Investigating Writing Professionals' Relationships with Generative AI: How Combined Perceptions of Rivalry and Collaboration Shape Work Practices and Outcomes

This study investigates how professional writers’ complex relationship with GenAI shapes their work practices and outcomes. Through a cross-sectional survey with writing professionals (n=403) in diverse roles, we show that collaboration and rivalry orientation are associated with differences in work practices and outcomes. Rivalry is primarily associated with relational crafting and skill maintenance. Collaboration is primarily associated with task crafting, productivity, and satisfaction, at the cost of long-term skill deterioration. Combination of the orientations (high rivalry and high collaboration) reconciles these differences, while boosting the association with the outcomes. Our findings argue for a balanced approach where high levels of rivalry and collaboration are essential to shape work practices and generate outcomes aimed at the long-term success of the job. We present key design implications on how to increase friction (rivalry) and reduce over-reliance (collaboration) to achieve a more balanced relationship with GenAI.

Introduction. Generative AI (GenAI) technologies are becoming an increasingly prominent topic of debate in writing professions. Central to this debate is how GenAI is reshaping professional roles, including their behaviors, work practices, and outcomes. A stream of scholarship adopts an automation perspective, arguing that GenAI’s capabilities can displace worker responsibilities, thereby threatening tasks and processes negatively [1, 56]. In contrast, an augmentation perspective argues that GenAI can create new opportunities for workers, expanding their scope and improving outcomes [18]. However, these perspectives adopt a top-down framing that centers GenAI’s technical capabilities rather than workers’ experiences. In practice, integrations are often bottom-up, where professionals actively interpret these technologies and reshape their roles [36]. Nascent HCI research adopts this bottom-up lens, showing how workers orientation towards GenAI and their relationship with it substantially shapes GenAI’s impact on work. Some workers view GenAI as a competitor that threatens their expertise, autonomy, and professional identity, fostering a rivalrous orientation [15, 91]. Individuals with this orientation resist GenAI when they perceive it as undermining their skills and devaluing their contributions [47]. Conversely, workers develop collaborative relationships with GenAI [92], understanding it as a supportive partner that helps them reflect on their work, refine their roles, and enhance outcomes [49]. A key gap in this scholarship, however, is that collaboration and rivalry have been treated as isolated phenomena. In practice, professionals often experience both simultaneously in different intensities, navigating parallel relationships with GenAI across different aspects of their work [15, 120]. Building on this work, our study investigates how writing professionals’ intertwined dual relationship with GenAI relates to their work. Specifically, we propose the following research questions:

RQ1. How do rivalry and collaboration orientations independently associate with work practices and outcomes? RQ2. How do rivalry and collaboration orientations, in combination, associate with work practices and outcomes?

Work practices were measured using job crafting and skill maintenance. Job crafting refers to bottom-up behavioral changes through which workers better align their work with personal meaning, values, and identity [128]. Skill maintenance complements crafting as an important practice that workers undertake once they have acquired and mastered a skill to reduce skill decay [6]. For work outcomes, we considered two important measures, perceptions of productivity and job satisfaction. To answer the research questions, we conducted a cross-sectional survey with 403 professional writers. For RQ-1, we proposed several hypothesis (H1-H6) and conducted confirmatory analysis using linear regression models. For RQ-2, we engaged in an exploratory analysis guided by the RQ-1 results. In particular, we applied Response Surface Analysis (RSA), which is based on polynomial regression,

Related work. In this section, we present related work to motivate our research questions and develop our hypotheses. First, we introduce the orientations of collaboration and rivalry towards GenAI. Next, we present prior work around both the orientations and their relationship with work practices (job crafting and skill maintenance) and work outcomes (productivity and satisfaction).

The growing ubiquity of GenAI is reshaping diverse domains of work, including creative [66, 120] and knowledge-based professions [70, 110], changing work structures and practices. These sociotechnical transformations can emerge through top-down mechanisms, where organizations or clients introduce GenAI tools and prescribe associated practices for employees or contractors [18]. Recently, these top-down mechanisms have been implemented in roles such as writing [37] and creative arts [52]. A significant portion of HCI research has studied the top-down deployment of GenAI tools in these work environments and resultant impact on working professionals [20]. At the same time, GenAI’s simplistic design and high-level capabilities, make them extremely accessible for workers. They enable bottom-up practices in which individuals exercise agency by adapting GenAI to reshape workflows to achieve desirable outcomes [36, 86]. However, these practices are shaped by how workers perceive GenAI in their work and how such perceptions guide their orientation and relationship with it. Historically, HCI literature shows that individuals adopt diverse orientations toward technology. One set of orientations assumes technology as a tool [5]. Under this framing, users focus on the instrumental and material aspects of technology, appropriating and reconfiguring its affordances within situated practice [39]. A second set of orientations treats technology as a medium [5], in which users attend to its expressive and communicative properties, viewing it as enabling or constraining common ground, coordination, and interaction [9]. In the context of AI, orientations increasingly position the system as a counterpart or social actor [5, 78]. Here, users respond to properties such as inscrutability, adaptivity, and hyper-personalization, and orient themselves towards these interactions and power dynamics by either aligning/integrating the system or challenging/avoiding it. As GenAI offers increasingly diverse capabilities, examining such bottom-up orientations and practices is critical because they challenge established norms and create new expectations within work roles. On one hand, workers perceive benefits in integrating GenAI into their work. They may demonstrate a collaborative stance by actively adopting GenAI, enhancing workflows, and using these interactions to improve outcomes [3, 111].

Method. To test our hypothesis, we conducted a cross-sectional survey study with professional writers who worked with GenAI in various capacities in their work. The study was approved by the IRB, preregistered 1, and administered via Qualtrics. We used Prolific to recruit professional writers in line with prior HCI research [38]. We employed stratified sampling techniques to find participants that had positive as well as negative perceptions around GenAI with an aim to capture enough participants with strong rivalry and collaboration perceptions. We used Prolific’s built-in sampling tools for this purpose. First, we selected the survey participants to be representative of the U.S. population by age, sex, and race. Second, we shared the survey to only participants above 18 and were employed full-time in a field that was writing-focused (e.g., journalist, author). Third, we also defined additional custom criteria through a pre-screener to shortlist participants with negative as well positive views around AI by asking them a few questions (e.g., “How do you view the overall effect of GenAI in your work?”). To determine a sufficient sample size, we conducted a power analysis a priori using G*Power with a desired effect size of .25, power of .99, and error probability < 0.05. Power analysis indicated that we needed at least 232 participants. We recruited a total of 450 writers (see Table 1) for the survey and paid them at Prolific’s suggested rate of 14.4$/hour (our survey took an average of 15 mins. to complete).

The survey instrument for the study was developed to assess writing professionals’ perceptions of their relationship with GenAI as well as their practices and attitudes relating to the technology. This process entailed the adaptation of existing validated measures that were not originally AI-specific, such as scales assessing interpersonal relationships of rivalry and collaboration and general job crafting scales, and the conversion of skill maintenance constructs into a self-report format. Scale adaptation proceeded in multiple stages. First, an initial qualitative study involving interviews with writing professionals was conducted to identify how the focal constructs manifested in writers’ work roles. Subsequently, triangulation between the coded interview data and the existing validated scales (or behavioral index) guided the adaptation of items to ensure conceptual alignment and face validity. Content validity was then evaluated by five subject matter experts and items were reworded until consensus was reached.

Overall, our survey measures focused on capturing four categories of constructs - (a) workers’ perceptions about GenAI (rivalry and collaboration), (b) their practices (job crafting), (c) their work outcomes (skill maintenance, productivity, job satisfaction), and (d) their GenAI usage rates and GenAI task involvement, included to illuminate descriptive patterns of different groups of writers. All items were measured with seven-point Likert scales. Descriptive and reliability measures of all the items are presented in the Table 2. For the instruments that were modified and adapted to fit the study, we have provided additional details on how we translated and validated them. All the items are listed in the supplementary material.

3.4 Positionality In this study, our objective was to understand how professionals’ orientations toward GenAI shape their work practices and subsequent outcomes. This question is particularly salient given ongoing concerns surrounding the development of large language models, such as exploitative data work [119] and training on copyrighted content [121], as well as their consumption, including issues of fairness [118], privacy [50], and misinformation [105], among others. To develop a comprehensive understanding of the research problem and produce equitable findings, we focused on capturing bottom-up and diverse worker-led perceptions and practices, whith diversity of perspectives, including workers both in favor of and opposed to incorporating GenAI into their work.

Discussion. Our results suggest there is substantial complexity in how writing professionals are orienting towards GenAI and forming relationships with it. In this section, we interpret these findings to provide (1) implications such orientations have on writing professions and shaping the profession as a result, and (2) design suggestions that can balance orientations of both rivalry and collaboration with GenAI to maximize GenAI-mediated outcomes in work.

At a broad level, professionals with either a rivalry or collaboration stance shaped their roles in similar ways, both by taking on new responsibilities (approach crafting, H5) and by reducing existing ones (avoidance crafting). These findings depart from prior rivalry literature [45] and our hypothesis (H4), which associated rivalry orientation only with approach crafting and not avoidance crafting. One explanation for this outcome is that a rivalrous orientation towards GenAI may prompt professionals to re-evaluate their role and reduce tasks that no longer add value in the face of GenAI competition (e.g., copyediting). Instead, they used GenAI or other technologies to handle such tasks. Section 3.3.2 presented partial evidence for this, where profiles with high rivalry reported having more GenAI experience than the baseline.

Evidence from the study also points to collaborators spending more time maintaining new skills required for GenAI integration, such as making GenAI outputs more human (Section 4.4.2). These “humanizing AI” activities may explain why collaboration stance was associated only with cognitive skill maintenance, as these individuals still relied on their traditional cognitive skills to make sense of GenAI outputs and incorporating them in their workflows. This finding contradicts prior results suggesting collaboration stance risks disengagement from cognitive skills [70, 120]. The differences in crafting practices between orientations were also reflected in the outcome measures, where only a collaborative stance was primarily associated with higher levels of productivity and satisfaction in the short term (see Section 4.5). This outcome highlights that while rivalrous professionals crafted their jobs with similar intensity, they did not experience the same levels of productivity or satisfaction, a combination previously linked to elevated strain and burnout risk [30]. Conversely, collaborators’ higher productivity and satisfaction coincided with patterns that may involve greater reliance on GenAI. This is concerning given that many GenAI systems are inherently designed to minimize friction and streamline adoption [21]. For instance, some tools offer scaffolded prompts (e.g., “Do you want me to provide an updated draft with the changes?”), making it effortless for users to interact. However, such nudges increase workers’ reliance on them. Taken together, these findings point to a recurring theme of the trade-offs between short term gains (productivity and satisfaction) and long term sustainability (skill maintenance) for workers who lean strongly toward a particular stance in their relationship with GenAI. In contrast, adopting a combined stance of collaboration and rivalry created a productive tension, yielding stronger associations with job crafting (Section 4.3) and productivity (Section 4.5). In particular, HighR/HighC profiles showed the strongest outcomes, indicating that collaboration was strongly associated with these outcomes, with rivalry showing an additional positive association. For skill maintenance (Section 4.4) and satisfaction (Section 4.5), the joint model explained variance comparable to the independent model, producing similar results. In other words, the presence of both orientations at higher levels yielded a more balanced effect.

Conclusion. This study presents critical evidence on writing professionals’ complex relationship with GenAI and its impact on their work practices and outcomes. In particular, an excessive leaning toward either rivalry or collaboration with GenAI was associated with greater imbalances, where individuals prioritized long-term career development at the expense of short-term outcomes, or vice versa, but not both. By contrast, combining rivalry and collaboration orientations was associated with improved balance, enabling writers to shape their practices for long-term growth while simultaneously benefiting from short-term outcomes. Based on these results, we present key design ideas, such as introducing micro-frictions to increase rivalry while maintaining collaboration levels to improve work pathways for professional writers.

Limitations. Our study has a number of limitations that highlight valuable areas for future research. First, we confined our respondents to writing professionals because their occupation is among the first to be severely disrupted by GenAI. Future research is required to evaluate whether our findings generalize to other professions, and to evaluate how professionals’ perceptions and practices evolve over time. Second, our findings are vulnerable to common methods bias due to reliance on self-reported measures collected at a single point in time. While cross sectional surveys are valuable for capturing perceptions at scale and across diverse respondents, they reflect subjective impressions rather than objective behavior, may be influenced by common method, recall, or social desirability biases, and do not support causal inference [124]. Future research using longitudinal designs, multiple raters, and behavioral data is needed to evaluate the directionality and robustness of the associations we observed. Our study was conducted on Prolific, and although we took measures to recruit a diverse range of writing professionals, it is possible that our sample was skewed towards professionals reporting high collaboration with GenAI due to Prolific’s tendency to attract participants who are more likely to adopt pro-GenAI stances.