Abstract
This article examines the everyday work of pay-per-click (PPC) specialists in the context of the growing automation and algorithmization of digital advertising. Drawing on 27 in-depth interviews, ethnographic observation and an analysis of over one thousand posts from professional online forums, it explores how practitioners negotiate responsibility and expertise under conditions of technological opacity. The study reveals that PPC work is undergoing a transformation from a technical to an interpretive profession. Specialists increasingly act as translators of algorithmic decisions to clients and managers, maintaining the appearance of control over systems whose internal logic remains hidden. Their expertise takes the form of embodied and collective knowledge, grounded in intuition, emotion and peer exchange in online environments. The article introduces the notion of algorithmic responsibility as an epistemic practice, a daily effort to make sense of automated decisions under structural ignorance. PPC work is read here as a theoretically generative site for examining the data economy, a space where knowledge, power and technology intertwine into new forms of relational expertise. This paper contributes to critical algorithm studies by demonstrating that humans do not disappear from automated processes but rather become their indispensable interpreters – the guarantors of meaning in a society dominated by data.
Keywords
Research context: Pay-per-click as a site of algorithmic opacity
Pay-per-click (PPC) advertising refers to a model of digital advertising in which advertisers bid in real time for the placement of ads, paying only when users click on an advertisement or complete a predefined action. PPC specialists are responsible for configuring, monitoring, and explaining the performance of these campaigns across major advertising platforms such as Google Ads or Meta Ads. Their work involves translating business objectives into measurable conversion events, managing campaign structures (including keywords, audiences, product feeds and creatives), and continuously interpreting performance data under conditions of volatility and partial observability (Turow, 2011; Crain, 2019).
In the early phase of online marketing, the work of a PPC specialist operating in systems such as Google Ads or Meta Ads was largely manual and controllable. It relied on keyword selection, manual bidding, geographic segmentation and the analysis of click-through rates (CTR) or conversions through simple analytical dashboards. It was a stage in which humans, not systems, decided how campaigns should look and how to optimize their performance. Knowledge was empirical, largely grounded in iterative testing and local heuristics. This configuration echoed longer-standing attempts within advertising to discipline uncertainty through metrics, optimization and managerial oversight, while still relying heavily on human judgment and local heuristics (Porter, 1995; McGuigan, 2023; Crain, 2019).
With the spread of algorithmic optimization systems (e.g., Smart Bidding, Performance Max, automatic extensions), the structure of work began to change fundamentally. Manual interfaces disappeared, replaced by predictive models that autonomously decide on audience targeting, timing and bidding strategies. The PPC specialist no longer builds campaigns from scratch but rather partners with an algorithmic system whose recommendations are presented as ‘best practices’, often without the possibility of modification (Elish, 2019). While platform-driven automation is often framed as a radical break with earlier practices, it extends longer histories of optimization and delegation in advertising. Mathematical models and calculative devices have long been used to rationalize subjective judgment and scale decision-making under conditions of complexity and uncertainty (Porter, 1995; McGuigan, 2023). What distinguishes contemporary PPC is not the presence of models as such, but the concentration of control within platform-owned infrastructures and the opacity of their continuous recalibration (Caliskan et al., 2025). Unlike earlier forms of advertising automation, contemporary PPC systems operate as continuously updating infrastructures whose parameters cannot be inspected, paused or selectively overridden by practitioners, intensifying both epistemic dependence and responsibility.
Consequently, the expert's role increasingly combines limited technical intervention with extensive interpretive and justificatory work, particularly in client-facing contexts. This transformation introduces new tensions: while specialists are formally responsible for campaign outcomes, they no longer have full control over how results are produced. As one user on forum noted: Since Performance Max and Smart Bidding appeared, my job feels like explaining the algorithm's decisions. The system decides for me, and I just have to justify it to the client. Sometimes it's pure magic – I don’t understand why something works or why it stops working overnight. Google changes things in the background, we adapt, and then the client asks: ‘why did sales drop?’ What can I say? ‘Because the algorithm wanted it’? 
. (Post forum, January 2025)
One central lever concerns the definition of success itself. PPC specialists are responsible for specifying conversion events, value hierarchies and attribution logics that determine what the system optimizes for. Decisions such as whether to prioritize purchases, leads, micro-conversions or revenue-weighted actions profoundly affect campaign trajectories, even though their downstream effects remain partially opaque. In this sense, practitioners do not directly steer outcomes, but they configure the evaluative framework through which outcomes are recognized and rewarded by the algorithm (Cluley, 2018; Espeland and Stevens, 2008).
A second set of levers relates to data provisioning and constraint-setting. Budget caps, target ROAS or CPA thresholds, audience signals, exclusion rules and campaign structure delimit the space within which optimization occurs. While platforms often frame these inputs as minor or optional, practitioners treat them as strategic points of intervention, carefully adjusting parameters to stabilize performance or to prevent undesirable forms of algorithmic experimentation. This work resembles what McGowan et al. (2024) describe as intermediary labour: managing tensions between platform logics and client expectations without direct access to decision-making mechanisms.
Creative assets and product feeds constitute another crucial site of intervention. Automated systems increasingly test, combine and prioritize creatives autonomously, yet the quality, diversity and framing of supplied materials shape how campaigns are evaluated and scaled. PPC specialists thus engage in anticipatory work, crafting assets not only for human audiences but also for algorithmic interpretation, attempting to infer how systems ‘read’ relevance, intent and quality signals (Beauvisage et al., 2024; Beer, 2017).
These levers are characterized by delayed feedback and ambiguous causality. Interventions may take days or weeks to register, may interact unpredictably with platform updates or may be overridden by system-level changes beyond practitioners’ visibility. As a result, agency in automated PPC is inherently probabilistic. Specialists act through hypothesis-testing, pattern recognition and cautious adjustment, rather than through linear cause-and-effect reasoning. This condition aligns with broader accounts of algorithmic work in which professionals operate as boundary workers, navigating between managerial responsibility and infrastructural opacity (Ryan et al., 2023).
Importantly, the persistence of these levers sustains professional accountability. Because PPC specialists retain the capacity to intervene, they remain organizationally and morally responsible for outcomes in the eyes of clients and organizations. Automation therefore intensifies rather than alleviates responsibility: practitioners must decide when to intervene, when to refrain and how to justify both action and inaction under conditions of uncertainty. This dynamic sets the stage for the emergence of collective sense-making practices and tacit forms of expertise, which do not replace agency but enable its continued exercise within constrained algorithmic environments.
The epistemology of PPC work increasingly relies on tacit and intuitive knowledge (Polanyi, 1966) rather than on clear rules and procedures. This knowledge is situational, embodied and collective – acquired not through formal instruction but through participation in a community of practice (Knorr-Cetina, 1999; Lave and Wenger, 1991).
PPC specialists today operate within a gap between responsibility and control. On the one hand, they are the public face of campaigns and accountable for outcomes; on the other hand, more and more decisions are made by self-learning systems whose internal logic remains undisclosed. As Frank Pasquale (2015) famously argued, algorithmic systems function as ‘black boxes’ – their decision-making processes cannot be externally verified, leaving users to guess what occurs inside.
This raises a fundamental question: who is to blame for a failed campaign – the specialist who followed the system's recommendations, or the algorithm operating under its hidden logic? Ethnographic and netnographic data reveal that practitioners have developed tactical ways of coping with this epistemic uncertainty, including:
shifting blame onto the system (‘the algorithm went crazy’), constructing defensive narratives (‘it's due to Google's internal tests beyond our control’), and reinterpreting results as long-term strategy adjustments (‘a drop in ROAS is part of a broader optimization plan’).
As one forum user expressed: The worst thing is that the client only sees the numbers. They look at a report and ask why CTR dropped by two points – and I have no idea which ads are even showing, because everything runs through automation. I’m supposed to explain something I don’t understand myself. I feel like Google's PR guy, selling the story of ‘the system learning’. (Post forum, January 17, 2025)
Yet, despite this asymmetry, PPC communities have developed epistemic resilience, practical and often informal methods to cope with opacity, such as A/B testing, pattern analysis, cross-data triangulation (e.g., GA4, CRM, heatmaps) and peer consultation. The foundation of these practices is embodied knowledge, an ability to ‘feel’ that something is wrong before it appears in the data, echoing Polanyi's dictum that ‘we know more than we can tell’.
As one in-house respondent stated: When Performance Max was introduced, the lack of trust wasn’t toward me but toward Google and how these campaigns would work. It's a kind of black box – you feed it assets, products, set target ROAS, and it all happens automatically. You don’t always know why, but you know what to do to make it work. (Interview, In-house 1, January 2025)
As Zuboff (2020) argues, the environment of digital advertising exemplifies surveillance capitalism – a regime where data extraction, monetization and control over attention are delegated to users themselves. In this setting, embodied and communal expertise becomes a form of epistemic resistance against opaque and asymmetrical systems of algorithmic governance.
In recent years, the work environment of PPC specialists has become one of the most dynamic sites of algorithmic experimentation, where tensions between human and automated knowledge, agency and control, and transparency and opacity (black-boxing) are particularly visible. Shaped at the intersection of Big Data technologies, machine learning, and platform economies, the PPC sector can be seen as a laboratory for algorithmic opacity, a field in which professional practice must constantly negotiate the boundaries of responsibility and knowledge (Pasquale, 2015; Seaver, 2017). Importantly, this opacity does not eliminate professional responsibility. Instead, it reorganizes it around practices of interpretation, narrative stabilization and the management of expectations across institutional boundaries, particularly in client-facing contexts (Espeland and Stevens, 2008; Bolin and Schwarz, 2023).
Theoretical framework: Algorithmic power and the epistemology of practice
To analyse PPC as an automated field of professional activity, it is necessary to adopt an approach that does not reduce the problem to technological functionality but considers the cultural, social and epistemic dimensions of work. The following theoretical perspectives provide complementary frameworks that help to capture the complexity of relations between algorithms, knowledge and responsibility in PPC practice.
PPC specialists operate within three main organizational forms: marketing agencies, in-house departments and freelancers. These forms differ not only in structure but in epistemic style – in how they relate to data, uncertainty and interpretation. The variation lies not in technical proficiency but in the depth and mode of knowing. The Social Worlds/Arenas Theory proposed by Anselm Strauss (1978) and later developed by Adele Clarke (2015) allows us to understand these differences as outcomes of collective negotiation and interaction. Social worlds are ‘domains of social life organized around shared activities, norms, and practices that are constantly under negotiation’. They are not only organizational spaces but also sites where meanings and professional values are produced. In the PPC field, what counts as effective, ethical or professional is co-created within such communities, through both internal standards and informal exchanges in digital forums.
This ongoing negotiation reflects the fact that PPC knowledge is rarely formalized. Specialists learn primarily through practice, observation and the exchange of experiences. As a result, their ethical and epistemic frameworks are continuously shaped through work itself. Respondents often emphasized honesty and transparency towards clients while simultaneously acknowledging their dependence on systems they could not fully explain. Statements such as ‘the system just worked that way’ or ‘something changed in the algorithm’ reveal that understanding the system and negotiating one's role within it are embedded in the human–machine relationship rather than external to it (Suchman, 2007).
A central analytical category for understanding automated professional environments is the notion of the black box (Latour, 1987; Pasquale, 2015). Black boxes are systems whose data flows and decision-making mechanisms remain hidden from users, even though their effects have direct social and economic consequences. PPC specialists, while formally responsible for campaigns, rarely have insight into how or why an algorithm made a given decision. The degree of this opacity varies among practitioners, depending on their access to data, tools and interpretive communities. A practical manifestation of the black box is evident in the questions and troubleshooting strategies shared on professional social media biggest forums in Poland. It's hard to optimise Performance Max campaigns because you never really know how users are targeted. If someone misconfigures remarketing lists and uses them as signals, there's no alert, no visibility. (Interview, Agency 5, March 2025)
Following this, classical sociology of knowledge (Berger and Luckmann, 1966) reminds us that all knowledge, including scientific and technical, is socially constructed. Data are not ‘raw facts’ but products of decisions about what, how, and when to measure, as well as how to interpret and report outcomes. The same logic applies to PPC analytics: what appears as objective data is often a negotiated representation, shaped by the technical affordances of platforms and the interpretive labour of specialists.
As automation increases, so too does the tension between the expectations placed on specialists and their actual agency. They are responsible for outcomes they cannot fully influence. Pasquale (2015) describes this as the ‘transfer of responsibility onto operators of systems that cannot be audited’. In the empirical material, this tension is expressed through defensive discourses that justify or explain algorithmic outcomes ex post. Specialists perform what can be called epistemic labour: they construct plausible narratives to account for automated decisions. Responsibility thus becomes a performative ritual rather than a function of control. A client asks why the campaign stopped. Everything looks fine in the dashboard, but results just dropped. I have to say something – maybe seasonality, maybe the system is learning. Deep down, I know something simply changed in the algorithm. (Interview, In-house 6, January 2025) The hardest part is talking to clients because when something breaks, there's no one to blame. I just repeat that ‘the algorithm is adapting.’ It's become automatic – explaining without explaining. (Post forum, May 2025)
In this sense, PPC can be understood as a micro-infrastructure of knowledge in the data society, a site where technology, economy and interpretation converge into a single epistemic process. Every act, from setting conversion parameters to defining report scope, is both cognitive and rhetorical: it produces not only numbers but credibility. I trust the reports I make myself. They include everything I care about, and I don’t have to dig through GA4. The best summary is that I trust my own reports the most. (Interview, Freelancer 5, February 2025)
Bruno Latour (1987) originally observed that black boxes cease to be questioned when they function effectively. In PPC, the logic is reversed: the more efficient the system, the stronger the sense of lost control. As long as the system ‘works’, its inner mechanisms remain invisible; effectiveness replaces understanding. For Pasquale (2015), this opacity is a political and economic condition; for Seaver (2017), a cultural form; for Latour, a stabilizing device; and for Elish (2019), a moral dilemma. Together they show that the black box in PPC is not only a technical artefact but an ethical problem, a system that limits auditability while demanding accountability from human operators.
Within this synthesis, the black box in PPC practice can be defined as a system that restricts transparency yet maintains expectations of moral and professional responsibility for its human users. Specialists become responsible for outcomes they cannot audit, explaining and legitimizing the actions of opaque systems through narrative and performance.
Methodology and empirical material
In studying automated work environments such as PPC, it is essential to adopt a methodological approach that integrates both the material dimension of technology and the social production of knowledge under conditions of opacity. The aim is not only to describe what specialists do but to understand their epistemic strategies, defensive narratives, communal heuristics and forms of technological self-awareness that emerge in relation to nontransparent yet decisive systems.
The research sought to understand how PPC specialists construct and maintain a sense of responsibility within a profession increasingly dominated by automation and algorithmic systems. The starting point was the observation that decisions once made by humans are now delegated to automated tools such as Smart Bidding and Performance Max, which do not reveal their internal logic. The study therefore asked: How do PPC practitioners negotiate knowledge, agency and responsibility under algorithmic opacity?
The project was qualitative and exploratory, drawing on the tradition of ethnographic studies in Science and Technology Studies (STS). It combined multiple qualitative and digital methods to reconstruct micro-practices of working with algorithms and to trace how knowledge and responsibility are distributed between human actors and systems. The core assumption was that what is invisible (for instance, the system's logic) can be studied through what is social – that is, through language, interpretive tactics, epistemic rituals and negotiations of meaning within professional communities (Knorr-Cetina, 1999; Clarke, 2015).
The project's goal was to capture the relationship between automation and professional epistemology among PPC specialists. Algorithmic opacity was treated not merely as a technical feature but as a social and cognitive problem that practitioners must actively manage.
The methodological design employed a triangulation of qualitative and digital approaches, allowing for a layered analysis of both subjective experiences and communal discourses. The study included three components:
This triangulation made it possible to transcend a single perspective and explore the profession from both ‘inside’ (subjective narratives) and ‘outside’ (discursive reproduction of norms and practices). Each method illuminated a different dimension of the relationship between human and algorithm: between individual and community, and between formal and embodied knowledge.
The empirical material consisted of 27 in-depth interviews with PPC specialists working in three environments:
Participants were recruited using the snowball method, drawing on professional forums and social media groups such as Facebook and LinkedIn. The selection was purposive: all participants worked daily with automation-driven tools (Google Ads, Meta Ads, etc.). Interviews were conducted online via Google Meet, lasting 45–90 min. Each was recorded, transcribed and anonymized. The Polish PPC environment, characterized by a strong reliance on peer-led learning and informal professional communities, provides a particularly revealing context for examining how interpretive labour compensates for limited institutional transparency.
The study followed the ethical principles of social research (British Sociological Association, 2017). Participants received full information about the project, voluntary participation and the right to withdraw at any time. All quotations were shortened and paraphrased to prevent identification.
Alongside interviews, the study analysed more than one thousand posts from public and semi-public online spaces. These materials were treated as netnographic data reflecting collective sense-making practices, informal language and everyday ways of interpreting algorithmic decisions.
For Facebook groups, the principle of ‘contextual visibility’ was applied: only posts shared with administrator consent were analysed, and all personal data were removed.
The analysis followed the framework of the Social Worlds/Arenas Theory developed by Strauss (1978) and Clarke (2015), which combines grounded theory with an interpretive focus on collective practices. Instead of searching for dominant narratives, the study aimed to map multiple positions, strategies and discourses that constitute the PPC professional community.
The coding process was open and iterative. Analytical categories emerged inductively from the data and were then refined through comparative analysis between groups (freelancers, in-house professionals, agencies). The resulting coding tree covered eight thematic areas that correspond to the social worlds and arenas of PPC practice:
The analysis focused on how specialists construct meaning in the face of automation. The aim was not to evaluate their technical competence but to uncover how concepts such as responsibility, control, intuition and expertise are negotiated across organizational contexts.
Coding and analysis were reflexive and collaborative. Categories were repeatedly verified against forum discussions to ensure consistency between individual experience and collective discourse.
A key dimension of the study was its autoethnographic reflection. The researcher had prior professional experience in the PPC industry, having worked in agencies, in-house departments and as a freelancer managing digital marketing teams. This dual position, practitioner and researcher, shaped both access and interpretation.
This familiarity with the field allowed access to high-threshold environments (closed groups, internal repositories, specialist jargon) and facilitated sensitivity to affective and ethical dimensions of work. Emotions such as frustration, relief, tension and guilt are central to understanding algorithmic labour.
The researcher thus acted as a technological translator: someone who can move between the logic of platforms and the language of sociology. Following the principle of methodological estrangement (Clarke, 2015), familiar practices were examined from a deliberate distance, allowing them to appear strange enough to be analytically grasped.
Autoethnography here did not serve as introspection but as reflexivity, situating knowledge within the researcher's position, history and body, in line with Donna Haraway's concept of ‘situated knowledge’ (Haraway, 1988). This perspective aligns with STS traditions that view researchers as participants in the co-production of knowledge rather than detached observers.
Combining individual, communal and reflexive layers of data made it possible to reconstruct the epistemic infrastructure of PPC work. This approach moves beyond occupational study to reveal how boundaries of knowledge and responsibility are drawn not between humans and machines, but within the practices that link them.
Studying ‘life with algorithms’ in this sense requires that ethnography itself become a form of epistemic translation, exposing how people create meaning within systems that were designed to replace them.
Findings: Epistemic strategies and professional micropolitics in the PPC environment
Research conducted within the PPC environment revealed a complex professional structure in which knowledge, power and responsibility are subject to ongoing renegotiation. In contrast to classical models of expert labour, the PPC expert operates in a space dominated by algorithmic systems – opaque, dynamic and, as Seaver (2017) puts it, ‘uninhabitable’ in an epistemological sense.
However, as the analysis shows, this does not imply the passivity of specialists. On the contrary, they develop a repertoire of epistemic practices, interpretive narratives, communal rituals and technical tactics aimed at managing uncertainty and dispersed responsibility.
Following the framework of Social Worlds Theory (Clarke, 1990; Strauss, 1978), the PPC field does not constitute a homogeneous community of practice. The study identified three distinct professional segments: agencies, in-house teams and freelancers, each operating within its own epistemic style.
Agency specialists typically manage numerous accounts and rely on procedures and internal benchmarks. Knowledge here takes an institutionalized form: it is scalable, rapidly updated and often delegated to tools. In an agency, campaigns are set up according to templates. When something works, you don’t overthink it – there's no time. What counts are results and standardization. (Interview, Agency 4, March 2025) In the group, it's like a lab. We share tests, screenshots, anomalies. Nobody knows how PMax works, but together we try to figure out what might be happening. (Post forum, August 2025)
The central challenge identified in the study was the opacity of algorithmic systems, commonly described in the literature as the black box (Pasquale, 2015). Empirical material shows that PPC specialists do not treat the algorithm as a mere tool but as a relational partner that requires care, intuition and constant interpretation. The algorithm is like a living creature. You have to feed it data, not overload it, not annoy it. If you change too much in the campaign, it sulks. (Interview, Freelancer 3, January 2025)
Instead of rational control, we encounter relational epistemology, a continuous reading of the signs generated by the system, the formulation of hypotheses and their testing under conditions of incomplete information. When a campaign falls apart, it's not always your fault. Maybe the system is learning, maybe it hit a different auction. But you still have to explain it, because the client won’t understand that the algorithm got offended. (Interview, In-House 7, March 2025) The client doesn’t want to hear ‘I don’t know what happened.’ So we learn to say ‘we’ve noticed a trend shift’ or ‘we’re testing a new model’ even when nothing has really changed. (Post forum, May 2025)
Such narratives function as epistemic masks, preserving the expert's status in a world where expertise no longer guarantees control over decision-making. As Pasquale (2015) observes, this is a form of ‘responsibility without transparency’, where professionals are formally accountable yet act retrospectively, interpreting events that have already occurred and could not have been predicted.
In conditions of non-auditability, embodied knowledge becomes central. This form of knowing is based on experience, sensitivity and heuristics. After years of work, you just feel it. A campaign starts behaving strangely, and you sense it before the data shows it. It's not knowledge you can transfer to someone else – you have to live it. (Interview, Agency 7, March 2025)
Finally, communities of practice play a crucial role in this epistemic ecosystem. Online professional groups function as collective epistemic infrastructures, providing access to knowledge, emotional support and shared interpretive norms. When something weird happens, I first check if others see it too. If ten people post that Google has cut reach again, then it's not my fault but the system's. (Post forum, February 2025) I’m reporting on something I don’t fully understand, but I have to look as if I do. (Interview, Agency 5, February 2025) The system suddenly changes something in the campaign. CTR drops, the budget drifts. The client asks what happened, and I say the algorithm is learning. It's a magic phrase that explains everything. (Interview, Freelancer 3, January 2025)
Similar phenomena appeared on industry forums. In one post, a user wrote: CTR dropped by half after the weekend. I didn’t change anything. The system is learning, so they’re testing something live again. (Post forum, February 2025) I can’t tell the client I don’t know, so I say it's a learning phase and ask for patience. Most of the time, it works. (Interview, Freelancer 6, March 2025) Is it just me, or is Google testing its AI on us? Nothing works the same as last week, and I still have to sell it to the client. (Post forum, May 2025)
In PPC practice, this blurred accountability takes the form of legitimation rituals: the creation of reports, charts and presentations designed to sustain an impression of control. Although these activities rarely lead to genuine understanding, they preserve the stability of professional relations and the symbolic coherence of the work environment. I prepare reports, explain changes, make presentations – all to make it look logical. Sometimes I feel more like Google's PR representative than a specialist. (Interview, In-House 9, March 2025)
As advertising systems become increasingly autonomous, PPC specialists’ knowledge shifts from procedural to embodied forms. This means that understanding no longer depends on knowing the algorithm but on developing sensitivity to its rhythms through repeated engagement with data interfaces. Knowledge becomes sensory, grounded in the body, emotion, and everyday gestures of work. After spending years in the data, you start to feel when something's off. You just sense it – like playing an instrument. (Interview, Agency 9, March 2025) After years of work, you know something's wrong before you see it in the numbers. It's intuition – you see the data, but you feel it earlier. (Interview, In-House 6, March 2025)
Several respondents described their practices in almost ritualistic terms, referring to the daily monitoring of campaigns, checking data at fixed times or intuitively ‘cleansing’ accounts. Such habits create a sense of continuity and control within processes that are inherently unstable. I always check the panel in the morning, even before coffee. It's like a ritual. When I see that something has shifted, I feel it right away. (Interview, Freelancer 9, March 2025) It's not something you learn from a course. It comes after a thousand campaigns and sleepless nights. Then you start to understand the system in your own way. (Interview, Agency 6, February 2025) I can’t explain it, but PMax performs better when I leave it alone. If I tweak too much, it starts acting weird. (Post forum, January 2025)
Intuitive knowledge therefore performs a compensatory function. In a world where system logic is hidden, it is experience and emotional response that enable action. The senses, body and affect replace access to code, turning intuition into an epistemic instrument. Embodied knowing becomes not only a form of competence but also a strategy for coping with algorithmic uncertainty. Through intuition, repetition and ritual, specialists generate stability where transparency is absent. This demonstrates that the epistemology of data environments is not solely rational but also affective and somatic. PPC specialists do not simply analyze data – they live within it, co-producing a sensory, relational mode of understanding.
A comparative analysis of the three segments of the PPC environment: agencies, in-house teams and freelancers, reveals the existence of distinct data cultures. Each organizational form produces its own epistemologies, practices and modes of sense-making in relation to algorithmic systems. Following the framework of Social Worlds Theory (Clarke, 1990; Strauss, 1978), these segments do not form a hierarchy but coexist as diverse interpretive orders within the technological landscape.
In agencies, work with algorithms is the most institutionalized. Systematic procedures, process standardization and measurability underpin the organizational culture. Knowledge circulates collectively and procedurally through shared spreadsheets, checklists and reporting tools. In the agency, campaigns are set up according to templates. When something works, you don’t overthink it. There's no time. What matters are results and standardization. (Interview, Agency 4, February 2025) We have a shared spreadsheet where everyone logs their observations – for example, that Performance Max performs better for premium products but worse for seasonal ones. It's like collective fortune-telling. (Interview, Agency 6, March 2025)
In-house specialists operate closer to organizational decision-making and business strategy. Their work is more context-specific, embedded within the goals of a single brand, but also burdened by managerial expectations. Within this culture, knowledge assumes a managerial form, articulated in the language of KPIs and ROI rather than experimental insight. I have to explain results to managers, but I can’t tell them I don’t know why something works. They expect clarity. (Interview, In-House 3, January 2025) I no longer feel like a specialist, but rather an intermediary between the system and management. (Interview, In-House 9, March 2025)
Freelancers represent the most flexible yet also the most precarious form of data culture. Their work relies heavily on intuition, experience and collective hypothesis testing within online communities. The absence of formal structures is compensated by participation in professional Facebook groups that function as collective laboratories. Someone posted a screenshot showing that Performance Max performs worse after headline changes, and now we’re all testing it. It's like crowdsourced knowledge. (Interview, Freelancer 7, March 2025) When something strange happens, I first check if others see it too. If ten people post that Google changed something, I know it's not just me. (Post forum, February 2025)
These practices confirm that online communities serve as key epistemic and emotional infrastructures for freelancers. They constitute horizontal networks of negotiation where collective interpretations of system behaviour emerge despite the absence of official explanations.
The epistemic diversity of the PPC field demonstrates that algorithmic knowledge does not exist in singular form. Agencies institutionalize stability, in-house specialists engage in epistemic diplomacy, and freelancers practice communal experimentation. In each of these cultures, data is not an objective resource but a medium of knowing, shaped through rituals, relations and localized sense-making practices.
The collected data indicate that in the context of growing automation, the work of PPC specialists becomes primarily interpretive. Yet this interpretive labour is not hermeneutic in the traditional sense but epistemic. Its goal is not to directly control the system but to sustain meaning and legitimacy within an environment characterized by technological opacity and instability. I no longer control campaigns the way I used to. Now I mostly observe and try to understand what the system is doing – and how to explain it to the client. (Interview, Agency 2, January 2025) When I talk to clients, it's less about changing data and more about understanding what it means. It's interpretation rather than optimization. (Interview, In-House 4, February 2025)
Opacity, therefore, becomes a stable feature of professional culture rather than a technical flaw. As the analysis of posts and interviews shows, practitioners learn to coexist with it by developing shared heuristics and communicative strategies that reintroduce order and reason. I don’t know why Performance Max sometimes works and sometimes doesn’t, but I know how to talk about it so it sounds professional. (Interview, Freelancer 4, February 2025) I used to think my job was to set up campaigns. Now I see it's about translating the system for the client and the client for the system. (Interview, Agency 7, March 2025) PMax changed the way it reports again – anyone else seeing drops in conversions? Not sure if it's a bug or a new update. (Post forum, June 2025)
The contemporary PPC specialist thus becomes a technological interpreter, a mediator between machine and social world. Their knowledge is performative (oriented towards producing an impression of rationality), relational (emerging through communication) and affective (grounded in intuition and emotion). PPC practice, in this sense, reveals a broader logic of life in the algorithmic age – not grounded in understanding technology, but in the capacity to produce meaning in its shadow.
Discussion: Epistemology under the algorithmic sky
Examining the PPC environment as an automated world of labour reveals new tensions in the relationship between technology, knowledge and responsibility. This is a professional sphere marked by epistemic uncertainty and the simultaneous pressure of performance and accountability. As the analysis shows, PPC specialists are not passive victims of black-box systems but social actors who develop tactics of resistance, translation and legitimation while navigating the micropolitics of automation.
In line with Seaver (2017, 2018) and Pasquale (2015), algorithmic opacity should be understood not only as a technical feature but as a social and occupational problem. PPC specialists do not suffer from a lack of information in the technical sense but from the absence of an available epistemology of action – one that would allow them not only to operate the system but also to take responsibility for its effects.
Hence the emergence of narrative epistemic work: post-factum explanations that legitimize decisions whose origins cannot be reconstructed. As Pasquale (2015) argues, this represents a form of responsibility without control, an illusion of agency within systems that do not grant it.
In the absence of audit tools, practitioners rely on embodied knowledge: experience, intuition, and the feel of the system. Contrary to popular assumptions that automation eliminates human involvement, the findings demonstrate that algorithmic labour requires intensive cognitive and affective engagement. This involves not only data analysis but also predicting system behaviour, managing client relations, and producing credible narratives.
This confirms the insights of Polanyi (1966) and Knorr-Cetina (1999): professional knowledge is not a static set of rules but a localized cognitive practice shaped through interaction with technology and community. Embodied knowledge functions as a form of social control, allowing specialists to regain stability in a world that no longer guarantees epistemic completeness.
The Social Worlds framework further reveals that access to epistemic security is unevenly distributed. Agencies have procedural support and technical resources but less flexibility; freelancers enjoy autonomy but face precarity; in-house specialists operate between strategic and operational demands. This inequality is not only organizational but epistemological: it concerns what counts as evidence, meaningful insight or credible explanation.
In this context, online communities operate as critical epistemic infrastructures. They compensate for the absence of official explanations and facilitate the circulation of practical, experiential knowledge unavailable in Google Ads documentation – for instance, tacit understandings of how ranking factors influence ad visibility.
Conclusion
This article examined how PPC specialists navigate responsibility under conditions of algorithmic opacity in platform-based digital advertising. Drawing on interviews, netnographic observation and technical documentation, it showed that practitioners remain accountable for campaign outcomes while lacking transparent access to the decision-making processes that increasingly shape those outcomes. Rather than eliminating professional agency, platform automation reconfigures it, shifting PPC work towards constrained intervention, interpretive labour and the management of expectations across institutional boundaries.
By situating contemporary PPC within longer histories of advertising automation and quantification, the analysis demonstrates that epistemic uncertainty is not new to the advertising profession. What is distinctive in the current configuration is the concentration of control within platform-owned infrastructures and the dynamic, continuously recalibrated nature of algorithmic optimization systems. These conditions intensify the gap between responsibility and control, making justification and narrative stabilization central components of professional practice.
The article contributes to scholarship on algorithmic work by specifying how agency persists under conditions of opacity. PPC specialists continue to act through indirect levers – defining success metrics, constraining optimization parameters and provisioning data – while simultaneously developing collective and tacit forms of knowledge to interpret system behaviour. These practices do not overcome opacity, but they enable continued action and accountability within asymmetrical systems of control. In this sense, interpretive and communal expertise should be understood not as a withdrawal from agency, but as a pragmatic response to its reconfiguration.
While PPC specialists cannot be treated as a straightforward microcosm of all algorithmically mediated professions, the case offers a theoretically generative site for examining how responsibility is organized when decision-making authority is delegated to opaque systems. The mechanisms identified here – responsibility without full auditability, probabilistic intervention and collective sense-making – are likely to recur across other platform-mediated domains, albeit in field-specific forms shaped by regulation, organizational structure and professional cultures. Comparative research is needed to specify the conditions under which similar configurations of agency and accountability emerge elsewhere. Rather than standing in for all algorithmic professions, PPC offers an analytically clear case in which responsibility, opacity and interpretive labour become unusually visible.
Finally, the Polish PPC context foregrounded in this study highlights the cultural and linguistic situatedness of interpretive labour. Local peer communities function as crucial infrastructures for sense-making, emotional coping and professional validation under automation. Attending to such contexts is essential for understanding how algorithmic responsibility is enacted in practice, not as an abstract ethical ideal, but as an everyday accomplishment negotiated at the intersection of platforms, markets and professional expertise.
Footnotes
Acknowledgements
I would like to thank my son, Janek, for his invaluable support, patience and inspiration throughout the process of researching and writing this article.
Funding
The author received no financial support for the research, authorship and/or publication of this article.
Declaration of conflicting interests
The author declared no potential conflicts of interest with respect to the research, authorship and/or publication of this article.