Daejeon, July 27–31
The accelerating proliferation of AI-generated images has transformed the way that we encounter visual media. The development of increasingly more powerful image generating tools (Figure 1) has lead to the creation of synthetic images now visually undinstinguishable from photographs, allowing synthetic images to circulate unmarked among journalistic photographs and political imagery. In this context, deceptive synthetic media poses significant civic challenges, destabilizing public confidence in informational infrastructures and increasing susceptibility to influence operations (Vaccari & Chadwick 2020; Bonfanti 2020). To tackle this crisis of veracity, industry and academic responses have predominantly pursued technical detection mechanisms ranging from AI-powered classifiers (Sadasivan et al. 2023) to deepfake detection architectures (Rana et al. 2022) and diffusion-forensic methods (Corvi et al. 2023). While promising, these systems remain fragile and perpetually one step behind generative models in constant evolution.
Nonetheless, forms of resistance are emerging as platform users debate and contextualize online media. Far from following an organized strategy, users engage in ad-hoc, improvised practices of cross-examining source material, comparing versions across posts, or annotating misleading claims. These actions reveal a growing reliance on contextual reasoning and distributed knowledge rather than on automated detection. What initially appears as scattered user reactions points to a broader shift: digital publics are beginning to take on interpretive roles traditionally reserved for experts, developing collective methods for assessing provenance and credibility in the absence of reliable technical safeguards.
Drawing on a qualitative analysis of 115 social-media posts debating the authenticity of visual content, we show that users increasingly revive and adapt humanistic techniques traditionally associated with art-historical connoisseurship. Once the purview of credentialed experts, these methods now unfold collectively within platform infrastructures and rely on distributed knowledge rather than singular elite authority. What appears as ad-hoc resistance to deceptive synthetic media in fact represents a broader revival and reconfiguration of connoisseurship: long-standing interpretive practices refined over centuries are being recreated in naïve, grassroots forms among digital publics, who deploy them to evaluate authenticity, provenance, and context in environments lacking reliable technical safeguards.
We introduce the concept of cultural connoisseurship to describe this emergent, participatory mode of authenticity assessment. Understanding this new mode of expertise is essential for designing the structural and epistemic conditions that can support it, while also mitigating the historical limitations of traditional connoisseurship, including its elitism and susceptibility to bias. Our study contributes an initial conceptual model of cultural connoisseurship and outlines the challenges and future directions necessary for advancing research in this field.
Figure 1. Side-by-side comparison of the output of image-generation models Nano Banana and Nano Banana Pro, posted by Twitter user @immasiddx on 24/11/2025.
Traditionally, art historical connoisseurship aims to determine the “work-identifying facts of production”: the who, when, and where of an artwork’s making (Brown, 2008). In its original sense, the act of authentication seeks to clarify authorship, context, and materiality of an artifact. The practice of assessment can be divided in two domains: scientific connoisseurship, which establishes authenticity through physical testing, material analysis, and dating methods; and connoisseurship of the eye, which relies on an expert’s trained intuition and accumulated knowledge. Typically, the expert’s discernement builds on three complementary facets: (a) historical traces, such as archival records or documented provenance; (b) interviews and authoritative sources, including expert testimony or institutional verification; and (c) physical clues, including brushwork or material composition. Together, these elements allow the connoisseur to make an expert judgement over aspects of authorship and production conditions.
Yet the role of the connoisseur as a singular authoritative expert has been extensively debated in art historical literature. For instance, research on seventeenth-century workshop practices demonstrates that authorship has rarely been singular or easily demarcated. Early modern distinctions between originals and copies often reflected fluid, collaborative, or partially delegated labor, complicating the assumption that authorship necessarily corresponds to a single hand (Tummers 2011).
Additionally, connoisseurship has been critiqued for its innate social and institutional elitism. As Reyburn (2021) notes, it emerged as an exclusionary “boys’ club” reinforced by institutional gatekeeping. Crucial to expert status are certification, recognition, and discriminatory acuity (Shanteau 2002). For centuries, these conditions restricted who could participate in the production of authenticity judgments, sustained by the innate elitism of the art world: a domain that defined itself in contrast to “the masses”, maintaining information asymmetries to secure expert authority (Arora & Vermeylen, 2012).
However, contemporary digital environments are reshaping these epistemic arrangements, leading to what has been described as the end of the connoisseur. Today, the immediate availability of contextual information - past posts, metadata, reverse-image searches, collective memory, journalistic reporting - reduces the information asymmetries that once made such context accessible to experts only. Instead of relying on the authority of a single specialist, users can rapidly assemble, compare, and debate evidence, drawing on distributed knowledge that was previously inaccessible or siloed. More significantly, today’s platforms create what Bowker et al. (2010) describe as new “information architectures”, or sociotechnical systems that structure how knowledge is produced, circulated, and legitimized. Within these architectures, we argue that connoisseurial methods are not disappearing but reconfiguring and expanding how authenticity judgments are collectively produced.
We look to social media posts to analyse the practices of examination of synthetic content in public digital spaces (e.g. Twitter), or dedicated communities (e.g. AI identification subreddits). Specifically, we build a corpus of 115 posts (50 top posts on /r/RealOrAI/ and / r/isthisAI/ and 15 hand-picked Twitter posts). Based on a qualitative analysis of these posts we map current practices of cultural connoisseurship to the three practices of connoisseurship examination: historical traces, interviews and authoritative sources, and physical clues. Our analysis shows that everyday assessments of synthetic media rely primarily on historical traces and physical clues, while appeals to authoritative sources are concentrated in cases involving politicians and other high-profile figures. In addition, we identify specific characteristics of the environment of cultural connoisseurship (Figure 2):
Synthetic media literacy. Users produce guides identifying the signs of AI-generated images such as unusual anatomy, blurred boundaries and inconsistent lighting. These training materials function as crowd-sourced pedagogy: a collective “training of the eye” that democratizes visual heuristics.
Collective moderation. Moderation tools such as ‘community notes’ allow users to append contextual information to unmarked synthetic media. These notes operate as crowd-sourced provenance work: tracing origins, documenting transformations, and situating images within broader social or historical narratives.
Proactive authenticity assertions. Creators increasingly record and share process videos or livestream their creations to preempt accusations that their work was AI-generated. These self-produced ‘provenance archives’ blur the line between evidence and performance, providing material testimony of authorship in a context where such claims are increasingly precarious.
Together, these practices show that digital publics are not inventing new epistemic logics so much as reviving and reconfiguring the core strategies of connoisseurship. Authentication becomes a communal activity characterised by its scale and participatory nature. At the same time, new challenges arise with the distributed, decentralized, and anonymous nature of these ad-hoc practices.
Figure 2a. User creates a guide to spotting AI Generated images. 2b. ‘Community Notes’ allows users to collectively flag misleading content with contextualization notes. 2c. User proactively records creation process to prevent risks of being accused of publishing deceptive AI art.
The emerging practice of cultural connoisseurship faces two interlinked challenges: scale and selective attention. First, the volume of synthetic media vastly exceeds the capacity of users to examine it. While some posts attract heightened scrutiny, most synthetic content likely circulates unnoticed. Second, community attention often clusters around figures with high visibility: celebrities, political figures, or other actors to whom audiences feel a prior investment. Those without visibility often struggle to protect their digital identities and creative processes. In this sense, cultural connoisseurship reproduces the blind spots of traditional connoisseurship.
At the same time, emerging forms of expertise are deeply shaped by platform architectures. Research in HCI demonstrates how digital affordances condition how users collaborate, moderate, and produce knowledge (Shibuya 2022; Petterson 2023; Brewer 2020). These affordances determine what kinds of evidence users can provide, which claims are amplified or suppressed, and how communal interpretations coalesce or fracture.
These challenges highlight the need to consider cultural connoisseurship as an emerging object of study. In the face of unreliable technical detection, audiences rely on informal attempts of contextual reasoning and shared knowledge to negotiate the credibility of visual media. Understanding how these interpretive practices function is essential for designing environments that support these public efforts.
The emergence of cultural connoisseurship signals a transformative shift in how authenticity is constructed and collectively negotiated. Based on a qualitative analysis of 115 social media posts, and by situating these emergent practices within the historical genealogy of connoisseurship, this paper shows that contemporary authentication work represents a reconfiguration of long-standing practice of connoisseurship in art historical scholarship. The techniques through which publics assess provenance, context, and veracity revive the core logics of connoisseurship, reconfiguring them as communal practices. This reframing underscores the crucial role of humanistic insight in shaping the tools, and infrastructures necessary to sustain veracity in the synthetic age.