Does generative AI mean the end for online ethnography?

Abstract

The social composition of online spaces such as social media platforms has undergone a recent dramatic transformation with the advent of generative AI. Online material generated by AI includes human-like chatbot interactions and manufactured text and images purporting to represent real-life people, places and events. Efforts to estimate the extent of such material are hampered by the considerable difficulty of reliably detecting machine-generated material. For some commentators, these developments threaten the ability of online spaces to offer meaningful social engagement. This paper explores whether, in this context, it is still possible to conduct online ethnography aimed at understanding culturally-embedded meaning-making. The paper argues against a generalised methodological exceptionalism for generative AI. Instead, some promising strategies are found in existing methodological approaches that treat authenticity as a problem experienced by ethnographers and participants. A reflexive approach to ethnographic treatment of authenticity remains a valuable stance in situations where suspicions about the presence of generative AI are raised. In particular, multi-sited approaches allow experience of varying and cross-contextual understandings of authenticity and autoethnography focuses attention on how we navigate the lived experience of uncertainty about the nature of online content. Second, the paper turns to more-than-human ethnographic approaches and finds ethnography positioned as an immersive means to embrace non-human actors, including AI-generated features, as an intrinsic part of online experience. Such approaches ask for reflexivity around what is at stake in making judgements about the ontological state of materials encountered online. The methodological strategies reviewed here suggest that there is a future for online ethnography in the face of generative AI involving ongoing methodological innovation without wholesale methodological exceptionalism, but that this requires both a multi-faceted reflexivity and caution in adopting human-centric approaches founded on principled separability of human and machine.

Introduction

Online ethnography emerged in the 1990s as a means to interrogate the cultural space that was then emerging on the Internet. While the initial focus was on the description of orderly social formations that might be termed online communities (Baym, 1995), subsequent developments challenged the dominance of the online communities metaphor (Postill, 2008), implemented multi-sited and cross-platform designs (Postill and Pink, 2012; Robinson and Schulz, 2009) and responded to the emergence of social media (Caliandro, 2018). These varied forms of online ethnography were able to explore an array of topics of social and cultural significance such as fandoms, peer-to-peer health and wellbeing support and the emergence of new social movements. Across these varied instantiations of online ethnography, it might seem quite obvious that ethnographers are interested learning about people by interrogating how they behave and interact with one another within and across different forms of mediated and face-to-face communication. A variety of interconnected issues complicate this simplistic assumption as applied to digital spaces, including: the challenge of identifying connections between social spaces online and offline; the role of technological platforms in constituting the conditions under which presence and interaction become possible; the mediation of the visibility of online interactions through a variety of somewhat opaque algorithmic feeds; and, the key topic of this paper, the proliferation of content generated by AI in the form of conversational chatbots, automated contributions to online forums and computer-generated material including images, writing and video. Now, more than ever, an ethnographer is confronted with online material that has an indeterminate or ambiguous relationship with people and places that it seems to represent. This paper offers a methodological response to the apparent challenge that this proliferation of computer-generated materials poses for the ethnographic endeavour.

Fundamental to classic versions of ethnography is the ethnographer’s effort to arrive at an understanding of observable features of the field setting according to their culturally embedded meaning. Geertz (1973) explains such efforts by referring to Ryle’s exploration of the potential array of meanings attributable to the movement of an eyelid that we know as a wink. When such aspirations to understand culturally embedded meaning are transferred to field sites that are partially or wholly online, the question arises whether meeting this aspiration involves needing to know whether we are in fact observing people. This point has often been taken for granted. As Rabe (2023) suggests, it remains a common shorthand to state that when we look online we are observing human behaviour. Caliandro (2018), for example, in exploring some useful metaphors for organising an ethnographic interest in social media, identifies community, crowd, public, self-presentation as tool and user as device. All of these metaphors, while they may be developed in ways that are nuanced through understanding of their co-constitution with the technological platforms that enable them, are recognisably founded in concepts of humans and human behaviours. In the contemporary Internet, however, AI-generated content ever more closely mimics human behaviours and becomes increasingly hard to detect. As the amount of such content, at least in some online settings, threatens to drown out the content generated without the use of AI, it becomes difficult to sustain a claim that what we see online is to be interpreted directly as representative of human behaviour. If we understand the point of online ethnography as being to study how people are behaving online then such a realist, human-centric approach to online ethnography is increasingly implausible. Fortunately, there are alternative conceptions already present within the methodological traditions of online ethnography that offer productive strategies for responding to, and even embracing, generative AI as a meaningful part of the online field site. Reflexive, autoethnographic and more-than-human approaches have considerable potential as means to explore the contemporary online space on its own terms and to investigate culturally embedded meaning-making even in the face of generative AI. A complete methodological exceptionalism for generative AI is therefore not necessary.

The first section of the paper explores the nature and extent of online machine-generated content, arriving at the conclusion that such material represents a substantial and increasing proportion of the interactions visible to an online ethnographer and that it is increasingly difficult to determine the human or machine origins of any given piece of content. The paper then outlines arguments rooted in the methodological literature on online ethnography, proposing that an immersive ethnographic understanding of what goes on in online spaces and how it comes to matter can still be as a plausible strategy for the age of generative AI. This methodological direction is a matter of choice, as Marres (2017) suggests: a decision about our methodological stance is not one that we can or should treat as a necessity determined by the technology but instead requires that we articulate what it is that we want from our methodologies under the circumstances we find ourselves operating within. This paper argues that we should not abandon online ethnography in the face of generative AI, but that it is nonetheless important to be reflexive about our methodological choices and clear what is at stake when we interrogate online spaces suffused with generative AI. The argument is presented through two intersecting themes in the methodological literature on online ethnography: first, an exploration of the ways in which ethnographers have addressed concerns about whether their online observations are, in a variety of ways, authentic; and second, the possibility of a “more-than-human” approach to ethnography that does not attempt to separate human from machine-generated content. Through these two themes, a set of strategies are advanced for ethnographic exploration that embraces generative AI rather than treating it as a problem to be managed away.

The focus within this paper is on versions of ethnography that are conducted through participant observation by an ethnographer and not employing generative AI as ethnographic tool or adjunct, although this is certainly a topic of interest and emerging field (Li and Abramson, 2025). The focus is also not primarily on the “synthetic ethnography” constituted by ethnographic study of generative AI through conversational engagement with it as outlined by De Seta et al. (2024). The approaches described by De Seta et al. (2024) focus on interrogating the workings of generative AI that we already know to be such, and are founded in ethnographic work aimed at opening the black box of algorithmic technologies. This approach may indeed be one component of an ethnographic embrace of generative AI in online spaces. However, the current article is targeted more broadly on challenges that might be encountered by an online ethnographer who is pursuing a sociologically-framed research question in the tradition of online ethnography and encounters material that may, or may not, be generated by AI. As the next section outlines, this is an increasingly common experience.

The nature and extent of computer-generated content in online spaces

AI-mediated communication can take a variety of forms from suggested wordings and addition of emojis through to wholesale generation of content (Hancock et al., 2020). While artificial intelligence has been a focus of computational research for decades, and rule-based conversational chatbots and machine-generated non-player characters in digital games have been present on the Internet for many years, the amounts and kinds of machine-generated content encountered in online spaces have changed dramatically in recent years. Before the advent of the recent wave of Large Language Models led by ChatGPT’s release to the public in 2022, chatbots that simulated human-generated content were already widespread: using a machine learning approach applied to archived data from Twitter in 2015, Varol et al. (2017) suggested that between 9% and 15% of active Twitter accounts at that point were bots. These bots included spammers and self-promoters in addition to those implicated in large-scale attempts to influence public debate. As Leslie and Meng (2024: 2) outline, however, since the launch of ChatGPT the rise of generative AI has been “explosive,” occasioning the “flooding of socio-digital space with empathy-simulating chatbots” that have the ability to fool people into thinking that they are dealing with another sentient human. According to Imperva (2025) automated “bot” traffic on the Internet now exceeds human activity at 51% and appears to have been boosted by the increasing capacities of generative AI making bot development more accessible to a wider array of users.

Estimates of how much content on social media is generated by AI vary widely and headline figures that circulate are often very hard to verify: a widely circulated statistic that 90% of online content may be synthetically generated by 2026 is often attributed to a Europol (2022) report and yet the statistic does not appear in this report. Other estimates appear in blogs and pre-prints yet to undergo peer review. In a pre-peer-review study, Sun et al. (2024) train a machine learning algorithm to detect AI-generated content across three online platforms and find that the proportion in each case is rising but varies widely across platforms, from a high of 39% in Quora to under 2.5% in Reddit. A pre-print from DiResta and Goldstein (2024) found AI-generated images generating high levels of engagement and circulation on Facebook, often seemingly without being recognised as AI-generated by those commenting on them. Another pre-print, by Thompson et al. (2024), highlights the extent of machine translation being used to generate multiple copies of material online in different languages. Using evidence derived from the existence of multi-way parallel translations of the same material on websites into different languages, they identify 57% of their corpus as arising in this way, framed as a “shocking” amount.

The absolute amounts of machine-generated material in any online space are thus hard to specify, but without doubt there has been rapid and significant growth in the extent of machine-generated content online. This increase in quantity, coupled with the increasing difficulty of reliable detection has occasioned some considerable disquiet regarding consequences for the quality of information online and the nature of interactions there. For some this threatens the usability of social media platforms as meaningful social spaces:

Once heralded as bastions of human interaction and connectivity, these platforms are increasingly becoming conduits for AI-driven content, prioritizing consumption over authentic social engagement. (Walter, 2025: 239)

In similar vein Tiffany (2021) describes “Dead Internet Theory”: a conspiracy theory about the overwhelming extent of fake content online that they argue has a certain feeling of plausibility in the sense that the online space has moved away from human-human interaction into increasingly consumer-oriented content that is automatically generated by chatbots. Clearly any researcher attempting to treat such a platform as a space of cultural meaning-making would want at least to be aware of the presence of such non-human actors and might well be daunted by the apparent loss of social authenticity. It is not so obvious how this awareness translates into methodological strategies for the ethnographer. One key question is whether in order to do research in such online spaces an ethnographer themselves needs to develop the skills of the machine learning model derived by Varol et al. (2017) and learn to sort the human from the bot. Accounts suggest that such aspirations may be doomed to failure: Meng et al. (2025) found that humans in their experiments were unable reliably to distinguish between human-generated and AI-generated online reviews. Jacobsen (2025) outlines the emergence of a nexus of research activity that is seeking means to identify deepfake materials algorithmically. They suggest, however, that such efforts are both doomed to failure in technical terms and also somewhat unhelpful in their reduction of a plethora of different forms of representation to a binary of fake or not.

If it is not possible reliably to tell whether any individual piece of online content or moment of interaction is machine-generated, how does this affect the ethnographic endeavour? How is the ethnographer to navigate a space where the interactions that they observe, and even those that arise within what appears to be a one-to-one interview might be created by a bot or constructed by generative AI in response to an unknown prompt? Ethnographers are attuned to the idea that people portray themselves according to their perception of the settings that they find themselves in, but are we attuned to the degree to which these portrayals might be AI-assisted? An awareness of the potential for generative AI to be involved shifts our framing of agency and intention in the construction of messages. This, in turn, has implications for our conceptualisation of identity performance and impression management. Whilst we might hitherto have been inspired by Goffman (1959) to understand an interactional turn within an interview as part of a performative social interaction between two people, the potential for AI to become a part of the interaction suggests a need for attention to additional forms of agency at play within the construction of the performance (Klowait and Erofeeva, 2025).

Whilst taking seriously the extent of machine-generated material online, it is important to acknowledge that AI-generated images, videos and interactional turns are produced by someone, somewhere feeding a prompt to AI or making a choice to employ these technologies to pursue their goals. The chatbots that generate online content are commissioned and paid for by someone who may have a wish for self-promotion or a desire for political disruption. Bots could thus be seen as machine-generated proxies of human actors, since the bots may be producing their own content but at the instigation of humans (Santini et al., 2020). Content, even AI-generated content, is derived from and made meaningful by humans embedded in their social and cultural contexts. Generative AI produces content by drawing on learning from an array of material, and is recycling and recombining existing ideas. Taking these factors into account, it is important not to develop an ethnographic strategy that enacts a principled boundary between the machine-generated on one hand and the truly human on the other. The two are inextricably entwined, humans are part-machine and machine-generated material is human too. For the ethnographer, the methodological problem arises more in the sense that we may not be able to tell how this entwining takes place nor have access to those social and cultural contexts of meaning-making in which the machine-generated material arose. This indeterminacy poses analytic challenges for an ethnographer striving to understand processes of meaning-making. In the next section two intersecting bodies of methodological literature relating to online ethnography are examined for their capacity to guide strategies to guide the online ethnographer’s approach to generative AI: the first relates to the treatment of concerns regarding authenticity in the practices of online ethnography; and the second to more-than-human approaches to ethnography.

Authenticity in online ethnography

As Beaulieu (2004) points out, various ways of specifying the nature of online ethnography are apparent in early work on the topic, spanning both those that treat online space as a site for observation or capture of cultural traces and those who set out to achieve intersubjectivity with those present in the online space. In the 1990s when the first online ethnographies were being conducted, it was relatively straightforward to assume that the content being observed and created in response to the ethnographer’s engagement was produced by human beings, sitting at keyboards typing out messages and reading one another’s words. At this point, the concerns of ethnographers working in pursuit either of capturing an online culture or of achieving intersubjective connection were often focused around the authenticity of this content, asking whether one might take at face value what people said online to one another and to the ethnographer (Paccagnella, 1997). To the extent that content was interpretable as not a transparent representation of an individual’s actual situation, this could be understood through framings such as fakes, lies, deceit and trolling, focusing on an agentic individual who set out to deviate from truth in an environment that was otherwise positioned as truthful. Alternatively, the ethnographer might consider all online content to be effectively a performance in context, and desist from making their own distinctions between truth and lies, leaving that to the participants in the setting. In some spaces, even before the current upsurge in content generated by AI, machine-generated content might go unnoticed by the unwary. The treatment of this array of issues, gathered here under the over-arching concept of “authenticity” have been a significant focus of methodological discussions regarding online ethnography. Cautioned by Beaulieu (2004), we should not expect to converge on a singular solution since objectives vary, but we may find within this literature strategies to inspire ethnographers within the contemporary online field and provide guidance for their treatment of generative AI. The remainder of this section focuses initially on approaches that conceptualise identification of machine-generated material as a skilled practice for ethnographers and participants, and then moves to explore ethnographic treatments that treat authenticity variously as emic and etic concept. Finally, the purchase offered by multi-sited and autoethnographic research designs is explored. Across this array of strategies, reflexivity is key to promote conscious attention to whose conceptualisation of authenticity is the focus and what is at stake in judgements of authenticity.

A classical conceptualisation of ethnography founded on immersive participant observation can be aimed at coming to understand what Boellstorff (2008) describes as the “cultural logics” of virtual worlds on their own terms. Boellstorff et al. (2012) adopt this form of classical ethnographic approach in their advice for ethnographers embarking on study of virtual worlds, premised on the notion that ethnographers are there to study cultures that are comprised of people engaged within technologically-mediated spaces. In this framing, the issue of working out what aspects of the observed phenomena are directly the product of human behaviours and what aspects might be technological artefacts becomes a component within the ethnographer’s efforts to work out what is going on. For example, when discussing the possibility that an avatar encountered in a virtual world might be exhibiting pre-programmed behaviours, Boellstorff et al. (2012: 103) suggest that “in many virtual worlds gestures or animations that are user controlled look different from those that are automated, and we should learn to recognize the difference.” In as far as the ethnographer is aiming to develop cultural competence within the setting, knowing what is user-controlled and what is automated is here positioned as a part of that cultural competence. Similarly, Nardi (2010) describes the ability of game players to spot the computer-controlled NPCs (non-player characters) as part of the cultural competence that is acquired through familiarity with the game world.

Echoes of this approach are seen in some more recent suggestions of ways to tackle the establishment of authenticity in social media profiles, for example in the efforts of Rabe (2023) to outline an ethnographic approach to cloaked profiles. The term “cloaked” alludes to profiles that seek to conceal their authorship or disguise their agenda and such profiles may be “more or less human-made” (Rabe, 2023: 540) including both the work of bots and human actors with various motivations including preservation of their own safety. As Rabe (2023) outlines, even in a face-to-face setting we may have challenges to assess the authenticity of what we are told, but online it can become much harder to assess status. Even so, some strategies for uncloaking profiles are outlined, involving close attention to presentational issues and detective work across platforms. Similarly, Farkas et al. (2018) identify strategies for determining that a Facebook page is cloaked, but then suggest that the researcher should primarily focus on the identity created on that page (“the design of the cloak”) without being able to access the account’s originators or explore more about the circumstances that occasioned the page’s existence. While Boellstorff (2008), Boellstorff et al. (2012) and Nardi (2010) largely focus on identifying machine-generated content as a participant competence that the ethnographer seeks to develop, Rabe (2023) highlights an area of skilled effort from the ethnographer that does not necessarily mirror participant competences. Whether or not this skilled effort is deemed appropriate depends on the objective of the ethnography.

An ethnographic strategy built on acquiring competency in identifying machine-generated material depends on there being an observable difference that ethnographer and/or participants within the setting are confident that they can become competent at discerning. Some nuance is required here as it is not necessarily deemed appropriate for the ethnographer simply to align their own standards for what counts as reliable data with participants’ approaches to understanding such issues. Drawing on classical ethnographic terminology as Boellstorff et al. (2012) advise, we might say that the ethnographer needs to be reflexive regarding whether to treat “user-controlled” and “automated” as emic categories that participants use themselves or as etic categories for building comparisons across settings. Observing differences in the ways that participants relate to various aspects of their technological environment may also be important in understanding cultural logics. For example, Kendall (2002) notes that the ability to make artful use of programming objects to animate the space of an online forum gives status within the group and also notes that knowing what is going on when such programming objects are deployed is the mark of an insider. An awareness of status differentials associated with the production or detection of computer-generated materials may therefore be significant, to the extent that the ethnographer is striving to develop what might count as insider knowledge in terms both of skills and the role that they play in social hierarchies. Again, reflexivity regarding the fit between concepts of authenticity and the ultimate goals of the ethnography is key.

As discussed earlier, contemporary discussions around AI-generated content suggest that the cultural competence to distinguish user-controlled material from that produced by automated means can no longer be taken for granted. The question then arises as to what the appropriate ethnographic strategies are if both the ethnographer and participants are unable consistently to identify AI-generated material and distinguish it from other forms of user behaviour. Here it is useful to turn to versions of online ethnography that do not insist on the ethnographer learning to make their own distinctions between human and machine-generated content and, as Farkas et al. (2018) outline, focus instead on the identity created. As outlined by Hine (2000), in the early days of online ethnography the issue of authenticity often arose in connection with concerns about whether ethnographers should trust what people say online about their offline lives. At the time (pre social media), the Internet was strongly associated with forms of identity play that involved adoption of online personae at odds with offline presentations of self. Hine (2000) argues that instead of the ethnographer attempting to verify mediated interactions, they can instead take these interactions as authentic on their own terms, in effect embracing whatever emic categories to underpin knowledge and trust are in operation within the field setting. The ethnographer can take an experiential approach that aims to draw on the same resources and embrace the same uncertainties as other users present in the setting. This strategy aims to avoid treating authenticity as a problem that needs to be solved through skilled detective work and verification through triangulation with other communication media and instead suggests that we think of authenticity as a culturally embedded concept that both participants and ethnographer may be grappling with.

When treating the judgement of authenticity as itself a culturally-embedded aspect of meaning-making, it becomes significant to note that the treatment of authenticity as an issue varies according to local conventions in different online spaces. For example, Baym (2000) notes a normative preference for use of real names and an aversion to anonymity as a feature of the soap opera online discussion forum that she studies, connected with an atmosphere of trust and personal disclosure. Kendall (2002) describes a trust that emerges in relation to an online forum both in virtue of long-term participation and an interweaving of online and offline interactions. Through their deployment of different modes of interaction and conventions of identification the participants in these forums are able to build their own locally specific form of trust in one another’s authenticity. The ethnographer’s role is thus not to determine whether people are right or wrong to place such trust in one another, but rather to learn how trust emerges as a context-specific practice. As Hine (2015: 116) suggests:

The ethnographer in such settings can follow a circulation of representations, looking at the divergent renditions of what the object means in different settings without making judgments about the accuracy or authenticity of any individual representation as a true portrayal of what the object actually is.

More recently, Polleri (2022) and Holmes (2025) suggest a similarly agnostic approach to ethnographic study of online misinformation. Applying this kind of contextually sensitive user perspective on the issues of trust and authenticity to the emergence of generative AI suggests that the conditions of trust become in themselves all the more an important topic for ethnographic enquiry. We might anticipate that an influx of AI-generated material into an established setting may shift the conditions of trust within such a context and occasion a climate of suspicion. Alternatively, generative AI might achieve a convincing persona and any human participants in an online space might either not know, or not care, that this is the case. Anecdotal observation of online forums such as reddit and mumsnet suggests that participants have indeed become attuned to the possibility of AI-generated content and that there is an emerging cultural practice of highlighting suspect material and labelling it as culturally inappropriate. For such participants, a preservation of these forums as domains of “authentic” human interaction is signalled as a cultural priority. Hierarchies of participants may emerge around the ability to spot AI-generated content, and some participants may accuse others of being either overly trusting or too suspicious. Social media influencers may develop new sets of practices aimed at convincing others that their contributions are to be taken as authentic (Taylor, 2022). Emergent cultures of trust and suspicion become the focus of the ethnographer’s interest, rather than judging who might be “right” or “wrong” in their judgements and indeed whether those making the judgements are even themselves human. Returning to a classical ethnographic approach, but this time one that starts in offline cultural spaces, Miller and Slater (2020) propose that we enter online spaces from the perspective that our interlocutors enter it and strive to see it through their eyes. Transferred to the issue of generative AI, such an approach suggests that we seek to understand it as a concern through the perspective that our interlocutors do and suspend any attempt to work out whether any given material is authentically human. As outlined by Heřmanová et al. (2022) and Taylor (2022), this approach treats authenticity as performative and processual rather than absolute.

This ethnographic strategy might inform an approach to content shared on social media. Aware of the possibility of the generation and manipulation of digital images, an ethnographer might set out to explore the visual culture surrounding a topic such as the online depictions of holidays. They might do this, not with the ambition to find out what holidays were actually like, but instead to understand the culture that prevails around holidays in online space, considering what aspects are shared and by whom and what responses they gain. The ethnographer can explore what counts as a good way to portray a holiday in this context, and take an immersive and experiential approach to exploring how such material travels and how it is marked as valuable or not through comments, shares and likes, understanding that the origins of material are opaque and that visibility is mediated through algorithmic control of visibility that no users can be fully cognisant of. Again, rather than seeking to identify and expose AI-generated material, the ethnographer is aiming at developing an immersive understanding of the emerging culture of this space. It is important, however, to be careful in positioning the significance of such an ethnographic study and not to claim a generalisability beyond the scope of the study as having captured something transcendent about the nature of holidays.

A further challenge for an observational stance on culturally embedded notions of authenticity potentially arises, to the extent that the AI-generated nature of materials may simply not be a topic for comment. Either lack of awareness, or nonchalance about such matters, or a process of mundanisation (Willim, 2024) that simply renders this material too ordinary to comment on may occur. On mundanisation, Willim (2024: 8) suggests that “a drift between attention and disregard, between embodied skills and ignorance” may occur such that a technological feature that seems at one time to require a skilled and discerning attention may come at another time to seem simply part of the way things are. Silences may be amenable to an autoethnographic interrogation of the practical and affective consequences of what it is that is being left unsaid (Hine, 2020). Alternatively, creative forms of probing can extend the classical ethnographic approach to acquiring cultural competence through participant observation into more active ways of surfacing what might be implicit in a situation. Probing, for Willim (2024) might involve performance and art works aimed at provoking questions and exploring the boundaries of what is possible. Such an approach to the issue of generative AI in online spaces might involve interventions aimed at exploring the consequences of ambiguity with interlocutors and provoking reflections on what might be at stake as boundaries between human and AI become less apparent. Again, it is a methodological choice whether to sit with the silence in the setting or to intervene.

When generative AI is embraced as a feature of the setting, ethnographic focus may then be drawn to it only in as much as it is rendered notable or problematic within the setting itself or in so far as the ethnographer is able to provoke reflections on its normalisation. The ethnographer is able to embrace the flow of material that the combination of user agency, automated chatbot and algorithmic feed brings to their attention, and explore how and where issues of authenticity arise as a feature of the setting. This involves attending to comments and reposts, noting where suspicions are raised and by whom and what responses they get. Is there an emergent discourse of knowingness around the presence of AI-generated material? How is authority around such matters manifested? Ethnographic understandings emerge through observations, interviews and immersion in the processes of navigating material through spaces and over time. A multi-sited and multi-temporal approach in the spirit of Marcus (1995) may feel appropriate in order to explore the varied spatial and temporal dimensions of trust in material that may appear, from observations within a single online space, to be trusted without question. It is important to bear in mind that the authenticity of digital materials may be topicalised in other spaces to those that the material itself is found in. Whilst in some spaces the ethnographer might find no attention given to the potential AI-generated status of materials, in other spaces serious concerns might be raised about what is happening in those spaces where AI-generated material is being mistaken for human-generated material, and vice versa. It is possible to treat these different spaces as “mutually contextualizing” (Orgad, 2009: 48) rather than treating one account as more truthful than or accounting for the other.

While a multi-sited approach that explores issues of authenticity across spaces is one response to the challenge of what to trust, another direction of methodological reasoning suggests an autoethnographic approach. Markham (1998: 17) takes an autoethnographic perspective on her exploration of early online spaces, describing her book as an attempt to “tell the story of what it took for me to get connected and of what and whom I encountered once I was there.” The “what and whom” here is significant in distinguishing the approach, since a key upshot of the exploration is a discussion of the multiplicity of online experience. Rather than aiming to close down on one singular interpretation of what online space is, Markham (1998) suggests that we recognise an ontological ambiguity such that online technologies can be seen variously as offering a tool or space or way of being. Markham (1998) draws autoethnographically on her own experience and her conversations in online space to reflect on the multiple and fluid nature of lived experience of online technologies as cultural form. This stance offers some potential inspiration for an autoethnographic approach to generative AI. An autoethnographic perspective on generative AI inspired by Markham (1998) might entail embracing an enduring uncertainty about the status of the materials we encounter online, or suggest that alternating between stances on their ontological status may be inherent in the experience. Our experience of encountering spaces suffused by generative AI might entail flipping between attempts to distinguish fakeness from authentic cultural expression and treating generative AI as simply another tool that people (who are more or less identifiable) might use to achieve something in that online space. Maybe too, in the terms of Markham (1998), generative AI can be thought of as a way of being. This reflection leads, as the next section explores, into the possibility of a more-than-human approach to online ethnography.

Human or more-than-human?

The early methodological writings on online ethnography predate the development of a strand of theorising suggesting that ethnographers should not confine their attentions to humans alone. This more-than-human or other-than-human perspective attempted to shift away from humanist exceptionalism and recognise a broader notion of the social, encompassing social worlds comprised of ensembles of humans and other species alongside material and technological aspects of environments (Lien and Pálsson, 2021). While human concerns remain a core concern of ethnographic writing as Lien and Pálsson (2021: 5) argue, this development “signals a shift from a concern with culture and sociality as a strictly human attribute.” Such concerns have increasingly found their way into methodological writings about online ethnography. Kozinets (2006: 131) was able confidently to assert that researchers could find online communities to study, where people “converse and share ideas,” when writing about netnography before the emergence of social media platforms and the advent of algorithmically mediated online space, Reflecting more recently on the evolution of the approach, Kozinets and Gretzel (2024: 2) somewhat more circumspectly outline the focus of netnography as “systematic, immersive, and multimodal use of observations, digital traces, and/or elicitations,” neatly sidestepping the extent to which people are directly observable and noting that notions of online community became less pertinent with the advent of social media. In replacement of the notion of observing people in their online interactions, Kozinets and Gretzel (2024) position immersion as the key orientation, asking the researcher to reflect on their own experiences of the online space and to engage with non-human interlocutors.

A call for approaches that incorporate algorithms and chatbots as social actors, was developed by Lugosi and Quinton (2018) in their “more-than-human” netnography. Actor Network Theory was drawn on to outline an approach that would attend to the socio-material relations that emerge in the coming together of humans and non-human actors such as apps, devices, platforms and algorithms. Such approaches would take seriously the extent to which human actors (ethnographers included) are enabled and constrained by systems of technological mediation that shape the possibilities for interaction and create the conditions for mutual visibility. Taken to a netnographic strategy, this stance for Lugosi and Quinton (2018) suggests a focus on the emergence of online publics that we would expect from the outset to take a heterogeneous socio-material form. This stance offers a productive strategy for online ethnography in the face of generative AI: the more -than-human approach offers a principled rationale for setting aside a search for “real” human activity and instead remaining agnostic as to the ontological status of the actors within that networked public.

The more-than-human approach suggests that we do not try to sort the human from the non-human according to some external criteria. An ethnographer can take an immersive approach to exploring emergence of agency within the online space. The walkthrough method (Light et al., 2018) takes an Actor Network Theory inspired more-than-human approach to understanding apps as socio-technical ensembles that acquire meaning through use. Using the walkthrough method entails operating with a sensitivity to the array of technical features, symbolic cultural references and interactions that guide one through an app and build meaning in the experience. The walkthrough method is deliberately multi-faceted, open to whatever it is that shapes the experience without prioritising any one aspect a priori. While Light et al. (2018) focused on the app as a particular socio-technical form to outline this approach, arguably it has potential to inform an ethnographic approach to other forms of digital space. Such an approach delivers a new light on the kind of autoethnography outlined in the previous section, highlighting a deliberate embrace of the more-than-human qualities of the online experience. In an online space where generative AI increasingly is an influential actor, this reflexive autoethnographic approach offers the prospect of taking an agnostic, experiential stance on whatever it is that is drawn to the ethnographer’s attention and shapes their experience, regardless of ontological status.

A more-than-human ethnographic approach can entail taking generative AI seriously as an interlocutor. Munk (2023) discusses an approach to finding out how AI sees the world through “hanging out” with it, engaging with it curiously and exploring what cultural threads are drawn into its representations. In similar vein, De Seta et al. (2024) engage with AI as a conversational partner with an aim of probing its ways of seeing the world through experimental engagements. Such approaches treat generative AI as a cultural being, with connections and history and a positioned way of seeing the world. Strategies of this kind are strong contenders as components of a more-than-human ethnography of generative AI, but it is important to recognise that such enquiries will not answer all of the questions one might have about how generative AI arrives at its representations and we will be left with many intractable uncertainties. The strength of these strategies is more in the spirit of actively engaging with a cultural territory in which human and machine are entwined than it is about debunking or revealing once and for all what is what.

Marres (2017) offers a framework for thinking about what is at stake, methodologically, if we take a more-than-human approach to digital sociality. Marres (2017) identifies that researchers venturing onto digital platforms for their research face a challenge posed by the ambiguity of the phenomena they witness: are these phenomena to be treated as social objects, such as communities, conversations and publics, or are they technologically-mediated phenomena to be considered as features of the digital platforms where they arise? The question of whether we are therefore studying technology or society occasions, for Marres (2017), three orders of response. First, the researcher might adopt an approach of critical extraction, seeking to sort out the different kinds of data and bracket off from analysis the contributions of chatbots to a Twitter thread for example as somehow separate from the social phenomenon being studied. As outlined in previous sections, this may increasingly be impractical. Second, Marres (2017) suggests that the researcher might embrace the performative role of the technology in constituting social phenomena and accept that their observations will be hybrid concoctions of technological and social. Chatbot contributions, according to this perspective, deserve to be taken seriously as constitutive parts of a Twitter conversation. This aligns somewhat with the ethnographic stance on authenticity as participant category outlined in the previous section. Thirdly, Marres (2017) advocates for an empiricist approach that does not seek a definitive answer to the question of whether we should seek to extract social phenomena from digital platforms or instead embrace their inherent socio-technical hybridity. The empiricist approach keeps this as a live question considering what is at stake for a given research endeavour by switching between perspectives.

Embracing the ambiguity of the situation and remaining agnostic as to how to understand the nature of the phenomena they are immersed in might be seems a plausible tactic for an ethnographer to take in the face of the upsurge in generative AI, suggesting a need to sustain a reflexive attention to what is at stake as they do so. While we may aim to take an immersive approach, as Kozinets and Gretzel (2024) advise, it is important for the ethnographer to explore how their perspective attunes with others within and beyond the setting if we are not to fall foul of the criticisms of autoethnography for a self-indulgent neglect of other ways of seeing the world (Collinson and Hockey, 2005). Crucially ethnography remains active and interactive, focused on gaining an immersive experience rather than scraping and capturing material for some distanced form of analysis. Ethnography is active, in that the ethnographer is consciously creating a field through deliberate moves aimed at enlightening understanding of the object. It is interactive, in that the ethnographer does not sit in silence with their own thoughts but exposes their emerging understanding to mutual scrutiny and challenge with interlocutors in the situation. In terms of the empiricist approach outlined by Marres (2017) there can be a dialogue between what appears to be at stake for the ethnographer’s imagined research objective in the treatment of the socio-technical aspects of the phenomenon, and what appears to be at stake from the point of view of a wider array of interlocutors. Where do these perspectives align and where do they differ? What is at stake for the ethnographer who chooses to insist on the hybridity of digital social phenomena in the face of an insistence from other quarters on the moral necessity of distinguishing human from machine-generated or, indeed, on allowing such ambiguities to remain cloaked? An ethically reflexive form of ethnographic knowing in such circumstances would need to be conscious of the consequences for the ethnographer’s project and their relations with the social phenomena under examination and to be cautious about taking a revelatory stance on what is “really” going on in any situation. We would do well to recall the partial, situated and provisional nature of ethnographic enquiry and embrace the multiplicity of knowledge.

Conclusion

This paper began with a deliberately provocative title, echoing a prevalent mood of concern that radical social change is underway occasioned by the recent remarkable advances in the capacities of generative AI to mimic other human cultural forms. To the extent that content generated by AI is both increasing in amount and increasingly hard to detect, it now seems more than ever implausible to attempt a realist, human-centred online ethnography that seeks only to focus on what humans say and do. Nonetheless, in the existing methodological literature relating to online ethnography there are grounds to reject a methodological exceptionalism for generative AI and to hope that online ethnography is far from at an end. Across an array of different methodological positions we find reasons not to abandon online ethnography but instead to continue to creatively adapt its approaches to enable interrogation of online contexts that are comprised of material of a variety of intermingled human and machine-generated forms.

Examining the methodological literature relating to authenticity in online ethnography, a variety of strategies applicable to the contemporary situation are identifiable. We can treat authenticity as a culturally embedded category, to become a topic for ethnography in so far as it becomes a topic within the setting. Generative AI, then, becomes a matter of concern as and when attention is drawn to it in the setting. If participants in that setting express concerns about it and develop skills in its detection then the ethnographer may wish to do likewise. Methodological choices here focus around the extent to which the ethnographer is aiming to align their own skills and understanding with those in the setting. Attention may be drawn to emergent hierarchies in relation to authenticity, skills and trust. Multi-sited approaches may allow exploration of the ways in which different settings are cross-contextualising, to the extent that material which is trusted in one setting may become an object of suspicion in another or instead acquires new authority as it circulates and transforms. Autoethnographic approaches draw attention to the experience of navigating online domains comprised of materials of uncertain origin. Across the board, reflexivity is required about what counts as an authentic observation and for whom and where generative AI intersects with these judgements.

More-than-human approaches to ethnography explicitly embrace the hybrid nature of sociality and provide inspiration for online ethnographies that recognise the intertwining of machine and human agencies through platforms, apps and devices and in autonomous agents and algorithmic feeds. Such approaches develop an immersive understanding, sometimes by directly probing and engaging with AI in the interests of developing a deeper understanding of how it sees the world. Again, a reflexive awareness of what is at stake across these various methodological choices is important, in order to be clear about the consequences for the claims to be made for the scope of the ethnography. A more-than-human ethnography in online spaces should be able to paint an evocative picture of this emergent cultural space, but it will come up against intransigent silences regarding exactly who, or what, is doing what with whom. Continued reflection is needed on how and to whom a distinction between human and AI-generated content might matter, what practical or moral weight the categories seem to bear and what we learn about emergent power dynamics through our thwarted efforts to make sense of these issues. Some of the puzzles of AI-generated content may be resolvable through forms of ethnographic enquiry involving engagement with AI as interlocutor, but often this will not be the case. Given the opacity of platform technologies, we may be left with profound uncertainties both about the extent to which AI is involved in any given situation we find online and the nature of its reasoning where we do identify it.

It is important to maintain efforts to engage ethnographically with digitally-mediated experiences because they are so profoundly embedded in contemporary everyday life. This paper has suggested that we have at least some of the methodological toolkit needed to do so. This is not, however, to be taken as suggesting that we do not need innovation. Ethnography is fundamentally an adaptive approach in which the research design is developed in conjunction with the conditions encountered in the field and the ethnographer’s evolving understandings. Each ethnography is methodologically innovative in its research design just as every field site is unique. In innovating, ethnographers develop strategies that draw from and contribute to an evolving methodological toolkit. The advent of generative AI may indeed give rise to further methodological innovations, such as AI-assisted approaches to fieldwork, shifts in our understanding of interviewing or new forms of multi-sited fieldwork. Our conceptual understandings of what an adequate research design is will continue to evolve. The practice of ethnography in contexts suffused by generative AI should provoke continued attention and reflection in ethnography’s methodological literature.