Beyond disruption and invisibility: Interactional continuity in everyday AI use in India

Abstract

Research on the social significance of Artificial Intelligence (AI) in journalism, communication, and organizational contexts is often organized around two emphases: one foregrounds disruption, framing standalone Generative Artificial Intelligence (GenAI) tools as transformative; the other foregrounds invisibility, treating embedded AI as backgrounded. We bridge these emphases with interactional continuity, a user-centered frame that explains cross-modality incorporation at the level of task episodes: AI use stabilizes through sequential placement in ongoing activities, while interface packaging shapes whether it is perceived and named as AI. Grounded in infrastructure theory, platform studies, and digital meaning-making, we examine everyday AI use in the Global South through 28 semi-structured interviews in an Indian workplace. We used iterative AI-assisted and manual coding to identify patterns. Importantly, we found that respondents used AI as episode-level task support within familiar platform routines, illustrating how disruption and invisibility can emerge from task sequencing and interface cues.

Keywords

artificial intelligence platform studies embedded artificial intelligence generative AI interactional continuity global south

Introduction

Research often treats Artificial Intelligence (AI) through two recurring analytical emphases.¹ In journalism, communication, and organizational scholarship, standalone Generative Artificial Intelligence (GenAI) tools have been framed as disruptive technologies and part of an institutional “AI turn” (Dodds et al., 2025; Lewis et al., 2025; Schaetz and Schjøtt, 2026). Science and Technology Studies (STS) and media scholars emphasize the invisibility of AI, locating it within infrastructural systems and institutional arrangements through which it operates while remaining consequential even when it is not directly perceptible to users (Ananny and Crawford, 2018; Seaver, 2017; Star and Ruhleder, 1996; Suchman, 2020).

These divergent framings are further complicated by the fact that AI is encountered in different forms. Some scholars treat AI as a broad category, while others disaggregate it into different types depending on visibility and use. We use AI as an umbrella for automated data-driven systems (Information Commissioner’s Office, n.d.; Zajko, 2022) and we distinguish embedded AI (i.e. feature-level functions integrated within platform interfaces) from standalone GenAI tools (i.e. prompt-based systems accessed as separate applications, such as ChatGPT or Gemini in their standalone forms) (Gillespie, 2014; Helberger, 2021). Importantly, the same underlying system may operate in both forms depending on how it is accessed (e.g. ChatGPT as a standalone application versus similar models embedded within platforms).

To address this gap, we introduce the concept of interactional continuity. Drawing on infrastructure theory, platform studies, and digital meaning-making (Murthy, 2022; Star and Ruhleder, 1996; Van Dijck et al., 2018; Yu, 2024), this concept explains how users integrate different forms of AI into ongoing task-based routines. Prior research has examined algorithmic bias and invisible labor (Gray and Suri, 2019; Noble, 2018), as well as AI governance, creative labor, and users’ relations to standalone GenAI tools (Murthy, 2024; Park et al., 2024, Jensen et al., 2025), but has not fully accounted for how AI becomes actionable across contexts of use.

Rather than treating AI as either disruptive or invisible, interactional continuity identifies the episode-level mechanism through which AI becomes consequential in practice. Within task-based routines, users render AI actionable through task sequencing and interface packaging across both embedded and standalone tools. This mechanism operates through three dynamics: the incorporation of embedded AI within task episodes; the everyday use of standalone GenAI alongside embedded AI beyond launch moments; and the primacy of task fit and performance over explicit naming.

AI is extensively studied in the Global North, specifically the U.S. and Europe, while research in the Global South remains limited and underrepresents the communities these systems target (Molla and Ahsan, 2025; Okolo et al., 2022). Furthermore, major technology companies are increasingly deploying AI in the Global South (Okolo et al., 2022), raising concerns about infrastructure capture and corporate influence over local innovation (Molla and Ahsan, 2025). We address this gap through a case study of AI use in India. Drawing on 28 interviews in Bangalore, we examine how users perceive and use embedded AI and standalone GenAI, and how purpose and recognition compare across modalities.

We find that respondents treat embedded and standalone GenAI tools as functional extensions of familiar activities, illustrating the dynamics of interactional continuity. In this workplace setting, AI uptake clusters at repeatable insertion points within task episodes. Task fit supports reuse, while interface packaging shapes whether tools are recognized and named as AI.

Literature review

Across journalism, communication, and organizational scholarship, GenAI is frequently positioned through zeitgeist frames that emphasize disruption, framing AI use as a rupture in everyday social and work life (e.g. Dodds et al., 2025; Hinds and von Krogh, 2024; Holmström and Carroll, 2024; Hosseini et al., 2025; Lewis et al., 2025; Usher, 2025). By contrast, platform and infrastructure studies foreground the invisibility of AI, locating it in the background operations of defaults that organize what users encounter, the moments of breakdown that make hidden systems visible, and the ongoing maintenance that keeps those systems appearing seamless (Petre, 2021; Plantin et al., 2018; Rader and Gray, 2015; Star and Ruhleder, 1996). When treated separately, these perspectives often starkly diverge by treating GenAI use either as a noteworthy rupture or, on the other hand, as total background mediation. However, this bifurcation can be avoided by examining how both are encountered within the same task episodes.

That being said, this split does help explain why critical AI, STS, and media scholarship treat AI as sociotechnical assemblages (Latour, 2005), showing how practice, institutional power, and governance shape what AI is and does, and why visibility and transparency are not straightforward remedies for accountability (Amoore, 2020; Ananny and Crawford, 2018; Seaver, 2017; Suchman, 2020).

This tension between disruption and invisibility highlights an unresolved problem: how AI becomes actionable within everyday platform practice, particularly across different modalities of use (Christin, 2020; Molla and Ahsan, 2025; Siles, 2023).

Interactional Continuity as episode-level mechanism

We conceptualize interactional continuity as an episode-level mechanism by which AI becomes routine when it enters repeatable task steps (e.g. drafting, searching, and refining) and is judged by whether it moves users to the next step. It explains stabilization without requiring sustained attention or explicit labeling as AI. Building on critical AI, STS, and media scholarship, which treats AI as practice-bound and power-laden rather than a self-evident object (e.g. Seaver, 2017; Suchman, 2020), interactional continuity shifts accountability from abstract visibility to the specific episode-level moments where systems become consequential.

Analytically, interactional continuity has three dimensions: task fit (whether a tool moves the activity forward), sequential placement (where it enters the episode), and interface packaging (whether it appears as a background default or as a distinct assistant or chatbot). These dimensions specify the insertion mechanics participants narrate when describing how AI becomes usable within concrete task episodes. Other factors emphasized in the literature (e.g. attitudes, trust, AI literacy) may shape broader orientations over time, but they do not as directly explain why a tool is selected at a particular step and recognized (or not) as AI in that moment.

The strength of the mechanism varies by task. Interactional continuity is most visible in repeatable, readily measurable work where users can iteratively judge output adequacy at each step (e.g. rephrasing, drafting, and quick information checks). It is less likely when tasks require verification, accountability, or high-stakes consequences, or when institutional constraints and access conditions limit experimentation.

Empirically, interactional continuity should be visible in four recurring patterns in interview narratives of use: (1) repeatable insertion points within episodes (e.g. draft → revise → send; query → select → act); (2) augmentation over replacement, with AI producing first-pass outputs and users reclaiming control through editing and selection; (3) naming–use decoupling, where assistant-style packaging increases explicit AI labeling even when embedded features are used more frequently than standalone tools; and (4) switching by friction, with movement toward embedded assistants when entry/exit costs of standalone tools are higher.

Interactional continuity is adjacent to domestication, practice theory, and appropriation/use-script accounts, but specifies a different analytic object: an episode-level, task-based insertion mechanism (Akrich, 1992; Berker et al., 2005; De Certeau and Rendall, 1984; Hirsch and Silverstone, 2003; Leonardi, 2011; Orlikowski, 2000; Schatzki, 2002; Silverstone and Haddon, 1996). It also differs from Caspi et al.’s (1989) person-level notion of interactional continuity as a stable pattern across the life course. Rather than tracing long-horizon routinization or imbrication, we focus on narrated selection and evaluation points within task episodes, which helps explain cross-modality incorporation and why some heavily used AI remains unnamed.

Empirical research across domains is consistent with the episode-level mechanism proposed by interactional continuity. Rehman and Khalil (2024) find that Pakistani students used Meta AI in WhatsApp study groups as an extension of familiar communication. In information systems research, Buana et al. (2024) find that users evaluate outputs against task expectations within ongoing sequences of work. Journalists use standalone GenAI tools within editorial workflows without changing sourcing (Dodds et al., 2025; Zhang et al., 2025). In creative settings, Laba (2024) shows how Midjourney users treated prompting as a learned interface skill rather than as a novel form of experimentation. Across these domains, users retain AI in their social and work lives when it produces a successful next-step output and abandon it when it does not, suggesting a common logic of episode-level insertion rather than wholesale workflow redesign.

Because platform design and policies tend to package AI as familiar and low-friction, they reinforce interactional continuity at the interface level. For example, Su and Chan (2025) show how OpenAI’s design and deployment choices are translated into platform governance frameworks that encourage assistant-style integrations over standalone, agentic deployments framed as disruptive. Ferrari et al. (2023) and De Seta et al. (2024) conceptualize standalone GenAI as embedded in infrastructural arrangements, while Grandinetti (2023) and Yu (2024) document how platforms frame AI as helpful, easy-to-use assistants folded into existing workflows. Interface packaging thus changes how the same ecosystem is encountered, sometimes as an event, sometimes as a default, helping explain why disruption and infrastructure scholarship can read it differently without treating AI as inherently one or the other.

AI as novelty: A critical review of disruption

Scholars conceptualizing GenAI through disruption/turn frames propose launch narratives and hype cycles that position systems like ChatGPT as novel breaks from earlier forms of automation, foregrounding rupture as an analytic lens (Dodds et al., 2025; Lewis et al., 2025; Usher, 2025). Related accounts describe these tools as unprecedented in function and anticipated impact (Cohen et al., 2024). Scholars also note that product-cycle announcements and “next-generation” model discourse circulate as rupture rhetoric, shaping expectations around what AI will change² in ways that align with sociotechnical imaginaries and future-oriented promises that organize action and legitimacy (Borup et al., 2006; Jasanoff and Kim, 2019). Cinque (2024) analyzes this dynamic as a medialogical event, staged through surprise and experimentation. These disruption-oriented framings treat GenAI’s significance as an eventful rupture, often foregrounding novelty over continuity in everyday use.

Countervailing scholarship illustrates how innovation talk collides with bureaucratic routine (Sloane, 2024), and how AI hype can shape organizational initiatives and decision-making, particularly under uncertainty (Schaetz and Schjøtt, 2026). More broadly, performativity theory explains how the initial spectacle of “the new” is rapidly folded into habit, attenuating the novelty of the technology over time (Couldry and Hepp, 2018; Rogers et al., 2014). User studies corroborate this pattern. Rather than viewing AI as a disruptive force that reorders activities or displaces existing tools, users develop complex mental models and relationships with it, framing it as a tool, a companion, or an advisor depending on their needs (Park et al., 2024). Similarly, Li and Zhang (2024) find that trust in standalone GenAI emerges from perceived care, responsiveness, and emotional resonance. Chatbots were described as encouraging or motivating, showing how affective interactions shape meaning. While disruption rhetoric is prominent in scholarly and public discourse, empirical accounts of practice often show rapid incorporation through task-episode fit and episode-level placement within sequences of action.

AI as infrastructure: A critical review of invisibility

Scholars conceptualizing AI as infrastructure (often operationalized as embedded algorithmic mediation) shift attention from launch narratives to the ambient systems that structure interaction, often without requiring explicit recognition (Plantin et al., 2018; Star and Ruhleder, 1996). This strand foregrounds backgrounded mediation through defaults, maintenance, and the way breakdowns make infrastructure legible. Embedded AI exemplifies this logic through autocomplete, predictive typing, ranking, and curation, which steer interaction through familiarity. Applied studies of platforms likewise describe AI as pervasive across systems, yet difficult for users (and even practitioners) to clearly locate or name (Jones et al., 2022). As Star and Ruhleder (1996) argue, infrastructure becomes visible only when it breaks, explaining why many AI functions operate below user awareness. Related work shows that when algorithmic behavior is consequential, it can remain experienced as an “invisible” part of system infrastructure that users infer (Petre, 2021; Rader and Gray, 2015).

Previous conversational search work also reveals complex information-seeking unfolding inside familiar messaging platforms (Radlinski and Craswell, 2017; Zamani et al., 2023). This takes multiple forms: interface politics that shape what becomes clickable (Yu, 2024), neutral-sounding personalization and rationalization that obscure automated inference (Simon, 2025), and probabilistic calibration of engagement that subtly guides user activity (Törnberg et al., 2025). These accounts show how algorithmic systems cultivate credibility through opacity, shaping behavior as a background condition of interaction rather than an explicitly recognized feature.

This normalization is hardly neutral. AI algorithms can magnify bias (Noble, 2018) while concealing the “ghost work” required to keep systems running (Gray and Suri, 2019), labor often performed in the Global South under exploitative conditions (Ardévol and Gómez-Cruz, 2014). Embedded inference can also preempt choice, while predictive algorithms infer sensitive traits related to politics, health risk, and emotional states, without explicit user input (Mühlhoff, 2023). Consequently, embedded AI operates as a form of habituated intelligence in everyday contexts, where the extraordinary becomes ordinary through routine use.

Overall, infrastructural and invisibility accounts treat embedded AI as backgrounded systems whose defaults (autocomplete, ranking, and curation) guide behavior without sustained attention. Most people notice these types of AI only when they fail. This invisibility has consequences: it normalizes interventions that can narrow choices, enable predictive profiling, propagate bias and social inequalities, and rely on hidden and exploited human labor (Gray and Suri, 2019; Mühlhoff, 2023; Noble, 2018; Star and Ruhleder, 1996). This work explains how embedded systems recede into everyday platform practice, but it explains less clearly how AI modalities enter specific stages of a task and how interface packaging cues their placement, an analytic gap that interactional continuity addresses.

AI in the Indian context

Though social media platforms mediate much of daily communication in India, embedded AI features are rarely explicitly framed as AI (Behera and Gartia, 2024; Fahad et al., 2023), reinforcing their role as background mediation. Policy and industry-facing national discourse frames GenAI as transformative (Wang and Downey, 2025). Yet sectoral evidence shows incremental augmentation rather than systemic overhaul: in education, capacity and uptake limits constrain deployment (Karan and Angadi, 2025); in healthcare, AI expands coverage and diagnostics without replacing existing systems (Bajpai and Wadhwa, 2021); while implementation across domains is conditioned by linguistic diversity, training-data bias, and digital inequalities (Khalid et al., 2025). This literature foregrounds a tension between AI as ambient infrastructural mediation and AI as disruption rhetoric, an interpretive gap that matters for how AI becomes recognizable, actionable, and reusable in everyday practice.

Research questions

To examine interactional continuity across AI modalities, we ask three research questions:

Research Question 1: How do users interact with and perceive embedded AI within platforms?

Research Question 2: How do users interact with and perceive standalone GenAI tools?

Research Question 3: How do interaction and perception compare across embedded AI and standalone GenAI tools?

The following analytic propositions flow from our definition of interactional continuity and guide our analysis:

Proposition 1: Naming AI will correlate more strongly with chat-style interface packaging than with frequency of use.

Proposition 2: Reuse stabilizes when AI advances the next step in a task episode, even if users don’t label it as AI.

Proposition 3: When AI is inserted into repeatable, task-advancing points, incorporation tends to augment existing workflows rather than replace them.

Data and methods

We conducted 28 semi-structured interviews in Bangalore, India, in July 2024, a month after Meta began testing and then broadly rolling out its Meta AI assistant into popular social media apps in India (Meta, 2024). With WhatsApp and Instagram among India’s most widely used platforms (Chougule et al., 2023), participants encountered Meta AI primarily as a chat-based assistant. ChatGPT had been publicly available since late 2022, but Meta AI introduced chat-style GenAI into messaging apps used by a wide range of socioeconomic groups in India (Mukhopadhyay, 2023). This timing allows us to examine how task fit, sequence placement, and interface packaging shape incorporation and naming during early use. We use a mixed-methods approach (see Figure 1), which combines four sequential stages: qualitative interviews and transcription, AI-assisted coding, manual coding, and qualitative analysis.

Figure 1.

Mixed methodological approach.

Stage 1: Qualitative interviews and transcription

We conducted interviews over 2 days at an Indian company with business interests related to marketing, sales, and real estate. We chose this workplace because employees with diverse occupational roles relied on ubiquitous apps for everyday communication and information-seeking, making it a useful setting to examine AI use in routine, time-sensitive tasks (e.g. drafting messages, coordinating logistics, quick searches). Respondents (22–56; mean 34) included 17 men and 11 women across sales, management, marketing, design, administration, and internships. Educational backgrounds ranged from secondary education/diploma training to undergraduate degrees and, for a smaller subset, postgraduate credentials.³ All employees were invited via an email from human resources emphasizing voluntary, confidential participation. Recruitment reflected availability. We treated education and role (Supplemental Table 2) as sensitizing dimensions and, during constant comparison, checked whether task-fit, placement, and packaging patterns varied across groups. Given the single-site and small sample of multilingual, English-speaking respondents, we report these as tendencies and not robust subgroup claims. All participants provided informed consent under an Institutional Review Board (IRB) approved protocol (STUDY00004601, University of Texas at Austin). We concluded the interviews once they reached saturation – that is, additional sessions yielded no new themes. Participants received no monetary compensation. Interviews were conducted in English with no supervisors present and averaged 22 minutes.

All interviews used a semi-structured guide with a shared question set. Questions first mapped routine platform use, then asked about embedded AI features, and, finally, about standalone GenAI. Episodic recall prompts elicited concrete sequence-level accounts: participants “walked through” a recent AI use episode, what prompted it, what they entered, how they evaluated the output, and what they did next (e.g. edit/copy/send/stop), with probes locating where AI entered the activity. We did not assume a shared technical definition of AI. Instead, we introduced the term as a conversational prompt alongside concrete tools (e.g. ChatGPT, Meta AI, and algorithmic recommendations) and invited participants to define and exemplify it in their own terms. Thus, AI in the data reflects participants’ own labels and proxies, enabling recognition and naming to be analyzed empirically.

We used Microsoft Word 365 Enterprise to machine transcribe interviews and add timestamps using the Indian English setting to maximize transcription accuracy. This software offers high levels of privacy protection, adhering to US Health Insurance Portability and Accountability Act (HIPAA) Security Rule requirements. To enhance protections, we saved the machine transcripts with pseudonyms. Only the authors had access to recordings and de-identified transcripts, which were stored in encrypted, password-based, access-controlled university folders. Random segments of transcripts were checked against audio recordings for accuracy.

Stage 2: AI coding

We developed a preliminary codebook based on prior literature (e.g. Mukhopadhyay, 2023) and initial transcript readings. We then imported verified transcripts into ATLAS.ti (Mac v25.0.1, build 32922) and applied a hybrid coding process combining AI-assisted tagging with researcher-led refinement. Following Williamson et al. (2025), we used the Coding Assistant with default settings (English; automatic application disabled) and ran separate AI-coding passes for each research question. The first author defined a coding question and label, prompted the assistant to highlight relevant segments, and then accepted, edited, or rejected every suggested quotation while writing brief memos (see Supplemental Appendix A for abridged examples). We piloted the codebook on five transcripts, refining definitions through memoing and discussion between authors. Because all AI-suggested segments were hand-checked and then manually re-coded in Stage 3, we did not compute separate precision or recall metrics.

Stage 3: Manual coding

Based on AI coding, we refined our analytic focus around four focal dimensions: feature type, purpose of use, recognition, and usage frequency. Then, we manually re-coded all interviews. Each AI use was coded per instance to capture intra-user variation. The revised codebook was pilot-tested for consistency and carried out by the first author, so we did not calculate inter-coder agreement statistics.

Stage 4: Qualitative analysis

Following Murthy’s (2022) framework, findings were integrated into a refined theoretical narrative, grounded in participant accounts and a sociotechnical lens. Working with the finalized coded dataset, the first author synthesized memos at the code and case level to identify themes in how participants engaged with embedded AI, embedded GenAI, and standalone GenAI tools. We then compared these patterns side-by-side to trace points of continuity, divergence, and exception across modalities, returning to full transcripts to preserve contextual nuance. This comparative analysis allowed us to interpret the tendencies observed in previous stages. The three dimensions were developed abductively through iterative engagement with theory and data. Infrastructure and platform scholarship sensitized us to interface design and backgrounded mediation, while patterns in interview narratives (repeatable insertion points; naming, use decoupling) guided iterative refinement through constant comparison.

Results

We present findings by research question: embedded AI (RQ1), embedded and standalone GenAI (RQ2), and cross-modality comparison (RQ3), emphasizing task-episode placement and interface packaging. Patterns were consistent across roles and education; given our small sample of single-site self-reports, we treat any observed differences as tendencies rather than generalizable claims.

Utility in practice: How users adopt AI for specific tasks

While some exploratory use occurred, respondents primarily adopted AI tools to fulfill recognizable tasks. From the 709 interview excerpts, we identified five purposes of use: content recommendation (AI-curated material surfaced while scrolling), search assistance (AI-supported information seeking), conversational information retrieval (digital assistant-style Q&A), content creation (AI-generated text or images), and text assistance (drafting, rewriting, or editing with AI). Use patterns varied by task and platform (see Supplemental Table 3 for representative excerpts).

Using embedded AI for task fulfillment

For embedded AI, use clustered around two purposes: content recommendation (82.1%; n = 23) and search assistance (78.6%; n = 22). These features operated as part of routine task episodes within platform use, entering through defaults (feeds, rankings, search) rather than as a distinct tool participants sought out.

Aakash (32) described Instagram content recommendations becoming integral to his weekend planning. Initially subscribing to location-based channels, he later relied on suggestions:

I am fond of going out to many places, so I used to subscribe to those channels, like places around Bangalore. [. . .] Instagram would show me nearby hotels, restaurants, places to go. I just check it on weekends before heading out.

Discovery merges with recommendation logic, allowing AI-based feeds to supplement everyday choices. Even when Amit (26) reduced his Instagram use, the platform continued to serve tailored content: “The pages that I follow, those are what I see. There are a few ads in between too, and if I like them, I follow. Mostly I just scroll through what’s already tailored to me.” This case shows how AI recommendations use accumulated behavior to stay relevant and shape perception with minimal input.

Dilip (29) exemplified search assistance through his daily Instagram routine:

I scroll for maybe 15 minutes a day, but mostly I search . . . I look up zoology, Bangalore zoos, hotels, the automobile industry, especially Indian brands. I prefer searching because recommendations just show everything, even reels I’m not interested in. I want specific, useful information.

He treats AI as a steerable tool, adapting to platform limits and using search to bypass recommendations that “show everything.” These cases show three orientations: absorption (Amit), strategic use (Aakash), and calibration (Dilip), as respondents accept, mobilize, or tune recommendations based on utility. Even habitual use involves selective negotiation with platform constraints.

Using embedded GenAI for task fulfillment

Embedded GenAI is generative in function, but its blue-circle chat packaging inside widely used Meta apps in India (e.g. WhatsApp and Instagram) shaped how participants encountered it, embedded in ongoing platform practice, lowering adoption barriers and aligning with existing task expectations. Respondents mainly used Meta AI for three purposes: conversational information retrieval at 67.9% (n = 19), content creation at 60.7% (n = 17), and text assistance at 39.3% (n = 11), beginning to answer RQ2 by showing that, for our respondents, embedded GenAI functioned as a low-friction insertion into ongoing communication tasks.

Conversational information retrieval unfolded within what is perceived as a private space (i.e. alongside WhatsApp and Instagram), making information-seeking immediate and personalized. Mohini (53) used Meta AI to understand a surgical procedure: “I asked, ‘What is that surgery?’ It gave me the meaning and explained why it is done. Then I asked ‘What are the side effects?’” She emphasized how complete and accessible the responses were, noting that the tool provided “whatever I required.” Her engagement was sequential and purposeful, illustrating that some participants also brought embedded GenAI into high-stakes contexts, though some of them described caution when consequences felt higher. In contrast, Naina (34) engaged tentatively, testing the tool by composing a coordination: “I just put one message saying that ‘let’s get ready for running to log out at shop 6:00 PM.’” She clarified that this was exploratory, not habitual, “I just checked how it works.” In these cases, engagement varied by immediacy and stakes, shaped by alignment with personal context. Familiarity with the platform allowed embedded GenAI to extend interactional continuity to high-stakes social contexts.

In content creation, Sangeeta (31) used the tool within WhatsApp for work: “I use image creator . . . for example, if I’m giving a promotion, I write the content, they give one paragraph.” When the output was too long, she revised her prompt, “I need this in two or three lines”, emphasizing the tool’s responsiveness: “It’s done within two to three seconds.” She appreciated the speed but also noted quality limitations in images. Her use is hybrid, folding generative outputs into broader task routines, not replacing them. Others engaged more experimentally. John (22) recalled prompting Meta AI to “generate a picture of Joe Biden being the president,” only to receive a blocked message: “Oops, I cannot generate the image.” He underscores encountering boundaries and adjusting expectations.

In text assistance, tone, correctness, and speed were crucial. Malik (50) described using the tool to revise a message for a loan request: “I typed some text, then I asked Meta AI, ‘this is my content, make it diplomatically.’” He copied the revised version and sent it to his bank manager. Though this was his only use, “Just one time. It was very useful”, revealing how infrequent engagement can support interactional continuity when precision is paramount. Rohit (28) used Meta AI to compose a job application: “It would take me 4–5 minutes to write the content. Meta AI gave it in 1–2 minutes.” He appreciated how it improved “punctuation marks and grammatical errors.” Rohit’s engagement focused on refinement, treating AI as a proofreading assistant rather than a creative partner.

These cases show that interactional continuity also structures embedded GenAI use through absorption (Malik), strategic use (Mohini, Rohit), and calibration (Naina, Sangeeta, John). By interpreting novel embedded GenAI tools in social media through the logic of task-based utility, users were preparing to use standalone GenAI and actively extending a utility-driven evaluative frame that traveled across tools, even as interface packaging shaped how those tools were recognized and approached.

Using standalone GenAI tools for task fulfillment

Standalone GenAI tools were used by 57.1% of respondents (n = 16) across all five purposes. Text assistance was most common, followed by content creation, conversational information retrieval, search assistance, and content recommendation. These categories overlapped, reflecting how the generative capacity of standalone systems blurred specific functional boundaries. While 7.1% (n = 2) respondents mentioned using Gemini, ChatGPT dominated as the default tool. The findings complement the answer of RQ2 by showing that standalone GenAI is evaluated primarily through task performance and episode-level fit within ongoing routines.

Conversational information retrieval often doubled as content recommendations, and content creation frequently merged with text assistance. These overlaps recalibrated perceived authorship: participants delegated first-pass drafting to the tool but retained editorial control, claiming authorship in the prompt, constraint, and cut. The tool functions as a capable drafter, with human voice and responsibility preserved through iterative editing. But not all experiments led to sustained use: Guru (32) used ChatGPT once to plan a trip: “I checked for Gokarna . . . What are the places that we can visit in Gokarna this day?. . . yeah, it was good, it was perfect”, but then stopped: “I have used it but not now . . . the purpose didn’t come out.” Amit (28) likewise kept the tool at arm’s length:

Mainly I use it for grammatical errors . . . because I think, you know, it can harm my personal creativity as a doctor . . . it’s easy to use it, but then I’ll be more dependent, I don’t want to be more dependent on the artificial.

Shivu (38) described moving away from standalone tools once embedded GenAI arrived:

Text it through my words and then rephrase it through AI . . . I have stopped using ChatGPT after WhatsApp AI . . . I don’t have to get into and then get out from there . . . in WhatsApp you can just copy and paste it, so that’s a convenience.

Others never moved beyond awareness. Rama (34) had seen the Meta AI blue circle icon and listened to an explanation of ChatGPT but dismissed both: “I’m not using, yeah . . . No, I don’t know,” later explaining that “time is also not there, for only I will use WhatsApp and Facebook and Instagram, that’s all.”

In contrast, Charan (29), a marketing professional, integrates text assistance and content creation using ChatGPT in his daily workflow. “I use it regularly for my marketing purpose . . . writing emails and anything in the organization.” His use ranged from drafting and revising copy to formatting tables and generating promotional lines like “Exciting offer. Visit our site today.” He described ChatGPT as a reliable assistant that “make[s] things easy . . . to be more productive [in] these 24 hours,” framing its value as efficiency. While he occasionally used Gemini, ChatGPT remained his default. Charan’s case underscores how individual familiarity and task alignment drive standalone GenAI tool absorption when utility is high.

Straddling both content recommendation and conversational information retrieval, some participants also used standalone GenAI for emotional reassurance, with engagement remaining task-oriented. Mohini (53) shows how affective needs, such as stress relief or companionship, led her to use ChatGPT. “Sometimes I will be depressed . . . I say I want to listen to some music. They [i.e. ChatGPT] will say this music is good for your refreshment.” She would then follow the recommendation. She also engaged with it as a social surrogate: “Sometimes we want to talk to someone, no? When we are sitting alone . . . that time we can chat.” These affective interactions were goal-driven, where presence and mood regulation reinforced continuity. Other users, like Aman (24), began exploring uses in search assistance and conversational information retrieval before shifting toward goal-oriented tasks in content creation and text assistance. “I first asked it to make a joke . . .” Then, his use turned toward productivity, including writing taglines, composing social media captions, and correcting grammar: “I write something, and I want to correct my grammar . . . it’s useful.” He assessed the value through its support for marketing tasks: “It gives good ideas for content . . . it’s good”. Lata (26), an architect, exemplified content creation. “It’ll just give me your formatted presentation . . . what slide needs to have what and all things. And then I can improvise from there.” She used it as a structured starting point, saving time while maintaining control over content.

These cases show that interactional continuity structures standalone GenAI use in three modes: absorption (Charan); strategic (Mohini and Aman); and calibration (Lata). Compared with embedded AI and embedded GenAI, functional boundaries blur, yet engagement remains task-first and performance-based.

Social influence in standalone GenAI adoption

Adoption was rarely solitary: invites, demos, and workplace/school cues did the onboarding and taught “use scripts.” Despite different network dependencies, social media’s value rose with peers present, while standalone GenAI offers immediate utility, peers still reduced uncertainty and supplied task templates. Most respondents encountered standalone GenAI through coworkers and friends who validated its usefulness. Amit (28), a lawyer, illustrates this: “In my company, we use ChatGPT. So, I started using it regularly, mainly for work e-mails, presentations, summaries.” He also promoted adoption among colleagues: “I sent them videos that this is ChatGPT and how to use it.” Guru (32), in finance, adopted AI via peers but used it more occasionally to refine professional communication: “I used it when I wanted to draft a proper email . . . it gives me a formal and better version than what I can do.” He likewise acted as an introducer: “When my friends don’t know what AI is, I show them this.”

Recognition: Seamless systems, named or not

Across embedded and standalone GenAI, participants adopted what worked and only sometimes labeled it as AI. Embedded AI tools were heavily used, yet were almost never labeled as AI, explicit naming occurred in 9% (n = 2) of respondents. By contrast, when AI appeared as a distinct tool, either as embedded or a standalone GenAI tool, users generally named it, while judging it by task fit (see Table 1). These contrasts answer RQ3 by showing that people use embedded and standalone AI for task-oriented purposes and evaluate them based on performance, while explicit naming as “AI” mainly follows interface packaging rather than use.

Table 1.

Recognition patterns by use status.

Modality	Status	Explicit	Implicit	Misrecognition
Embedded	Users (n = 23)	2	20	1
	Non-users (n = 5)	1	3	1
Embedded GenAI	Users (n = 23)	23	0	0
	Non-users (n = 5)	0	0	5
Standalone GenAI	Users (n = 16)	16	0	0
	Non-users (n = 12)	6	0	6

The AI recognition spectrum

Participants did not share a stable definition of AI. Instead, they identified it through recognizable tools (e.g. ChatGPT, Meta AI), interface cues (e.g. a chat window), or perceived capabilities (“it rewrites,” “it suggests,” “answers like a person”), sometimes framed as brands/assistants and sometimes as outcomes or platform metaphors (“the app knows,” “the algorithm decides”). These vernacular understandings anchor our analysis of recognition and naming below.

AI recognition ranged across three positions: explicit naming, implicit recognition, and misrecognition. In explicit naming, respondents directly identified tools or interfaces as AI, yet still evaluated them in terms of usefulness. Amit (26), for instance, described social platforms themselves as “AI apps”: “they study us, they know us and they give what we [want].” Others pointed to the WhatsApp or Instagram icon as “AI” but framed it as another feature to be tried, or ignored, depending on task relevance. In implicit recognition, respondents described algorithmic behavior without applying the AI label. Swetha (24) initially referred only to “my algorithm” on Instagram, explaining how her feed shifted toward religious and political content. She discussed how “it” learned from her activity and changed what she saw, signaling an understanding of automated curation without naming it as AI. In misrecognition, people mislocated where it was present. Swetha contrasted ChatGPT and Gemini with WhatsApp’s embedded assistant and corrected herself mid-answer: “maybe now I’m realizing that, oh, WhatsApp has its own AI,” recognizing that a familiar feature she already used was an AI system. Mohan (25) immediately labeled the icon “artificial” but added that he had “no interest, no ideas” to try it and insisted he was “not using AI,” even as he scrolled through AI-driven reels each night. These patterns suggest that recognition is modality-conditioned: embedded, feature-level AI is more often implicit or misrecognized, while assistant-style GenAI, whether platform-integrated or standalone, is more often explicitly named.

The intersection of usage and recognition suggests that the ability to name a tool as “AI” is not a prerequisite for its adoption. Task fit and context drive engagement, while interface design can shape whether AI is treated as an unnamed background system or as a named actor. Interactional continuity helps explain how task-first adoption can persist under low recognition when AI is embedded as background mediation.

Discussion and conclusion

AI use was described as recurring through repeatable insertion points within task episodes, while interface packaging shaped whether the same incorporation was recognized and named as AI. Embedded AI carried much of the platform-native work (especially recommendation and search) and was engaged as part of scrolling and searching rather than as a distinct system. On the other hand, GenAI concentrated on message- and document-facing tasks (e.g. conversational retrieval, text assistance, content creation) but was still folded into familiar routines. Features inside platform interfaces were rarely named as AI, whereas assistant-style tools, embedded or standalone, were more often labeled. This helps explain why the same environment can read as disruption or invisibility: packaging makes AI register as an eventful tool or recede into ambient mediation. Naming differences were not misunderstandings; participants used functional cues and familiar use cases rather than technical definitions, so naming could diverge from frequency of use.

Revisiting interactional continuity

In participants’ accounts, this mechanism is strongest for reversible, quickly evaluable tasks (e.g. drafting or rephrasing text, first-pass outlines) and weakens when verification, accountability, or higher stakes require corroboration, leading users to limit or avoid AI. First, task fit drove reuse across modalities: respondents kept tools that advanced the next step and abandoned outputs that failed task expectations. Consequently, the continued use of AI depends on episode-level evaluability, particularly for tasks where the output can be quickly verified. Second, AI was typically inserted into specific steps of existing workflows rather than reorganizing them wholesale. Participants used it for first-pass drafting and refinement, then reclaimed control through editing and selection; embedded features entered earlier (scrolling/searching), but still in task-advancing ways.

Third, interface packaging shaped recognition. Naming clustered around chat-based AI interfaces, even when used less consistently than embedded features, which were often described as “my algorithm” or “the app.” Packaging governs when AI is foregrounded as an actor, while task fit governs whether it is kept in the routine. This specifies how assistant-style packaging can re-materialize AI as a named entity within backgrounded mediation. Interactional continuity links “turn/disruption” and “infrastructure/invisibility” scholarship through the same episode-level mechanism.

Limitations and future research

Our analysis is constrained by a small, single-site urban office sample and interview-based self-reports, so findings are not generalizable and capture sequence-level accounts rather than in-situ decision processes. Future work should pair interviews with observation, diaries, or trace-based methods (e.g. chat logs and screen recordings) to examine sequencing under naturalistic conditions. Researcher positionality may shape interpretation, and comparative work across regions, sectors, and labor conditions would help test scope and boundary conditions.

Conclusion

For scholars and policymakers, our findings suggest the utility of shifting from frames of trying to distinguish whether people “use AI” to where tools enter task episodes and what they deliver. Labels are weak evidence of engagement. Rather, task placement and outcomes are stronger. Measurement and governance should treat embedded and chat-based GenAI systems like ChatGPT, Meta AI, and Gemini as a continuum and evaluate them in the contexts where decisions are made. For media studies and STS scholars, our development of interactional continuity links “turn” and infrastructure accounts by reframing disruption and invisibility as encounter outcomes shaped by packaging and sequence placement. Ultimately, our Indian workplace case study helps reveal how: low friction insertion into time-pressured, platform-first routines strengthens interactional continuity, while higher-stakes contexts requiring verification constrain it. This clarifies adoption and recognition as outcomes of insertion points, immediate evaluability, and interface packaging.

Supplemental Material

sj-docx-1-nms-10.1177_14614448261448545 – Supplemental material for Beyond disruption and invisibility: Interactional continuity in everyday AI use in India

Supplemental material, sj-docx-1-nms-10.1177_14614448261448545 for Beyond disruption and invisibility: Interactional continuity in everyday AI use in India by Emilia Edwards and Dhiraj Murthy in New Media & Society

Footnotes

Acknowledgements

The authors acknowledge the respondents for their time and valuable contributions to this study.

ORCID iD

Emilia Edwards

Ethical considerations

This study was reviewed and approved as exempt by the Institutional Review Board (IRB) at The University of Texas at Austin.

Consent to participate

Written informed consent was obtained from all participants prior to the interviews.

Consent for publication

Not applicable.

Author contributions

E.E. led the study design, data analysis, and writing under the supervision of D.M. D.M. conducted the interviews, transcribed, and anonymized the data, and provided conceptual guidance, methodological feedback, and critical revisions to the manuscript.

Funding

The authors disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This work was supported by Good Systems, a research grand challenge and the Moody College of Communication at the University of Texas at Austin.

Declaration of conflicting interests

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

Data availability statement

The data that support the findings of this study are available from the corresponding author upon reasonable request.

Supplemental material

Supplemental material for this article is available online.

Notes

Author biographies

Emilia Edwards is a PhD candidate in the School of Journalism and Media at The University of Texas at Austin. Her research examines media and photographic practices in relation to emerging technologies, with a focus on artificial intelligence, visual communication, and everyday meaning-making.

Dhiraj Murthy is a Professor of Media Studies, Sociology, and Information at the University of Texas at Austin and Director of the Computational Media Lab, where he investigates the evolving intersection of social media and AI technologies. His research leverages machine learning and computational social science to analyze how artificial intelligence shapes digital discourse and behavior on global platforms. His work provides critical, sociotechnical insights into the impact of AI, including within the Global South.

References

Akrich

(1992) The de-scription of technical objects. In: Law

(ed.) Shaping Technology/Building Society: Studies in Sociotechnical Change. MIT Press, pp.205–224.

Amoore

(2020) Cloud Ethics: Algorithms and the Attributes of Ourselves and Others. Duke University Press.

Ananny

Crawford

(2018) Seeing without knowing: Limitations of the transparency ideal and its application to algorithmic accountability. New Media & Society 20(3): 973–989.

Ardévol

Gómez-Cruz

(2014) Digital ethnography and media practices. In: Valdivia

(ed.) The International Encyclopedia of Media Studies. vol. 7, John Wiley & Sons, pp.498–518.

Bajpai

Wadhwa

(2021) Artificial intelligence and healthcare in India. ICT India working paper no. 43. Columbia University.

Behera

Gartia

(2024) The era of digital transformation: Navigating the social media landscape in India. In: Padhi

Datta

(eds) Social Media in India: Dynamism, Issues and Challenges. N. B. Publications, pp.1–16.

Berker

Hartmann

Punie

, et al. (2005) Domestication of Media and Technology. McGraw-Hill Education.

Borup

Brown

Konrad

, et al. (2006) The sociology of expectations in science and technology. Technology Analysis & Strategic Management 18(3–4): 285–298.

Buana

Prasetyo

Muhsinin

, et al. (2024) User satisfaction evaluation of Meta AI integration in WhatsApp: A quantitative EUCS approach. Journal of Information Technology, Computer Engineering and Artificial Intelligence (ITCEA) 1(2): 77–84.

10.

Caspi

Bem

Elder

Jr (1989) Continuities and consequences of interactional styles across the life course. Journal of Personality 57(2): 375–406.

11.

De Certeau

Rendall

(1984) The Practice of Everyday Life. University of California Press.

12.

Chougule

Kurane

Gaikwad

, et al. (2023) Comparative study of social media platform users using statistical tools and techniques. International Journal of Progressive Research in Engineering Management and Science 3(6): 174–178.

13.

Christin

(2020) The ethnographer and the algorithm: Beyond the black box. Theory and Society 49(5–6): 897–918. https://doi.org/10.1007/s11186-020-09411-3

14.

Cinque

(2024) Emerging Digital Media Ecologies: The Concept of Medialogy. Routledge.

15.

Cohen

Khavkin

Movsowitz Davidow

, et al. (2024) ChatGPT in the public eye: Ethical principles and generative concerns in social media discussions. New Media & Society. Epub ahead of print 21 September. https://doi.org/10.1177/14614448241279034

16.

Couldry

Hepp

(2018) The Mediated Construction of Reality. John Wiley & Sons.

17.

De Seta

Pohjonen

Knuutila

(2024) Synthetic ethnography: Field devices for the qualitative study of generative models. Big Data & Society. https://doi.org/10.1177/20539517241303126

18.

Dodds

Zamith

Lewis

(2025) The AI turn in journalism: Disruption, adaptation, and democratic futures. Journalism. Epub ahead of print 15 May. https://doi.org/10.1177/14648849251343518

19.

Fahad

Mustafa

Luqman

(2023) Religious-political discussion on Instagram and WhatsApp and perception of religion among youths in Delhi. Journal of Asian and African Studies. Epub ahead of print 17 October. https://doi.org/10.1177/00219096231200592

20.

Ferrari

van Dijck

van den Bosch

(2023) Observe, inspect, modify: Three conditions for generative AI governance. New Media & Society 27(5): 2788–2806.

21.

Gillespie

(2014) The relevance of algorithms. Media technologies: Essays on Communication, Materiality, and Society 167(2014): 167.

22.

Grandinetti

(2023) Examining embedded apparatuses of AI in Facebook and TikTok. AI & Society 38: 1273–1286.

23.

Gray

Suri

(2019) Ghost Work: How to Stop Silicon Valley from Building a New Global Underclass. Harper Business.

24.

Helberger

(2021) On the democratic role of news recommenders. In: Thurman

Lewis

Kunert

(eds) Algorithms, Automation, and News. Routledge, pp.14–33.

25.

Hinds

von Krogh

(2024) Generative AI, emerging technology, and organizing: Towards a theory of progressive encapsulation. Organization Theory 5(4): 26317877241293478.

26.

Hirsch

Silverstone

(2003) Information and communication technologies and the moral economy of the household. In: Hirsch

Silverstone

(eds) Consuming Technologies. Routledge, pp.25–40.

27.

Holmström

Carroll

(2024) How Organizations Can Innovate with Generative AI. Business Horizons.

28.

Hosseini

Horbach

Holmes

, et al. (2025) Open science at the generative AI turn: An exploratory analysis of challenges and opportunities. Quantitative Science Studies 6: 22–45.

29.

Information Commissioner’s Office (n.d.) Explaining decisions made with AI: Part 1 – the basics of explaining AI. Available at: https://ico.org.uk/for-organisations/uk-gdpr-guidance-and-resources/artificial-intelligence/explaining-decisions-made-with-artificial-intelligence/

30.

Jasanoff

Kim

(eds) (2019) Dreamscapes of Modernity: Sociotechnical Imaginaries and the Fabrication of Power. University of Chicago Press.

31.

Jensen

Murthy

Baker

(2025) Automating remix: generative AI, creative labor, and the decay of aura. Information, Communication & Society 1–17.

32.

Jones

Luger

(2022) AI “everywhere and nowhere”: Addressing the AI intelligibility problem in public service journalism. Digital Journalism 10(10): 1731–1755.

33.

Karan

Angadi

(2025) Artificial intelligence integration into school education: A review of Indian and foreign perspectives. Millennial Asia 16(1): 173–199.

34.

Khalid

Chakraborty

Rani

(2025) An exploration of ethical implications and social impact of artificial intelligence (AI) in the Indian context. In: Garcia

Pandey

Prasad

, et al. (eds) Open AI and Computational Intelligence for Society 5.0. IGI Global Scientific Publishing, pp.265–290.

35.

Laba

(2024) Beyond magic: Prompting for style as affordance actualization in visual generative media. New Media & Society. Epub ahead of print 29 October. https://doi.org/10.1177/14614448241286144

36.

Latour

(2005) Reassembling the Social: An Introduction to Actor-Network-Theory. Oxford University Press.

37.

Leonardi

(2011) When flexible routines meet flexible technologies: Affordance, constraint, and the imbrication of human and material agencies. MIS Quarterly 35(1): 147–167.

38.

Lewis

Guzman

Schmidt

, et al. (2025) Generative AI and its disruptive challenge to journalism: An institutional analysis. Communication and Change 1(1): 9.

39.

Zhang

(2024) Finding love in algorithms: Emotional contexts of close encounters with AI chatbots. Journal of Computer-Mediated Communication 29(5): zmae015.

40.

Meta (2024) Meta AI arrives in India: Leading AI assistant now at your fingertips. Meta. Available at: https://about.fb.com/news/2024/06/meta-ai-arrives-in-india-leading-ai-assistant-now-at-your-fingertip/

41.

Molla

MAM

Ahsan

(2025) Artificial intelligence and journalism: A systematic bibliometric and thematic analysis of global research. Computers in Human Behavior Reports 20: 100830. https://doi.org/10.1016/j.chbr.2025.100830

42.

Mühlhoff

(2023) Predictive privacy: Collective data protection in the context of artificial intelligence and big data. Big Data & Society. Epub ahead of print 16 April. https://doi.org/10.1177/20539517231166886

43.

Mukhopadhyay

(2023) ChatGPT adoption in cultural domains in India: A diffusion of innovation analysis. Asian Journal of Management and Commerce 4(2): 112–117.

44.

Murthy

(2022) Coding Twitter data using qualitative and computational methods: A mixed methods framework. In: Sloan

Quan-Haase

(eds) The SAGE Handbook of Social Media Research Methods. Sage, pp.533–545.

45.

Murthy

(2024) Social Media Cultures. Lexington Books.

46.

News (2025) OpenAI Launches GPT-5, A Potential Barometer for Whether AI Hype Is Justified. Associated Press.

47.

Noble

(2018) Algorithms of Oppression: How Search Engines Reinforce Racism. New York University Press.

48.

Okolo

Dell

Vashistha

(2022)Making AI explainable in the Global South: A systematic review. In: COMPASS ‘22: Proceedings of the 5th ACM SIGCAS/SIGCHI conference on computing and sustainable societies, Seattle, WA, 29 June–1 July, pp.1–14. Association for Computing Machinery.

49.

Orlikowski

(2000) Using technology and constituting structures: A practice lens for studying technology in organizations. Organization Science 11(4): 404–428.

50.

Park

Kim

, et al. (2024) From anthropomorphism to theomorphism: How users perceive and relate to generative AI. In: Proceedings of the 2024 CHI conference on human factors in computing systems (CHI‘24), Honolulu, HI, 11–16 May, pp.31–18. Association for Computing Machinery.

51.

Petre

(2021) All the News That’s Fit to Click: How Metrics Are Transforming the Work of Journalists. Princeton University Press.

52.

Plantin

Lagoze

Edwards

, et al. (2018) Infrastructure studies meet platform studies in the age of Google and Facebook. New Media & Society 20(1): 293–310.

53.

Rader

Gray

(2015) Understanding user beliefs about algorithmic curation in the Facebook news feed. In: Proceedings of the 33rd annual ACM conference on human factors in computing systems, Seoul, Republic of Korea, 18–23 April, pp.173–182, Association for Computing Machinery.

54.

Radlinski

Craswell

(2017) A theoretical framework for conversational search. In: Proceedings of the 2017 conference on conference human information interaction and retrieval, Oslo, 7–11 March, pp.117–126. Association for Computing Machinery.

55.

Rehman

Khalil

(2024) Use of meta AI in WhatsApp in learning and teaching English grammar: An experimental study. Liberal Journal of Language & Literature Review 2(4): 139–156. Available at: https://llrjournal.com/index.php/11

56.

Rogers

Singhal

Quinlan

(2014) Diffusion of innovations. In: Stacks

Salwen

Eichhorn

(eds) An Integrated Approach to Communication Theory and Research. Routledge, pp.432–448.

57.

Schaetz

Schjøtt

(2026) AI hype and its function: An ethnographic study of the local news AI initiative of the associated press. Digital Journalism 14(2): 220–237. https://doi.org/10.1080/21670811.2024.2443163

58.

Schatzki

(2002) The Site of the Social: A Philosophical Account of the Constitution of Social Life and Change. Penn State Press.

59.

Seaver

(2017) Algorithms as culture: Some tactics for the ethnography of algorithmic systems. Big Data & Society 4(2): 2053951717738104.

60.

Siles

(2023) Vivir con algoritmos: Una aproximación relacional a la vida cotidiana. Editorial Universidad de Costa Rica.

61.

Silverstone

Haddon

(1996) Design and the domestication of information and communication technologies: Technical change and everyday life. In: Silverstone

Mansell

(eds) Communication by Design. The Politics of Information and Communication Technologies. Oxford University Press, pp.44–74.

62.

Simon

(2025) Rationalisation of the news: How AI reshapes and retools the gatekeeping processes of news organisations in the United Kingdom, United States and Germany. New Media & Society. Epub ahead of print 8 May. https://doi.org/10.1177/14614448251336423

63.

Sloane

(2024) Controversies, contradiction, and “participation” in AI. Big Data & Society. Epub ahead of print 25 February. https://doi.org/10.1177/20539517241235862

64.

Star

Ruhleder

(1996) The ecology of infrastructure: Problems in the implementation of large-scale information systems. Information Systems Research 7(1): 111–134.

65.

Chan

(2025) Assembling platform governance as private ordering in the age of generative AI: Platform interdependence in policy evolution. Information, Communication & Society. Epub ahead of print 3 June. https://doi.org/10.1080/1369118X.2025.2513672

66.

Suchman

(2020) Algorithmic warfare and the reinvention of accuracy. Critical Studies on Security 8(2): 175–187.

67.

Törnberg

Söderström

Barella

, et al. (2025) Artificial intelligence and the state: Seeing like an artificial neural network. Big Data & Society. Epub ahead of print 16 May. https://doi.org/10.1177/20539517251338773

68.

Usher

(2025) Generative AI and journalism: Hype, the always already new, and directions for scholarly imagination. Digital Journalism. Epub ahead of print 9 April. https://doi.org/10.1080/21670811.2025.2490604

69.

Van Dijck

Poell

De Waal

(2018) The Platform Society: Public Values in a Connective World. Oxford University Press.

70.

Wang

Downey

(2025) Mapping the sociotechnical imaginaries of generative AI in UK, US, Chinese and Indian newspapers. Public Understanding of Science 34(7): 930–948.

71.

Williamson

van Rooyen

Dry

(2025) In tandem with artificial intelligence: A working framework for coding in ATLAS. ti™. International Journal of Qualitative Methods 24: 16094069241306551.

72.

(2024) AI and Economic Governance: Navigating the Visible and Invisible Hands in a Digital Era (No. tvchw_v1). Center for Open Science.

73.

Zajko

(2022) Artificial intelligence, algorithms, and social inequality: Sociological contributions to contemporary debates. Sociology Compass 16(3): e12962.

74.

Zamani

Trippas

Dalton

, et al. (2023) Conversational information seeking. Foundations and Trends® in Information Retrieval 17(3–4): 244–456.

75.

Zhang

Huang

Zhu

JJH

(2025) How journalism researchers navigate the AI hype: Research orientations and intervention recommendations. Information, Communication & Society. Epub ahead of print 10 April. https://doi.org/10.1080/1369118X.2025.2482666

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.02 MB

0.00 MB