Algorithmic control refers to a form of digital management practice in the gig economy, whereby online labor platforms, based on embedded technical rules and standardized procedures, utilize AI algorithms driven by big data to autonomously execute control functions over gig workers who are dispersed across time and space (Graham et al., 2017; Wood et al., 2019; Pignot, 2023). Building on this foundation, Pei et al. (2021) defined perceived algorithmic control as gig workers’ comprehensive perception of how platform algorithms exert real-time, dynamic control over the online labor service process through standardized guidance, tracking evaluation and behavioral constraint.

In algorithm-driven gig work, interactions between workers and platforms heavily rely on systematic and automated rules. Research indicates that perceived algorithmic control is not merely a supervisory mechanism but can also function as a “technical support partner” by offering clear path planning and real-time feedback, thereby assisting workers in understanding and achieving performance goals (Pei et al., 2021; Chen et al., 2025). Against this backdrop, we anticipate that perceived algorithmic control can promote gig workers’ voice behavior. Existing empirical studies provide preliminary evidence for a positive relationship between the two. First, prior research has shown that perceived algorithmic control can enhance gig workers’ work engagement (Chen et al., 2024; Lin et al., 2025). Work engagement, characterized by vigor, focus, and dedication, activates individual resources and strengthens the motivation to improve work contexts, thereby facilitating voice behavior (Parker et al., 2010; Bindl et al., 2012; Liu et al., 2021). Second, perceived algorithmic control can also increase gig workers’ job embeddedness (Liang et al., 2025), which reflects the alignment, connectedness, and perceived sacrifices between individuals and their work context. Highly embedded employees are more concerned with the long-term development of the organization and are more willing to offer constructive suggestions (Kiazad et al., 2015; Tan et al., 2019; Tabrizi et al., 2023). Based on this, the following hypothesis is proposed:

Perception of algorithmic fairness refers to an individual’s subjective evaluation of whether an algorithmic system adheres to principles of fairness in data processing, decision-making, and outcome delivery (Wei et al., 2021). In the context of the gig economy, gig workers often operate in environments characterized by information asymmetry and cognitive constraints, making it difficult to conduct a comprehensive and rational assessment of algorithmic fairness (Jones-Jang and Park, 2023). According to FHT, when information is limited or contexts are uncertain, individuals are more likely to rely on salient and accessible cues to make heuristic fairness judgments (Lind, 2001; van den Bos et al., 1997). The theory emphasizes that experiences of procedural, distributive, and interactional fairness constitute the core basis for an individual’s overall fairness judgment (Long, 2004).

In algorithmic management contexts, perceived algorithmic control provides gig workers with these three critical fairness cues. First, algorithmic standardized guidance delivers procedural fairness cues by providing clear, consistent, and transparent work standards (Pei et al., 2021; Chen et al., 2025), enabling workers to better understand the basis of algorithmic decisions. Second, algorithmic tracking evaluation conveys interactional fairness cues through real-time, specific, and credible performance feedback (Wiener et al., 2021), reflecting the system’s respect and attention toward workers. Third, algorithmic behavioral constraint provides distributive fairness cues via clearly defined and consistently enforced incentive and penalty mechanisms (Griesbach et al., 2019), allowing workers to perceive a balance between effort and reward. Since these three types of cues originate from the same authoritative system (platform algorithms) and are presented continuously and synchronously in daily work, gig workers can readily integrate them into an overall perception of algorithmic fairness.

Fairness Heuristic Theory further suggested that once individuals form a positive overall fairness judgment, they enter a “group mode,” characterized by greater acceptance of authority and rules, enhanced identification, self-esteem, and trust, and an increased tendency to engage in pro-social behaviors that benefit the organization (Lind, 2001; Long, 2004). In algorithmic management contexts, perceiving algorithms as fair leads workers to regard platform algorithms as reliable and worthy of cooperation, thereby strengthening their motivation to contribute suggestions for collective benefit and encouraging constructive voice behavior (Walumbwa et al., 2012; Duan et al., 2007). Empirical studies also indicate that fairness perception, as an individual’s perceived organizational fairness, is a key organizational factor stimulating voice behavior (Bashshur and Oc, 2015; Colquitt et al., 2001). Therefore, this study posits that perceived algorithmic control, by providing gig workers with procedural, interactional, and distributive fairness cues, fosters their overall perception of algorithmic fairness, which in turn enhances prosocial motivation and promotes voice behavior. Based on this, the following hypothesis is proposed:

Voice endorsement embodies the platform’s positive feedback and value affirmation toward individuals’ voice behavior (Ni et al., 2024), representing the organization’s acceptance and recognition of the suggestions raised by individuals (Liang et al., 2024). Specifically, this construct encompasses three primary forms: psychological acknowledgment of the suggestion’s value, practical implementation of the suggestion into work routines, or escalation of the suggestion to higher-level authorities to promote its adoption (Burris, 2012; Jia et al., 2025).

First, a high level of voice endorsement signals that platforms effectively accept and implement the suggestions put forward by individuals (Lam et al., 2019; Jia et al., 2025), which enhances gig workers’ trust in the platform’s decision-making processes. For gig workers with perception of algorithmic fairness, they can understand the standards and process of voice adoption, and trust makes them more actively participate in voice behavior, thus more willing to share their ideas. Second, for gig workers with perception of algorithmic fairness, a high level of voice endorsement makes them feel that their ability and value are recognized by the organization (Zhang et al., 2020; Wang et al., 2024). This recognition elevates individuals’ self-esteem and self-efficacy to a certain extent, motivating them to respond positively to challenges and be more inclined to put forward further constructive suggestions. Moreover, organizational acknowledgment enhances individuals’ confidence and stimulates a proactive pursuit of personal growth and development. When gig workers perceive the idiosyncratic competitiveness of colleagues, they often experience a potential status threat, prompting greater efforts to participate in voice behavior to enhance their capabilities and ultimately maintain or improve their standing and competitiveness within the organization (Liu N.-T. et al., 2022; Liu P. et al., 2022). Based on this, the following hypothesis is proposed:

As discussed above, the preceding discussion on the mediating role of perception of algorithmic fairness aims to reveal the psychological mechanisms through which perceived algorithmic control influences gig workers’ voice behavior. The discussion of the moderating role of voice endorsement further clarifies the conditions under which perception of algorithmic fairness exerts a stronger impact on voice behavior. On the basis of the above analysis, this study further proposes a moderated mediation model. Specifically, voice endorsement indicates that individual’s suggestions are positively evaluated and adopted by the platform. This process enables gig workers to perceive the value and effectiveness of their voice, thereby enhancing their self-efficacy and sense of accomplishment. At the same time, it demonstrates the practical significance of their suggestions to the platform, which fosters their self-esteem and strengthens their trust in the platform. When perceived algorithmic control activates gig workers’ perception of algorithmic fairness and thereby promotes voice behavior, high levels of voice endorsement further boost their self-esteem and trust, making them more willing to offer constructive suggestions. Therefore, when the platform voice endorsement level is high, the positive effect of perceived algorithmic control on gig workers’ voice behavior through perception of algorithmic fairness is further strengthened. Based on this, the following hypothesis is proposed:

The target population of this study consisted of gig workers in China, specifically food-delivery riders and ride-hailing drivers (e.g., Didi drivers). A mixed-method approach combining online and offline survey distribution was adopted. To ensure data validity, participants were required to be currently working as a food-delivery rider or a ride-hailing driver. No minimum tenure threshold was imposed during recruitment. Respondents were asked to report their work tenure on the focal platform.

Online questionnaires were distributed via the Credamo data collection platform. Within Credamo, we targeted respondents who were currently working as food-delivery riders or ride-hailing drivers, and we used platform settings to restrict multiple submissions from the same account and IP address. The survey link was released in several waves to reduce the likelihood that the same users would repeatedly receive the invitation. Offline questionnaires were distributed in person by trained research assistants at locations where gig workers commonly gather, such as dedicated rest stations for food-delivery riders and charging stations for drivers. Research assistants briefly introduced the purpose of the study, confirmed that potential participants were currently working as food-delivery riders or ride-hailing drivers, and then invited them to complete the questionnaire on site on a voluntary basis. No recruitment posters or other passive methods were used, all offline participants were recruited through direct, face-to-face invitations.

All respondents participated anonymously and received a small material reward upon completion. A total of 306 questionnaires were collected (241 online and 65 offline). To ensure data quality and participant attention, strict quality control was applied to exclude invalid questionnaires, such as those with unrealistically short completion times or straight-lining response patterns. Ultimately, 260 valid responses (84.97% of the initial sample) were included in our empirical analysis. To assess potential sampling bias between online and offline recruitment channels, we conducted a series of homogeneity tests. Independent-samples t-tests revealed no significant differences in age (t = −0.668, p = 0.504) or work tenure (t = −0.381, p = 0.704) between the two subsamples. However, chi-square tests indicated significant differences in gender [χ²(1) = 4.47, p = 0.034], marital status [χ²(1) = 4.48, p = 0.034], and education level [χ²(4) = 42.86, p < 0.001], but not in job type [χ²(1) = 0.59, p = 0.441]. To address these demographic differences and mitigate potential confounding effects, we included all these variables as control variables in our subsequent regression analyses to account for these pre-existing differences, thereby ensuring the robustness of our findings.

All measures used have been validated in previous research. Given that all administered items were in Chinese, translation and back-translation procedures were followed to ensure the quality of translations (Brislin, 1986). Specifically, the items were first translated into Chinese by a bilingual researcher and then translated back into English by another independent bilingual scholar. The two versions were compared, and minor discrepancies were resolved through discussion to ensure the Chinese items accurately reflected the original meanings. Each measure used a 5-point Likert-type scale ranging from “strongly disagree” to “strongly agree.”

Table 1 depict the descriptive statistics and coefficients among study variables for the samples. See Table 1 for details, perceived algorithmic control was significantly correlated with voice behavior (r = 0.462, p < 0.01) and perception of algorithmic fairness (r = 0.709, p < 0.01). Perception of algorithmic fairness was significantly correlated with voice behavior (r = 0.639, p < 0.01).

Before hypothesis testing, we conducted confirmatory factor analysis using Amos to test the measurement model in both samples. As shown in Table 2, the hypothesized four-factor model (perceived algorithmic control, perception of algorithmic fairness, voice endorsement, and voice behavior) fit the data well (χ²/df = 1.665, RMSEA = 0.051, SRMR = 0.057, CFI = 0.932, TLI = 0.926) and was a significantly better fit than all other alternative models, indicating the four variables were distinct from each other.

Hypothesis development was conducted using ordinary least squares regression and bootstrapping analysis in SPSS 26.0, the results are presented in Table 3. Model 2 revealed that hypothesis 1, which predicted that perceived algorithmic control was positively related to gig workers’ voice behavior, was found to be supported (β = 0.593, t = 8.076, p < 0.001). As Table 3 shows, perceived algorithmic control was positively associated with perception of algorithmic fairness (β = 0.762, t = 15.465, p < 0.001), which in turn was positively related to voice behavior (β = 0.765, t = 13.18, p < 0.001). The results showed that perception of algorithmic fairness mediated the linkage from perceived algorithmic control to voice behavior (indirect effect = 0.574, SE = 0.096, 95% CI = [0.390, 0.763]). Thus, hypothesis 2 was supported.

Hypothesis 3 predicted that voice endorsement would positively moderate the relationship between perception of algorithmic fairness and voice behavior. The results of model 4 in Table 3 showed that the latent interaction between perception of algorithmic fairness and voice endorsement was significantly related to voice behavior (β = 0.183, t = 3.282, p < 0.01). The simple slope test in Figure 2 shown that the effect of perception of algorithmic fairness on voice behavior was significant when voice endorsement was high (M + 1SD) (simple slope = 0.672, SE = 0.159, p < 0.001), whereas the effect was non-significant at the low level (M-1SD) (simple slope = 0.324, SE = 0.173, p = 0.061). These results support Hypothesis 3.

Following the procedures recommended by Preacher et al. (2007), we utilized the PROCESS macro in SPSS to examine the full moderated mediation model. This analysis was conducted using 5,000 bootstrap resamples to generate 95% bias-corrected confidence intervals. As reported in Table 4, the results revealed that when voice endorsement was high (M + 1SD) (indirect effect = 0.504, SE = 0.123, 95% CI = [0.263, 0.737]) and low (M−1SD) (indirect effect = 0.255, SE = 0.128, 95% CI = [0.005, 0.512]), the indirect effects of perceived algorithmic control on voice behavior via perception of algorithmic fairness was significantly positive. While, the difference in direct effect between “high” and “low” values of voice endorsement was significant (indirect effect = 0.249, SE = 0.127, 95% CI = [0.007, 0.501]). Thus, Hypothesis 4 was supported.

This study employed semi-structured in-depth interviews with 22 gig workers. We adopted a purposive sampling strategy to recruit participants, aiming to intentionally recruit information-rich cases that were closely aligned with the research topic. Specifically, we visited locations where gig workers gather, such as food-delivery rider rest stations and ride-hailing driver charging rest areas, approached and invited individuals to participate in the interviews. We focused on two types of workers, specifically food-delivery riders and ride-hailing drivers, because they represent the most typical forms of gig work in China’s platform economy and effectively illuminate the common experiences of platform workers under algorithmic management (Huang, 2023).

Sample characteristics are as follows: Among the 22 interviewees, 21 were male and 1 was female. Age distribution: 20–30 years (n = 8), 30–40 years (n = 3), 40–50 years (n = 10), 50–60 years (n = 1). In terms of occupation, 14 were food-delivery riders (Ele.me: n = 7; Meituan: n = 7) and 8 were ride-hailing drivers (Didi: n = 8). Each interview lasted approximately 15–30 min and was transcribed. Table 5 presents detailed background information on the interviewees.

Prior to each interview, we informed all participants about the study purpose and confidentiality measures and obtained their informed consent. All interviewees were assigned pseudonyms, and interview data were used solely for academic purposes.

Based on the research questions, this study designed an interview outline focusing on three core dimensions: perceived algorithmic control (e.g., “How do you perceive platform management while working on this platform?”), perception of algorithmic fairness (e.g., “Do you think you are treated fairly while working on the platform?”), and voice behavior (e.g., “When you encounter problems at work, do you provide feedback to the platform? Could you give an example of a suggestion you have made?”). Extensive open-ended questions were incorporated to encourage respondents to fully express their lived experiences, capture underlying information and emergent themes, and allow for deeper exploration through follow-up questions, ensuring data richness and depth.

This study employed systematic grounded theory (Strauss, 1987), adopting an inductive coding approach to allow themes to emerge from the data. The specific process involved open coding, axial coding, and selective coding, to gradually construct a theoretical framework. Coding was performed independently by two coders who received training and engaged in discussions to establish unified coding standards. When disagreements arose, coders discussed and reached consensus through deliberation. Representative examples of the three-level coding process, including the correspondence from original interview quotes to each coding level, are displayed in Tables 6–9.

Following the practices of Boyatzis (1998) and Owens and Hekman (2012), the analysis ultimately identified nine specific categories under the three core structures of perceived algorithmic control, perception of algorithmic fairness and voice behavior. Figure 3 summarizes the themes that emerged from our data.

Building upon the quantitative findings, this study further incorporates qualitative analysis to uncover the subtle and complex psychological mechanisms linking gig workers’ subjective perceptions to their voice behavior. Specifically, it elucidates how perceived algorithmic control influences their willingness to voice through perception of algorithmic fairness. The interview results reveal that gig workers commonly perceive the management control of platform algorithms in terms of standardized guidance, tracking evaluation and behavioral constraint, which in turn shape their perceptions of algorithmic fairness. This cognitive process not only enhances their motivation to express opinions but also enhances their initiative to interact with the platform. It is worth noting that the study also finds that the common job insecurity of gig workers may pose potential interference to their voice behavior, which provides important clues for exploring the mechanism of individual psychological factors in the context of algorithmic governance in the future.

First, this paper extends the research on perceived algorithmic control to the domain of proactive behaviors, and enriches the literature on the relationship between perceived algorithmic control and voice behavior. Existing research on perceived algorithmic control has predominantly focused on its impact on gig workers’ job attitudes and performance outcomes, such as work engagement and turnover intention (Lang et al., 2023; Zhang et al., 2024; Yu et al., 2024). However, whether perceived algorithmic control can stimulate workers’ proactive behaviors aimed at improving organizational processes remains underexplored. At the same time, although recent studies have begun to examine voice behavior among gig workers, they tend to emphasize the channels (Bucher et al., 2024) and forms (Gegenhuber et al., 2021; Karanović et al., 2021) through which voice is expressed, while offering limited insight into why gig workers choose to voice and what intrinsic motivations drive such behaviors. In the context of algorithm-driven digital labor, this study systematically analyzes how perceived algorithmic control influences gig workers’ voice behavior through cognitive processes. In doing so, it not only broadens the theoretical application of perceived algorithmic control in the field of proactive behaviors but also provides a new theoretical perspective and analytical framework for understanding how gig workers demonstrate willingness to participate under platform governance structures.

Second, this paper reveals the psychological cognitive mechanism of perceived algorithmic control on voice behavior. While prior research has established that algorithmic control shapes workers’ attitudes and behaviors (Cram et al., 2022; Zhang et al., 2024), it has largely overlooked the underlying psychological processes. Drawing on fairness heuristic theory, this study introduces perception of algorithmic fairness as a mediating variable, clarifying how individuals form overall fairness perceptions based on previous experiences of procedural, distributive, and interpersonal fairness, thereby activating or inhibiting voice behavior. This paper not only illuminates the cognitive mechanisms underlying this influence process, but also substantiates the explanatory power of fairness heuristic theory in the context of digital labor, thereby enriching its application in emerging organizational forms.

Third, this paper clarifies the boundary conditions of gig workers’ voice behavior under perceived algorithmic control. By introducing voice endorsement as a moderator, this paper demonstrates that when platforms respond positively to worker suggestions, workers perceive their input as valued, which amplifies the positive effect of perception of algorithmic fairness on voice behavior. This finding not only identifies important boundary conditions in the influence path of perceived algorithmic control, but also provides a theoretical basis for understanding how platform organizations promote gig worker participation through voice feedback mechanism. Moreover, it extends the contextual understanding of voice behavior formation mechanism within the field of organizational behavior.

First, platforms should balance efficiency with human-centered care in the design of algorithmic control to enhance gig workers’ acceptance of algorithmic rules. This paper finds that gig workers are not entirely passive recipients of algorithmic management; rather, they cognitively evaluate the fairness embedded in algorithmic systems and decide whether to make suggestions accordingly. Therefore, platforms should avoid overemphasizing control and efficiency when designing task allocation and performance evaluation algorithms. Instead, they should enhance the transparency and explainability of algorithmic rules. For example, clearly explaining the rationale behind platform rules and the mechanisms for individual performance evaluation can foster a stronger sense of fairness among workers, thereby increasing their willingness to participate in voice.

Second, platforms should strengthen gig workers’ positive perceptions of algorithmic fairness by improving organizational mechanisms. The findings indicate that perceived algorithmic control does not necessarily provoke resistance; the key lies in whether workers perceive the algorithm as “fair.” Therefore, platforms should not only optimize the logic of algorithmic systems but also establish corresponding institutional mechanisms to support communication and participation. For instance, incorporating worker representatives in rule-making processes, organizing regular feedback sessions, and publishing algorithm transparency reports can all enhance the credibility and legitimacy of platform governance, thereby promoting stability and constructive interactions of gig labor relations.

Third, platform managers should establish responsive voice mechanisms that include timely acknowledgment, implementation updates, and explanations for non-adoption. This study finds that voice endorsement significantly amplifies the positive effect of perception of algorithmic fairness on voice behavior, indicating that workers’ willingness to continue speaking up depends on whether their voice is “heard.” Therefore, platforms should establish both official and unofficial voice feedback mechanisms, which should not only offer channels for expression, but also provide timely responses regarding whether suggestions are adopted and, if not, explaining the rationale. Such practices help cultivate an inclusive and respectful interaction culture, enhancing workers’ organizational identification and willingness for long-term collaboration.

First, this paper adopts a cross-sectional data collection approach, which limits the ability to capture the dynamic evolution of variables. Particularly in identifying causal relationships among perceived algorithmic control, perception of algorithmic fairness, voice behavior, and voice endorsement. Future research could adopt a longitudinal design using multi-wave data to analyze dynamic processes. This would allow for a clearer understanding of the temporal sequencing and interaction pathways among these variables, thereby enhancing the rigor of causal inference.

Second, regarding sample selection and data quality, the current study focused primarily on two typical platform occupational groups: food-delivery riders and ride-hailing drivers. We did not impose a minimum tenure threshold or specific content-based attention checks during recruitment. We did, however, apply procedural quality controls, such as screening for unrealistically short response times and straight-lining patterns, and included tenure as a control variable in our analyses. Nevertheless, this narrow occupational focus and the absence of stricter inclusion criteria may limit the generalizability and precision of the findings. Future research should expand the scope to include other gig roles, such as online content creation and home services, and consider implementing minimum experience requirements and explicit attention checks to further improve external validity and data reliability.

Third, in the qualitative interviews, the interviewed gig workers generally expressed concerns about job insecurity, which is related to high dependency and lack of labor security. However, the quantitative part of the current study has not yet sufficiently incorporated this psychological state into the theoretical model and tested it empirically. Future research is recommended to further focus on the generation mechanism of insecurity in the context of digital platform governance, as well as its potential impact on gig workers’ trust and voice behavior. Such efforts would provide a more comprehensive describe the psychological adaptation and behavioral reactions of gig workers in an algorithm-dominated environment.

Fourth, the generalizability of the qualitative findings is constrained by the gender imbalance in the interview sample, with only one female participant, reflecting the male-dominated on-site demographics of the sampled industries. While the core theoretical mechanisms under investigation are not predicated on gender, future research should purposively include more female gig workers to explore potential gender-based nuances in their perceptions and experiences.