Everyday encounters with deepfakes: young people’s media and information literacy practices with AI-generated media

The development and growing presence of generative artificial intelligence (AI) has reshaped the ways in which information is created, thereby changing the media landscape. Generative AI refers to AI systems that can be utilized to create new content such as videos, audio, images, code, text and simulations, often based on users’ text-based instructions, or “prompts” (Feuerriegel et al., 2024). With recent advancements in large language models ^[1] (LLMs) powered by vast datasets, generative AI applications have become increasingly sophisticated and capable of producing highly human-like content. Generative AI tools such as ChatGPT ^[2] and DALL-E ^[3] have been widely applied to media content creation. In this context, the term “AI-generated media” (AGM) has emerged to describe the media content, in the modalities of text, audio, image and video, that is produced by an AI system (Partadiredja et al., 2020, p. 1).

The applications of AGM in different scenarios have enabled new avenues for innovation. For example, the documentary Welcome to Chechnya attempted to protect people’s identities by employing generative AI algorithms to create faces that replaced the real faces of anti-LGBTQ movement survivors (Danry et al., 2022). Additionally, virtual influencers active on social media platforms, such as Lil Miquela, are reportedly created by generative AI technology (Ferrari and McKelvey, 2023). This application facilitates the customization of a figure’s personality, thereby reducing associated expenses such as costs related to human influencer management or agent fees.

However, AGM that is referred to as “deepfake”, “deep fake”, or “synthetic media” is widely recognized as a potentially risky AI artefact (Westerlund, 2019; Sayler and Harris, 2020; Vaccari and Chadwick, 2020). In accordance with Vaccari and Chadwick (2020, p. 2), the term “deepfake” is broadly used to describe a kind of hyper realistic media, driven by machine learning algorithms, in which someone’s facial expressions are pasted onto another person’s face. These AI-powered artefacts have appeared widely as pornographic, political or humorous media content (Westerlund, 2019). Deepfakes can be used maliciously in disinformation delivery as part of fraud campaigns and bullying. In these contexts, the inauthentic information in AGM content, which can appear strikingly convincing, poses risks to individuals, organizations and society (Vaccari and Chadwick, 2020, p. 2; Langa, 2021). For instance, Chen and Magramo (2024) reported an instance where deepfake technology was employed during a video conference to replicate the voice and likeness of a company’s chief financial officer, deceiving a finance employee and leading to a $25m scam in Hong Kong. Reportedly, individuals have also exploited disinformation generated by deepfakes in attempts to influence election outcomes (Scott, 2024). Beyond direct malicious use, employing this type of AGM is considered problematic as it may sow distrust among audiences. This concern is reflected in the recently adopted European Union (EU) AI Act, which includes transparency requirements for the use of “deepfakes”, defined in the regulation as “AI-generated or manipulated image, audio, or video content that resembles existing persons, objects, places, entities, or events and would falsely appear to a person to be authentic or truthful” (EU AI Act, 2024, Art. 3[60]).

The general public’s increased access to advanced generative AI technologies has led to a tremendous increase in the volume of AGM content and has consequently become more accessible to young people. Many younger people, particularly those belonging to Generation Z, have experienced significant impacts from digitalization; many have actively engaged with various digital platforms from an early age (Swart, 2023). Concerns as to how AGM might impact youth are thus increasing. Young people are regarded as a vulnerable group; as they are still navigating stages of mental, physical and social development, they may lack the skills to handle harmful information encountered online (Espenschied, 2024). It is therefore vital to better comprehend how young people encounter AGM in their everyday lives, including the ways in which they deal with these media.

This study aims to contribute to the knowledge area related to youths’ media and information literacy (MIL) practices in terms of AGM in everyday life. It explores young people’s practices when encountering AGM in their daily lives, focusing specifically on their engagement with and understanding of deepfakes. It thus seeks to address their practices of navigating deepfakes in an everyday setting.

The following research questions are addressed in this study:

The article is structured into six sections. The next section reviews previous research and provides the theoretical background for this study, focusing on generative AI and young people, information encountering in the digital age, and MIL. The Method section outlines our research design and the process of participant recruitment, data collection and analysis. The Results section presents our findings, highlighting young people’s MIL practices in relation to the context of encountering deepfakes, their reactions to these encounters, and their understanding of deepfakes. The Discussion section offers reflections on the findings and is followed by the Conclusion.

Academic discussions surrounding generative AI have surged with the emergence of revolutionary AGM tools such as ChatGPT. Scholars from multiple disciplines have published works discussing different generative AI applications and their impacts on society (e.g. Fischer, 2023; Hirvonen et al., 2023; Lao and You, 2024). Notably, as young people constitute one of the prominent user groups targeted by these advanced technologies (Bozkurt et al., 2023, p. 104), a large body of studies has focused on generative AI and young people, with one strand focusing primarily on young people’s understanding of generative AI. For example, Higgs and Stornaiuolo (2024) conducted a study on high school students’ ethical and critical reflections on generative AI and found that, while young participants acknowledged the construction of human creativity in the context of machine-assisted composition, concerns about AI as a threat to human creativity, anxieties over inauthentic AI-generated content and the potential harms and risks posed by such technologies were frequently raised in discussions. This prior research shows that young people have developed certain understandings of this emerging technology with consideration of its potential consequences, mostly considering its negative aspects.

Another significant branch of existing research focuses on young people’s engagement with generative AI, mainly exploring how they use generative AI technologies. Ali et al. (2024) conducted a workshop with 34 high school students to examine their engagement with text-to-image generation algorithms. The study found that students were able to use prompts effectively with generative AI tools to express their imaginative visions. This process also enhanced users’ understanding of generative AI: following these learning activities, participants reflected on both its benefits and potential harms (Ali et al., 2024). Overall, prior research on young people’s engagement with generative AI has taken place predominantly in structured educational contexts (see, e.g. Putjorn and Putjorn, 2023; Newman et al., 2024). However, there is a notable gap in the research when it comes to understanding young people’s experiences with generative AI outside of these settings.

Additionally, while substantial existing literature focuses on how generative AI assists young people’s content creation, there is relatively less emphasis on their everyday experiences with AI-generated content. This being said, as deepfakes have become more prevalent in online environment in recent years, they have also become the subject of studies. For example, in their experiment examining users’ perceptions of deepfake personas, Kaate et al. (2023) found that participants identified deepfakes primarily by noting the character’s gender-related traits, cultural background, human or non-human characteristics and expressiveness. Such research focuses predominantly on how people detect AGM. However, most prior research has focused on how the general population, rather than young people specifically, engages with deepfakes, often in controlled research settings. Within this research landscape, our study aims to address this gap by focusing on young people in everyday settings.

The advancement of algorithmic systems has greatly increased information flows, often causing unintentional content exposure. Meanwhile, these systems encourage users’ active scanning through micro-targeting and personalization of information (Hirvonen et al., 2023, p. 1156). In this context, many individuals do not encounter information—such as that conveyed through AGM content—through deliberate searching, but rather by coming across it during routine online activities (see, e.g. Janssen et al., 2024; Hameleers et al., 2024). If the encountering of content has become increasingly prevalent in this algorithm-driven information environment, it is essential to address this phenomenon when examining MIL practices.

The incidental acquisition of information, accidental discovery of information, and serendipitous encounters all describe situations where individuals bump into information without actively seeking for it. McKenzie (2003, pp. 26–27) uses the term non-directed monitoring to describe such moments, where individuals come across information unexpectedly—whether through casually observing others, overhearing conversations, or passively engaging with information sources like newspapers with no specific intent. These unplanned encounters reflect a broad spectrum of passive information behaviours also captured in earlier conceptualizations such as Savolainen’s (1995) monitoring the context, Toms’ (1998) chance encounters, Wilson’s (1997) passive attention, Ross’s (1999) finding without seeking, and Erdelez’s (1996) notion of information encountering. Among these, Erdelez’s concept (e.g. 1995, 1996) offers a particularly rich framework for understanding how individuals encounter information in everyday life.

Erdelez has defined “information encountering” as:

Erdelez's (2005, pp. 181–182) model of information encountering encompasses steps that take place during this process, including:

1. Noticing: finding the information potentially relevant to the information user’s problem;

2. Stopping: stopping current search when the information user notices a piece of information related to the need;

3. Examining: analysing the encountered information’s relevancy and accuracy;

4. Capturing: saving the information that is worth saving; and

5. Returning: reverting to the initial information search for the foreground problem.

Subsequent research has sought to further refine and expand upon this model. For example, Awan et al. (2019) examined the information encountering and encountered sharing behaviour of 120 MPhil and PhD research students in an online environment and developed Erdelez’s model (2005) by inserting the process of “keeping” and “sharing” between “capturing” and “returning”. However, as Makri and Buckley (2020) pointed out, these models can be idealized, since the process of addressing encountered information may be influenced by factors such as stress, mood or time pressures and disruptions that can occur at any stage. It is therefore valuable to delve into the specific cases that help researchers grasp the variations between information encounters in different contexts. Inspired by this, this study seeks to explore the nuances and tensions in people’s encounters with AGM within the complex information landscape shaped by emerging technologies, rather than reconstructing the model built by Erdelez (2005).

A portion of existing research has explored people’s experiences with information encountering in everyday settings, with health information often being the focus of such research. While research on information encountering often highlights its positive outcomes—such as aiding in problem-solving (see, e.g. Erdelez, 1999)—previous studies also acknowledge that such encounters can involve misinformation. Positioned in a post-COVID-19 era flooded by mis- and disinformation, Malki et al. (2024) examined women’s serendipitous encounters with health misinformation on social media platforms and found that such encounters were pervasive across various platforms (Malki et al., 2024). These encounters often triggered negative emotional responses and were navigated based on the knowledge participants gained from their personal experiences and social media comments (Malki et al., 2024).

Previous studies have highlighted that encountering has become a significant means by which individuals are exposed to information in today’s online environment (see McKenzie, 2003). For instance, Hassoun et al. (2023) investigated the methods by which young people sought and evaluated online information and found that they tended to rely on encountering information incidentally. Despite these academic contributions, the concept of information encountering has garnered less attention within the field of information behaviour and practice research (e.g. Hakim Silvio, 2006; Jacobs et al., 2017) compared to the perspective of purposive information seeking. This study endeavours to address this research gap by recognizing information encountering as a central situation to young people’s exposure to AGM.

In 2011, Erdelez et al. (2011) identified the potential for incorporating information encountering to the model of information literacy, noting that this opportunistic discovery of information had not been addressed in any previous models. This is an applicable idea also for the integrated concept of MIL which combines the principles of media literacy and information literacy (Frau-Meigs, 2019; Singh and Banga, 2022) and highlights the intertwined nature of media and information.

According to The United Nations Educational, Scientific, and Cultural Organization (UNESCO), MIL is:

Building upon this foundational concept, Grizzle et al. (2014), in collaboration with UNESCO, has expanded the definition of MIL to encompass not only knowledge and skills but also the attitudes necessary for citizens to understand the role of media and information providers in democratic societies. This expanded view emphasizes the importance of critically evaluating content, ethically locating and using information, applying information and communication technology skills, and engaging in self-expression and democratic participation.

Previous research on young people’s MIL has focused primarily on how youths navigate online information, particularly when confronted with mis- and disinformation. Melro and Pereira (2019) concentrated on Portuguese young people’s perception of disinformation in news and pointed out that most participants presented limited critical analysis. In addition, Nazari et al.’s research (2022) covered perspectives of young people’s engagement with fake news, indicating how they evaluated the credibility of different sources and perceived and dealt with fake news (Nazari et al., 2022). Such studies further highlight the importance of critical thinking and concerns towards MIL.

Existing research related to MIL and generative AI has been conducted mostly in the field of education, often focusing on the ways in which generative AI impacts the development of educators’ teaching instruction. For example, Söken and Nygreen (2024) propose three instructional innovations from the perspective of critical media literacy based on their teaching experiences with generative AI: establishing an appropriate tone, designing in-class activities to engage critically with generative AI, and revising assessment and grading policies. However, even with the emergence of generative AI bringing significant changes to various digital infrastructures, often seen as a “social turn” (Frau-Meigs, 2019, p. 10), there has been little exploration of young people’s MIL concerning generative AI artefacts.

Inspired by Lloyd’s perspective (2010), this study views MIL as a practice, within which information encountering can be considered as an integral component. A practice lens pays attention to people’s meaningful “publicly accessible performances”—those which are shaped by social structures and cultural backgrounds (Rouse, 2007, pp. 504–505). In this sense, a practice lens underscores examining actions—composed of complex elements—in particular contexts. Building on this, Lloyd (2010) has argued that information literacy should not be viewed merely as a set of skills or competencies; rather, it should be understood as a complex social information practice occurring within specific contexts. However, this perspective is currently underrepresented in the field of MIL research.

Lloyd’s perspective on applying a practice lens is primarily grounded in Schatzki’s practice theory (2002). According to Schatzki’s articulation, “[a] practice is a set of doings and sayings”; and since these doings and sayings often lead to further actions within the contexts in which they occur, the set of actions extends beyond just its doings and sayings (2002, p. 73). In this explanation emphasis is placed on the contextual elements, stressing that practice is inherently intertwined with the context (Schatzki, 2002, p. 73). Moreover, from this perspective, a practice can be regular or irregular, unique, and constantly changing doings/sayings, tasks (Schatzki, 2002, p. 74), and can be integrative (encompassing multiple actions, project, end, and emotions, e.g. teaching, working) or dispersed (centred on a single type of action, e.g. describing, ordering, questioning) (Schatzki, 2002, p. 88).

In line with these ideas, Lloyd (2010) has argued that:

More specifically, in line with Schatzki (2002), she emphasizes: (1) the complexities of these practices, suggesting that they should not be examined from a single perspective but rather through various social elements that might influence them; (2) the context in which these practices occur; and (3) the role of knowledge and meaning-making within this specific setting (Lloyd, 2010).

From this perspective, a practice lens requires researchers to interpret people’s information-related actions through a sociological and contextual framework (Talja, 2005; Lloyd, 2010). This perspective is equally applicable to studies of MIL. It is essential to consider not only people’s actions but also their practical and embodied understandings of experiential knowledge, which reflect their MIL. By emphasizing this dual focus, this study aims to provide a more nuanced understanding of how young people navigate and interpret their interactions with generative AI and deepfake content in their everyday lives.

The study adopted a qualitative research strategy to investigate young people’s practices with AGM in their everyday life. As such, the research addresses its core questions by “developing an understanding of the meaning and experience dimensions of humans” lives and social worlds’ (Fossey et al., 2002). In this study, the youths’ everyday experiences with AGM, including their understandings of and practices with deepfakes, were interpreted and analysed qualitatively, with in-depth interviews being conducted to collect empirical data. According to Legard et al. (2003, p. 147), this enables researchers to gain, through conversations, an understanding of interviewees’ thoughts, feelings, views and experiences. In our case, it helped us to interpret the “meaning, actions, and context” (Popay et al., 1998, p. 345) of the young people who participated in our research.

Participant recruitment and data collection took place in Finland, a country known for its strong emphasis on citizens’ media literacy (Palsa and Salomaa, 2020). These activities were conducted in 2021 by a research group from University A. For the participants, taking part in the project and data collection was part of their work internship as a specific work practice programme ^[4] required by students’ formal K–9 education. Three senior researchers and a doctoral researcher from the research group delivered a presentation at School B to introduce this internship opportunity to the university and associated research. Twenty-one grade nine students, aged 14–15 years, showed interest in joining. Consent forms were completed by the students who agreed to join the study as well as their legal guardians. They could withdraw their consent at any time without impacting their engagement in the internship.

During the internship week, students had an opportunity to get to know University A as a working environment and thus to better understand the researchers’ work. In addition, various activities related to robotics, AI and business were offered to the participants. These encompassed different themes, such as learning about AI and robotics, thinking about AI-related business plans and taking part in research. This internship took place in a classroom-like setting at University A.

Twenty participants (N = 20) consented to be interviewed and responded to questions following a pre-prepared interview outline that included five main themes: “AI in everyday apps”, “AI and personalized content”, “AI in finding media content”, “synthetic media/AI-generated content” and “impacts of AI on young people”. The present study focuses only on participants’ responses under the “synthetic media/AI-generated content” theme.

When embarking upon the “synthetic media/AI-generated content” section, and to gain an initial understanding of young participants’ experiences with general AGM artefacts, we first asked general questions such as “Do you know about media content that is generated using AI?”, “If so, can you talk about your experiences?” and “Would you like to show an example (an app or video, etc.)?” During this process, the participants were allowed to scroll back their mobile phones to recall memories about their previous experiences and to show us their screens. Subsequently, as part of an experimental section participants were asked to watch two deepfake videos and verify the authenticity of the media content.

Then, we collected data on their experiences with deepfakes in everyday life, this data serving as the empirical material for the research. All conversations were audio recorded with participants’ permission. The interview data were first transcribed verbatim and then coded by the first author, guided by the two research questions, using a combination of inductive and deductive content analysis. Each sentence of the transcript was segmented as a unit of analysis. Transcripts regarding the scenario of the exposure to deepfakes were categorized under “situation” and the medium of the exposure to deepfakes under “platform”, while details about different types of deepfakes were sorted under “content”. For integrating the concept of “information encountering” in this analysis, the analysis moved to a deductive stage at which the categories were grouped under the central theme of “encountering deepfakes”. The empirical data concerning the situations in which young people are exposed to deepfakes, the platforms where this exposure occurs, and the types of deepfake content they come across are organized under the central theme of “encountering deepfakes”, with three categories: “situation”, “platform”, and “content”.

Transcripts related to participants’ ways of characterizing activities with deepfakes as casual actions were categorized as “casual interaction with deepfakes”, while those describing how young people deal with deepfakes and their thoughts on them were categorized as “action on deepfakes”. These two categories were further inductively grouped under the central theme of “responding to deepfakes”.

Additionally, young people’s descriptions of deepfake content were inductively analysed from three perspectives — “content creation”, “impact”, and “AI techniques”—and categorized under the central theme of “understanding deepfakes”. Together, these three central themes are interpreted to represent young people’s MIL practices when encountering deepfakes in their everyday lives. The central themes and categories are summarized in Table 1.

Most of the research participants already had experiences with deepfakes at the time of data collection. Most had not created any media content, nor had they sought out this content purposefully, instead encountering deepfakes unintentionally in their everyday lives. This section unfolds the analysis of young people’s MIL practices from three perspectives: the context of encountering deepfakes (4.1), the reaction to encountered deepfakes (4.2), and understanding deepfakes (4.3).

This section presents the findings in relation to RQ1, pointing to the practices young participants engage in when encountering deepfake content in their everyday lives. These results first explore the context in which young people engage with deepfake content, focusing on the situations in which they encounter deepfakes (4.1.1), the platforms where these encounters take place (4.1.2), and the types of content involved (4.1.3). Given that the young participants were found to often browse information on recommendation-algorithm-driven social media platforms—where deepfakes frequently circulate—they were primarily exposed to deepfakes on these platforms through incidental encounters. The deepfake content encountered by the young participants predominantly includes celebrity, humorous, political, and educational themes.

The results point to the situations in which young participants were exposed to deepfakes. It is important to note that their descriptions reflect their personal perception of deepfakes—it is possible that they may have come across deepfake content without recognizing it as such or may define deepfakes in diverse ways. Most participants were exposed to deepfakes in “not planned or anticipated” ways (Erdelez, 1995, p. 3) in their daily lives, suggesting a pervasive yet passive interaction. Only one participant (F18) explicitly claimed to have never seen a deepfake. B6 expressed uncertainty about accessing deepfakes: “I just skipped them, like I didn’t probably pay much attention to them. But from my recent sites, no [I have not seen them]” (B6). The remaining 18 participants stated that they had encountered deepfakes in their everyday lives.

Most participants claimed that they had not searched for this media content but had encountered it accidentally (Hassoun et al., 2023). They usually noticed the deepfake content without expecting it while browsing online (see Erdelez, 2005). C7’s response to the researcher’s question serves as a representative example:

In addition, A3 highlighted, “Always come across. I don’t search” (A3). Likewise, some respondents (such as E14 and G21) pointed out that they came across these media artifacts in their daily lives. Some participants described encountering deepfakes in more detail. For instance, D11 and F17 underlined their incidental encounters: “Usually they’re just random [content]” (D11) and “Accidentally yes, only [recommended by the system]” (F17).

Some participants who had come across deepfakes reported infrequent encounters with these media. Several participants (such as A2 and C9) expressed that they seldom bumped into these media artifacts. For example, G20 responded, “I don’t feel like I see them often” (G20). Some young students pointed out a concrete frequency, for example, “probably every couple of months” (E13).

Unlike most participants, two young participants mentioned having intentionally searched for deepfakes. As G19 stated, “Intentionally I’ve, I have searched [for them]” (G19). Alongside purposive seeking, G19 also encountered deepfake content incidentally, though not frequently. G19 also expressed having shown it to a friend. This “sharing” (Awan et al., 2019) happened after G19 became interested in deepfake content. However, G19 had seemingly not captured (see Erdelez, 2005; Awan et al., 2019) any preferred deepfake content, attempting to search for an instance on social media rather than providing a saved or bookmarked example. Unlike G19, E13 conducted an intentional search, driven by curiosity, only once upon first hearing the term “deepfake”.

Most of the young people had been exposed to deepfake content created by others, with only a few having attempted to create deepfakes themselves. One deepfake creator, E13, shared their experience: “Yeah, I tried to put a photo of my friend with. Like … tried to put my friend’s face on Mao Zedong ^[5] […]” (E13). E13 added that the artifact did not work well due to a lack of knowledge about deepfake creation. G20 acknowledged having made a deepfake video but did not publicly release it. Most participants claimed that they did not create deepfakes themselves. For example, when asked “Have you tried to make them yourself?”, A3 said “No […], I’m not that good” (A3), pointing to the need for specific capabilities to create them. Others, such as C7 and C8, gave only a brief negative response.

What is significant in these extracts is an indication that deepfakes had already blended into the digital landscape at the time of data collection. The participants did not purposely seek for deepfakes; instead, they mostly encountered them passively through platforms making use of recommendation algorithms (see 4.1.2).

Based on the interviews, most young participants encountered deepfakes primarily on social media platforms. Typically, this content was experienced to pop up on the platform and become part of the feed. As C7 stated, “But if I’ve seen [a deepfake] like a recommended video, I’ve watched it” (C7). Participant B5 shed light on a particular social media platform where deepfakes are encountered: “[…] When you go through the Instagram feed, trying to find something entertaining to do, like with reviews or going through your Instagram feed, and find there’s bound to be at least some point where you find [deepfake-like] memes […]” (B5).

In line with the ways in which details are shared on a specific social media platform, D10 recalled: “[…] On TikTok, for your page, just scroll on them, you see one [deepfake]. [ …] You just scroll down, then you see, like, one of those [deepfakes]” (D10). G20, however, encountered deepfakes within a gaming context— “But I played games where you fake stuff into someone else’s voice or something”—which they considered a deepfake element.

Most participants mentioned YouTube and TikTok. Participants including A1, A2, C7, G19 and G21 claimed that they encountered deepfakes via YouTube, such as on “[…] the channels made [ …] for entertain[ment]” (G19), while participants, such as C7, D10 and E14, mentioned TikTok, indicating that a portion of deepfakes circulated as short videos. Participants also noted Twitter (e.g. A3), Facebook (e.g. B5) and Instagram (e.g. B5, C9) as platforms where they had encountered deepfake content.

Participants shared their experiences with various types of deepfakes as they perceived them, which predominantly encompassed celebrity, humorous, political and educational themes. Based on the interviews, most of the encountered deepfake content was in video format. However, one participant mentioned the presence of deepfake content in apps: “There were, like, a couple of apps that we could use [… those] into the celebrity [stuff … could] make [… a celebrity] say stuff […]” (G20). In addition, deepfake features in games—such as voice imitation—were also identified by one participant (G20) as a type of encountered content, as discussed in section 4.1.2.

Many participants, such as C7 and D10, noted that the most memorable deepfake content they found involved celebrities. This content usually consisted of a video showing “famous people” (A1) saying or doing something (C7, E15). E13 gave an example of an encountered deepfake:

Young people had also come across political content, which usually showed a political figure’s image, such as Vladimir Putin ^[7] (A2) or Donald Trump ^[8] (F17) giving a speech. The responses from A2 provide an example:

This participant identified a video of Putin as a deepfake, although the authenticity of the video remains uncertain. This raises interesting questions about how young people recognize and categorize deepfakes—particularly when real footage can be mistaken for manipulated content, or when disbelief may stem from a desire to reject uncomfortable information. Another participant, D10, offered further insight by referencing a similar example: a deepfake video of Putin found on a TikTok joke account, describing it as “some person who puts Putin’s face onto [someone else’s] and makes it look real” (D10). What’s more, participant F17 brought up encounters with another type of deepfake content: “Yeah, most of them are conspiracy theories” (F17).

Some of the young participants mentioned that they had encountered humorous deepfake content, mostly memes. Participants, such as D10 and G21, shared their experiences: “It was I think this year, maybe last year, it was a popular, like, meme […]” (D10), and “[…] there was like a bunch of memes” (G21). These memes usually featured well-known people, such as celebrities (D10) and political figures (G21), engaging in humorous activities such as singing funny songs or saying amusing things (D10).

Last, participant C9 mentioned encountering educational deepfake content: “[…] I mean, I’ve seen some YouTubers talk about it” (C9). In these videos, deepfakes were often used as examples, with people providing explanations or engaging in discussions about how these media work and their potential impacts.

This section presents findings related to RQ1, exploring young people’s reactions to encountering deepfakes by examining their casual interactions with (4.2.1), and responses to (4.2.2) the deepfake content they had come across. After encountering deepfake content, most of the young participants either watched it casually or paid little attention to it. However, specific types of deepfakes or deepfake-related content prompted some participants to take action, such as choosing to watch or deliberately avoid the videos.

The young participants generally adopted a casual approach towards deepfakes, noticing them but watching them without significant concern while acknowledging their potential negative impacts (see 4.3.2). Many participants expressed that they watched deepfakes when they come across them on social media platforms (such as B5, C7, D10, D11 and E13). For instance, when asked, “What do you do if you come across a deepfake?” D11 responded, “I just watch them” (D11). Likewise, interviewee B5 said, “Oh, I just look at it. I’m like, ha-ha, that’s kind of funny, and then just, like, leave it […]” (B5). D10’s words resonated with the aforementioned interviewees’ responses as well: “[…] If the thing [in the deepfake content] is funny … like, and then I’ll just, like, laugh about it. But then usually forget [about] it, like in the next 5 min” (D10).

Additionally, some participants expressed that they did not pay much attention to the encountered deepfake content, with B6 and D12 claiming, respectively, “[…] like, I didn’t probably pay much attention to them” (B6) and “Well, I don’t think about it that much” (D12). Similarly, E14 “[…] I just scroll[ed] the video” (E14) when coming across deepfake videos.

A group of participants described more complex situations in which they might respond to deepfakes encountered in everyday life. Some interviewees’ actions depended on the encountered content, with E15 stating, “Well … it depends on the content”, further explaining the criteria used to judge the encountered deepfake content:

Identifying the types of content also appeared to influence one participant’s decision as to how to respond to deepfakes. G19 apparently watched only funny deepfake videos. For certain types of content, such as those that are “too political” (G19), G19 actively avoided watching them, even attempting to modify the social media algorithms by, for example, clicking “not interested” to reduce the frequency of such content appearing.

Interestingly, F17 took different actions based on the timing of encountering deepfakes:

This participant emphasized being initially intrigued by deepfakes, even attempting to better understand them by learning more about the topic: “[…] If there’s an interesting video on deepfakes, for example, maybe a short documentary about them, I’ll watch it […]” (F17). However, F17 also expressed a lack of interest in watching “the deepfakes themselves” (F17), though they did not attempt to modify the algorithms of social media platforms.

Participants such as C8 and D12 shared their thoughts about the credibility of the deepfake content. For example, D12’s determinedly stated distrust in encountered deepfakes: “But you know, […] I don’t […] believe on those” (D12). However, D12 did emphasize an interest in the production of deepfakes, stating, “I just look at how they are done and like, […] I’m interested in […] how they make those, but I don’t believe them, of course” (D12). These responses indicate a sense of scepticism among some young participants regarding the presence of deepfakes in their lives.

This section shows the findings in response to RQ2, exploring young people’s understanding of “deepfake” through three key perspectives: content creation (4.3.1), impacts (4.3.2), and the role of AI (4.3.3) in the production of deepfakes. It should be noted that the participants’ interpretations reflect personal perceptions, rather than technically accurate definitions of how deepfakes are created or function. Participants may rely on assumptions, prior exposure to such media artefacts, or broader media narratives—areas that warrant deeper exploration in future research. These understandings may not fully align with the actual mechanisms behind AGM. Nevertheless, their perspectives offer valuable insight into how deepfakes are understood and interpreted in everyday life.

Despite varying levels of familiarity with the term, participants’ understanding of deepfakes centred on their “fakeness” and the manipulation of human faces. Before collecting the empirical data, we were concerned that the participants might not have heard of the term “deepfake”, as it was an emerging form of media that, at the time of data collection, was not widely discussed. However, the interviews showed that around three-fourths of the participants had at least heard the term. Participant C9, who was not familiar with the term, was very curious and asked the researchers to explain it. Participant F18, who had not previously encountered this type of media, attempted to guess what it might be. The remaining participants unfamiliar with the term realized they had actually encountered deepfakes in their daily lives, though they were previously unfamiliar with the term “deepfake”.

Most of those who had heard the term “deepfake” described it from the perspectives of the features and process of deepfake content creation. Regarding the features, a small number of participants talked about the evident feature of “fakeness”. For example, G19 and D12 shared their thoughts: “They are fake” (G19) and “[… it’s] like, [a] really fake thing. […] Extremely fake thing” (D12).

In terms of the process of creating deepfake content, many participants pointed to the process of human face creation in deepfakes. A few participants thought that this involved applying a filter or required specific software. As A2 described, “[…] you basically put a filter on them, and they look like other people” (A2). Pointing out the different tools of content creation, E15 explained the perception of how deepfakes work, stating that faces are processed using Photoshop-like software: “Well, basically you can kind of make anyone do anything, basically just by photoshopping their faces [ …] It’s just based on their faces when you are doing this” (E15).

Another group of participants focused on the combination of different elements within the deepfake content, such as the image, the sound and various components of the image. Some thought that the faces were taken from someone and a voice added. For instance, C7 explained that deepfakes “take someone’s face and make them speak” (C7). In line with C7’s view, F17 suggested that deepfakes allow someone to use “their faces to talk” (F17). Others, such as E13 and F16, suggested that deepfakes were made by “imposing somebody’s face on top of somebody else’s body and […] make it look like they’re them” (E13) or by putting a person’s face “into another image” (F16). Notably, one participant described such image creation in great detail: “[deepfakes take] multiple photos [ …] for facial features, like, different perspectives, help[ing] it, like, get the depth. Then they just mask it over a model or an actor” (D10). Furthermore, this respondent explained that the voice, such as a “celebrity voice” (D10), was synthesized for the image. Likewise, F17 mentioned voice creation for the faces: “[The creators] are using a voice changer to make their voice […]” (F17). Only one person considered that these “animated” faces were generated by a computer (D11).

These responses show participants’ common comprehension of the fake perspective of deepfakes and the manipulation of human faces. Although they may have lacked a full understanding of the complexity of these media artefacts, they perceived deepfakes as inauthentic media by noticing their visual and audio features.

The participants’ responses covered both the positive and negative impacts of deepfakes. Notably, the participants were more inclined to share their thoughts about negative consequences, expressing concerns over potential misuse and mis- and disinformation.

In terms of the positive impacts of deepfakes, the participants’ responses pointed primarily to entertainment. Participants such as C8 offered a general description of a scenario where deepfake is positively applied: “[…] If they’re used for […] good things, it’s like fun […]” (C8). In addition, students such as D10 and G19 pointed out the entertaining side of deepfakes. D10 took a meme of a celebrity as an example:

Likewise, G19 directly pointed out that deepfakes are “made […] for being entertain[ment]” (G19).

Concerning the negative impacts of deepfakes, the participants addressed their views on the issues related to general negative impacts, undistinguishable and misleading information, malicious use, and bias and stereotypes. Respondents such as B6 and F16 shared their general perspectives on the negative impacts of deepfakes. B6 noted that “So deepfakes, as any fake media, can cause a lot of attention [ …] it can just [cause] chaos in general” (B6), while F16 stated that, “depending on, like, what you use it for, it can be a bit, like, wrong, I guess” (F16). Among the participants who articulated their views in more detail, most expressed concerns about indistinguishable and misleading information in deepfake content, which they saw as potentially problematic. For example, E13 expressed views on deepfakes’ quality: “It’ll probably get even better quality and then it’s going to be way harder to distinguish what’s real and what isn’t” (E13). E15 conveyed similar ideas: “[… Deepfakes] can say something that can cause problems, and some people just don’t realize that it’s fake” (E15). An example of malicious use was conveyed by G21, who stated that the disinformation in deepfakes might be used to entrap someone: “So, you can, like, do a lot of, like, messed up stuff with it. Like framing someone for example” (G21). Participants such as B5, C8 and G20 suggested the possibility of malicious use of deepfakes, including “identity theft” (B5), “bullying” (C8), and spreading disinformation related to “celebrities” (G20). Last, two participants talked about bias and stereotypes brought about by deepfakes. While F18 pointed out that deepfakes are “stereotypical” (F17), D12 stressed that “somebody’s own personal life can, like … get biased by those videos” (D12).

These understandings indicate that although the participants may have found deepfakes entertaining, they stressed their detrimental potential, particularly in terms of disseminating mis- and disinformation or manipulating public perception.

Participants recognized AI as a key component in generating deepfakes, though their explanations of its technical role remained speculative and superficial. Except for F16 and F18, who had no knowledge about AI’s application in deepfakes, other participants shared their thoughts about its role. While all participants talked about how AI functions in deepfakes, none mentioned AI techniques, such as specific algorithms, in any detail. Notably, while touching on the topic of AI, some participants were prone to express their views speculatively or with uncertainty, using phrasing such as “AI is probably […]” (A2), “Well, I’m not sure […]” (C7), “I mean, it probably […]” (C9) and “I’m, like, maybe” (C8).

Most participants noted that AI was possibly employed to process the images of deepfakes, pointing primarily to two perspectives, video creation or image analysis, with video creation being the most mentioned. Some claimed that AI might be used to generate and change people’s faces (such as A3 and B6). One participant pointed out the word “filter” while articulating their views: “[AI technologies] create or make the filter possible to be in the person” (E14). Some explained the process of face creation in a more detailed way. For example, G21 described AI’s function as image scanning and stitching: “If the moving part [of someone’s photo] was the head, it would scan the head. Then you would also scan the video where the actual moving is happening. [AI technologies] make [them fit] together” (G21). Participants also mentioned AI playing a role in editing (such as C8), with one participant explaining that AI is used in the “whole editing of [the video]” (E15).

The second perspective of image processing indicated by participants was image analysis. While talking about the details of the videos, B5 pointed out that “[… AI] can analyse things” (B5). Other participants mostly mentioned that AI conducts image analysis of facial elements. For example, C7 indicated that it was possible for AI to learn “facial movements” to make the deepfakes “more realistic” (C7), while B5 stated that “[…] analysing facial structure is one of the things that AI can do best” (B5).

AI was also described as a tool to process sound, especially the human voice. During enquiries into the role played by AI, participants such as A3 unequivocally pointed to “the voice” (A3) as their direct response. Some participants expressed that AI could be utilized to generate or change characters’ voices in deepfakes (such as A1 and D12). For example, according to A1’s description, AI can “change the voice” (A1), while D12 said that AI is the “main thing” that supports “voice generators” (D12).

What is more, some participants pointed to the general role of AI beyond the sphere of deepfake. Apart from deepfakes, B6 thought that “AI can play anything, even if it has a big role in, like, videos”. In addition, participants such as G19 shared about future applications of AI, such as creating “fake faces” for “inauthentic humans” as spies put into “a different territory” (G19).

The speculative tone in these excerpts indicates that while the young participants grasped the involvement of AI technology in deepfake creation, they were less sure about the specifics of how it functions. This suggests that participants lacked detailed knowledge about its algorithmic processes. Some participants, such as G21, showed a fundamental understanding of AI as a tool for combining visual elements, though without technical depth. This reflects a broader trend in which the public has a vague notion of AI’s functionality in media production but lacks the expertise to fully articulate its mechanisms (Westerlund, 2019).

The results of this research show that the young people who participated in this study had encountered deepfakes in their daily digital experiences, suggesting that generative AI had already shaped their everyday lives at the beginning of the decade, in 2021. Some of the deepfake content encountered by the participants appears particularly eye-catching, seizing youths’ attention.

In relation to RQ1, which explores young people’s practices while engaging with the encountered deepfake content, this study first identified situations in which these practices occurred, describing how deepfake content was often encountered incidentally on social media platforms; in these situations, the young participants reported infrequent encounters with deepfakes in their everyday life and mostly engaged with such content in a casual manner; a group of participants took action in response to certain deepfakes, depending on their content—for example, actively watching them if the content was humorous. These notions on the young participants’ encounters with deepfakes align with Hassoun et al.’s (2023) findings on the exposure to online information, indicating broader trends in young people primarily encountering online content incidentally rather than through intentional searching. Furthermore, our findings highlight the role of information systems in shaping how people encounter deepfakes. For instance, the recommendation algorithms used by social media platforms can actively push deepfake content to young users.

Second, this empirical study noted that young people’s engagement with encountered deepfakes shows nuances that differ from the models of information encountering proposed by Erdelez (2005) and Awan et al. (2019). For example, Erdelez’s model (2005) describes the process of information encountering as beginning with noticing, where the information user identifies content that appears relevant to their problem or need. In this study, deepfakes may not have appeared to meet most of the young participants’ information needs or address specific concerns. Nevertheless, such encounters can still hold meaning or exert influence. For instance, participant G19 had intentionally searched for deepfakes and even shared the content with a friend, suggesting that deepfake content may hold informational or social value for some individuals; additionally, Awan et al. (2019) proposed that the stages of “keeping” and “sharing” should be inserted between “capturing” and “returning” in the information encountering process. However, in our findings, most young participants neither kept nor shared the deepfake content they encountered—instead, they simply scrolled past it.

Overall, the results regarding young people’s MIL practices resonate with Makri and Buckley’s (2020) discourse, which posits that the situations of information encountering within concrete scenarios may vary from Erdelez's (2005) model. The idealized model can be shaped by different factors, such as the media content of deepfakes, the timing of their encounters, and the participants’ curiosity towards them. Taking this into account, it is essential to incorporate the aspect of information encountering into our understanding of the current digital online environment, which is increasingly shaped by generative AI.

In relation to RQ2, which focuses on young people’s understandings of “deepfake”, the participants demonstrated a certain level of awareness of the foundational knowledge of generative AI technology that is used to create such content. Contextualized in the current digital environment, young people show that their understanding covered not only the information conveyed by deepfakes but also the technology by which deepfakes are powered. Unlike the understanding gained through content creation practices using generative AI, as shown in Ali et al.’s research (2024), the young participants in this study could reflect on the media content they had received, developing an understanding both the emerging forms of media and the technology behind them.

In addition, this study highlights the need to rethink the concept of MIL. While UNESCO’s definition of MIL emphasizes citizens’ competencies to “access, retrieve, evaluate and use, create as well as share information and media content in all formats” (2012, cited in Frau-Meigs, 2023, p. 4), it does not fully address the notion of information encountering. Given the phenomenon in which such encounters occur —particularly within algorithmically driven, recommendation-based digital environments—the concept of “information encountering” should be more explicitly integrated into the MIL framework, aligning with the argument put forth by Erdelez et al. (2011). Furthermore, approached with a practice lens, in today’s digital context, technological elements are deeply intertwined with digital media and, consequently, with young people’s practices of engaging with these media artefacts. In the case of deepfakes, it becomes difficult to separate the act of engagement from the underlying technologies that shape these interactions. For instance, in this study, some of the young participants expressed distrust towards deepfakes based on their understanding of how such content is created. This suggests that their practices are influenced by their technological awareness. Therefore, we argue that individuals’ understanding of technology should be integrated as a critical sub-element within the MIL framework.

While discussing their understandings of deepfakes, the young participants in this study primarily focused on the negative impacts brought by deepfakes—such as mis- and disinformation issues, malicious use and bias and stereotypes—rather than the potential opportunities offered by this AGM. This perspective echoes a body of previous research showing that young people primarily discuss the negative sides of technology when sharing their perceptions of generative AI (e.g. Higgs and Stornaiuolo, 2024; Vaccari and Chadwick, 2020). However, as generative AI offers numerous benefits, particularly in the realm of content creation, it is important that young people are encouraged to develop an awareness of these technological strengths and to develop their AI literacy accordingly. As MIL inspires citizens to engage critically with media content (UNESCO, 2012; cited in Frau-Meigs, 2023, p. 4), we urge educators to strengthen generative AI education, guiding young people to think about this novel technology and its artefacts in a more balanced manner.

By combining the perspectives of both RQ1 and RQ2, we observe a paradox in the context of encountering deepfakes: while many of the young participants acknowledged the potential harms associated with deepfakes, they often adopted a casual stance—neither fully rejecting nor endorsing the content. Even when some recognized deepfakes as containing misinformation or disinformation, they still tended to treat them with indifference. These findings are not in line with Malki et al.’s research (2023), which found that people tend to exhibit negative emotional responses when encountering misinformation. This nuanced engagement shows the complexity of youths’ information practices with the information generated by emerging technologies in their everyday lives and reflects a broader trend in how young people respond to new technological phenomena—with ambivalence, curiosity and, sometimes, indifference. It also presents that young people’s practices are multifaceted, moving beyond the simplistic binary of either resisting or accepting these new media. To this extent, it is crucial for researchers and educators to understand these varied reactions, which will help in developing a more effective MIL pedagogy that addresses this complexity.

This study contributes to the body of knowledge on young people’s MIL practices concerning AGM in their everyday lives. By employing a qualitative method, it examines young people’s encounters with deepfakes and presents findings on MIL practices from two perspectives: young people’s engagement with deepfake content in their everyday lives and their understanding of “deepfake”. Regarding their engagement, this research first characterizes the context in which young participants are exposed to deepfake content: in their daily setting, participants primarily encounter deepfake content—often featuring celebrities, humour, political themes or educational material—on social media platforms, instead of intentionally searching for it. Second, the results show how young participants respond to these deepfakes. While most participants approached deepfake content casually or not paid much attention to it, certain types of deepfakes or related content prompted some to take action—such as deliberately choosing to watch or avoid the videos. Some participants showed their distrustful attitude towards deepfakes in their daily lives. In terms of young people’s understanding of deepfakes, the results indicate that most participants had a basic understanding of deepfakes, pointing to ideas about content creation, impact and the AI techniques involved.

Theoretically, the findings of this research enhance our understanding of MIL practices in the context of young people’s everyday encounters with AGM. We argue that young people’s understanding of technology should be integrated into the MIL framework. Additionally, our findings point to the potential value of revisiting the concept of information encountering in light of the current online environment. From a practical perspective, this research provides valuable insights into how young people engage with and evaluate AGM, which can inform the design of targeted interventions by policymakers, educators and organizations to improve young people’s MIL, enabling them to more effectively navigate the evolving landscape of AGM.

As for the transferability of this qualitative study, it is important to note that the empirical data were collected in 2021 and with a specific group of young people who took part in an AI-focused work practice programme. As a result, the findings may not accurately reflect young people’s current experiences with AGM, particularly given the significant and ongoing changes in the digital landscape driven by generative AI technologies. Regarding the credibility of this study, some responses collected during the interviews remained somewhat superficial, as not all participants had a strong grasp of the interviewing process. Since deepfake content was not a primary component of young people’s everyday media consumption—at least at the time of data collection—it proved challenging for them to recall such encounters in detail.

Based on these reflections, future research ought to investigate young people’s current everyday experiences and practices with deepfakes and reflect them to the findings of this study. This approach would both highlight the significant evolution of generative AI-driven media and examine shifts in young people’s MIL practices.