 62 (2023), p. 101071https://doi.org/10.1016/j.jslw.2023.101071View PDFView articleView in ScopusGoogle Scholar
- Woo et al., 2024D.J. Woo, H. Susanto, C.H. Yeung, K. Guo, A.K.Y. FungExploring AI- generated text in student writing: How does AI help?Language, Learning and Technology, 28 (2) (2024), pp. 183-209https://doi.org/10.64152/10125/73577View at publisherCrossrefView in ScopusGoogle Scholar
- Yan, 2023D. YanImpact of ChatGPT on learners in a L2 writing practicum: An exploratory investigationEducation and Information Technologies, 28 (11) (2023), pp. 13943-13967, 10.1007/s10639-023-11742-4View at publisherView in ScopusGoogle Scholar
- Yin, 2014R.K. YinCase study research: Design and methods(5th ed.), Sage (2014)Google Scholar
- Zhang & Tur, 2024P. Zhang, G. TurA systematic review of ChatGPT use in K–12 educationEuropean Journal of Education, 59 (2) (2024), Article e12599https://doi.org/10.1111/ejed.12599View in ScopusGoogle Scholar
General Document Comments 0
on the uploaded document.
0 archived comments