Generative AI adoption and solvers’ popularity on supply-driven crowdsourcing platforms: the dual role of price signals

Supply-driven crowdsourcing platforms have reshaped how creative work is sourced and delivered in online labor markets (Ge et al., 2015). Unlike demand-driven platforms, where customers initiate tasks and service providers respond (Deng and Joshi, 2016), supply-driven platforms rely on solvers proactively marketing their capabilities by posting service listings (commonly referred to as “gigs”) (Brunzel, 2024). For example, on Fiverr alone, over 4.3 million active buyers offer services across more than 600 categories (Fiverr, 2023). A key characteristic of these platforms is that the services offered are often creative work, particularly in competitive domains such as graphic design and content creation (Brunzel, 2024). The focus on creative tasks exacerbates the information asymmetry inherent in these settings, where it is difficult to assess a solver’s skill.

To stand out from competitors and capture potential clients’ attention in a saturated online marketplace, solvers must utilize observable signals to demonstrate competence and to differentiate themselves (Piazza et al., 2022). Solver popularity, which reflects the level of attention a solver receives from potential clients (Xu et al., 2022), emerges as a critical determinant of success. Typically measured through platform metrics such as likes, comments, and ratings, solver popularity not only reflects the effectiveness of a solver’s signaling strategies but also plays a central role in shaping consumer decisions (Ladhari et al., 2020; Xu et al., 2022).

The advent of Generative AI (GenAI) is transforming creative tasks on supply-driven crowdsourcing platforms. Powered by advanced deep learning models, GenAI tools enable solvers to generate novel content more efficiently (Lysyakov and Viswanathan, 2023; Zhang et al., 2025). Consequently, many solvers use GenAI to assist with the creation, modification, and/or refinement of their creative work. On platforms such as Fiverr, the use of GenAI is publicly disclosed on the solvers’ service pages, either in the form of a display image co-created with GenAI or in the service description about the use of GenAI tools, suggesting that the adoption of GenAI is not just a technical choice to improve output but a deliberate strategic choice to showcase capabilities. This signaling strategy, however, presents a crucial dilemma. While signaling GenAI adoption is intended to convey enhanced creativity and productivity (Noy and Zhang, 2023; Peng et al., 2023), it has been met with potential consumer skepticism regarding authenticity and originality (Eloundou et al., 2024). Therefore, the effect of GenAI adoption on solver popularity remains unclear, leading to our first research question (RQ1):

To address this question, we draw on signaling theory, which involves one party (the sender) conveying private or unobservable information to another party (the receiver) through observable signals (Spence, 1973). In our study, we conceptualize the solvers’ indication of GenAI adoption as a signal to potential clients. We examine how signaling of GenAI adoption interacts with price to shape perceptions about the solvers and their creative products (Hsiao et al., 2024; Tian et al., 2023). Typically, higher prices signal premium quality, while lower prices indicate efficiency and tend to attract cost-sensitive customers. However, when pricing signals are combined with GenAI adoption, the interpretation becomes more complex. GenAI may be perceived as enhancing creativity and justifying higher prices, yet its efficiency gains could also enable cost-saving strategies that lead to lower prices (Chui et al., 2023; Rane, 2023). These competing logics lead to our second research question (RQ2):

To investigate these questions, we compiled a 14-month panel dataset from an international crowdsourcing platform from August 2022 to September 2023. This period includes seven months before and after the platform introduced a designated AI service category, marking a key shift in how GenAI was adopted. Focusing on solvers in design-related categories, we apply propensity score matching (PSM) and multi-period difference-in-differences (DID) to identify the causal effects of GenAI adoption on solver popularity and examine how price influences this relationship.

Our study offers several contributions. First, it offers a fresh perspective on GenAI’s impact on crowdsourcing. We provide empirical evidence that GenAI adoption enhances solver popularity on a supply-driven platform, highlighting the signaling value of this adoption beyond productivity gains. Second, we extend signaling theory by demonstrating the dual role of price in its interaction with GenAI adoption. When GenAI-assisted gigs are priced higher than their non-GenAI-assisted counterparts, price enhances the positive effect of adoption (boosting effect), but when priced lower, it weakens the signaling effect (boomerang effect). Third, the findings from our heterogeneity analysis extend the literature on digital inequality and enrich the discussion on the differential impacts of technology adoption.

A supply-driven crowdsourcing platform is a distinctive type of online marketplace where solvers are the primary drivers of the services offered (Brunzel, 2024). Unlike demand-driven platforms, where customers initiate tasks and solvers bid for them, supply-driven platforms rely on solvers proactively showcasing their capabilities by posting gigs. Hence, solvers must differentiate themselves through strategic self-branding and service positioning (Piazza et al., 2022). The concept of popularity becomes paramount. Popularity, typically measured by metrics such as likes, comments, and scores, plays a central role in influencing consumer decisions and solver success (Hong and Pavlou, 2017; Xu et al., 2022). A high level of popularity conveys trustworthiness, assuaging buyer concerns inherent in transactions with unknown solvers (Xu et al., 2022). Therefore, in the context of supply-driven crowdsourcing, popularity is not merely a metric but a critical determinant of solver success.

The rise of GenAI tools, such as ChatGPT and DALL-E, can change the work of solvers on supply-driven crowdsourcing platforms (Wagner et al., 2021; Yuan and Chen, 2023). GenAI can generate creative content through advanced machine learning techniques, particularly deep learning models (Stokel-Walker and Van Noorden, 2023). GenAI’s advanced creative capabilities set it apart from traditional AI, as GenAI can create new and seemingly original content that exhibits a level of creativity once thought unique to humans (Agrawal et al., 2022; Benbya et al., 2024; Zhang et al., 2025). GenAI can serve as a powerful tool for solvers, potentially boosting their productivity. The human-GenAI collaboration typically occurs during the creative process. For instance, a designer could use GenAI to instantly generate multiple concept prototypes, then manually refine and finalize the chosen design, iterating on ideas quickly (Zhao et al., 2024).

Research suggested that GenAI-driven solutions can rival or even surpass humans in creative problem-solving novelty (Benbya et al., 2024). GenAI also introduces new challenges and uncertainties for creative labor (Brynjolfsson et al., 2023; Hui et al., 2024). On the one hand, workers who used to specialize in creative domains might find their services less in demand if they can be largely automated by GenAI (Hui et al., 2024). On the other hand, widespread availability of GenAI can increase competition; less skilled workers may use GenAI to produce passable work (known as “workslop” by Niederhoffer et al., 2025), flooding the market with cheaper products (Brynjolfsson et al., 2023). Therefore, the impact of GenAI on supply-driven crowdsourcing platforms warrants further investigation.

Unlike traditional workplaces, online supply-driven crowdsourcing platforms operate as decentralized marketplaces, where each solver independently decides whether to adopt GenAI. In this context, adopting GenAI is not only a strategic decision aimed at enhancing service capabilities but also serves as a visible signal of quality to potential customers (Noy and Zhang, 2023). This signaling is made explicit when solvers choose to disclose their use of GenAI tools in their gig descriptions. However, it is unclear how consumers perceive GenAI adoption by solvers. On the one hand, it may increase the perceived value of services through improved performance (Noy and Zhang, 2023); on the other hand, it may lead to skepticism if GenAI-assisted work is seen as less authentic or original (Hui et al., 2024). Despite the growing interest in GenAI, its broader implications for solver outcomes in decentralized labor markets remain underexplored.

In supply-driven crowdsourcing platforms, where consumers often find it difficult to assess a solver’s skills, the price set by the solver serves as an important informational cue. Price is not merely a monetary value but a strategic signal solvers use to position their offerings (Ba and Pavlou, 2002; Li and Hitt, 2010). The influence of price on consumer choice is complex and can lead to opposing outcomes (Völckner, 2008). Prior research suggested that price can produce a “boosting effect,” where higher prices enhance perceptions of quality and increase purchase intention (Ba and Pavlou, 2002; Li and Hitt, 2010). Conversely, price can also trigger a “boomerang effect,” where it backfires if perceived to be too high or misaligned with other cues, causing consumers to question the product’s value and thus weakening its appeal (Fan et al., 2022; Jang and Chung, 2021).

Moreover, insights from price literature suggest that which of these effects occurs is context-dependent, as consumers tend to focus on the information most relevant to their choices (Hsiao et al., 2024; Jang and Chung, 2021). For example, in the mobile gaming market, Jang and Chung (2021) found that high absolute prices of add-ons positively influence sales by signaling superior quality, while the price of new add-ons—compared with that of the base product or existing add-ons—exerts a negative impact. As more solvers adopt GenAI on supply-driven crowdsourcing platforms, the relative prices between GenAI gigs and non-GenAI gigs emerge as a crucial indicator for understanding how consumers interpret GenAI adoption signals.

Numerous studies indicated that GenAI reduces production costs, thereby lowering the prices of GenAI-generated products (Frey and Osborne, 2023; Wamba et al., 2023). The lower prices enabled by GenAI can boost order volume and overall revenue, allowing a broader range of consumers to access previously cost-prohibitive products (Zhang et al., 2023). This price strategy can enhance market competitiveness (Rane, 2023), given that GenAI-generated products offer comparable quality at reduced prices. However, Kanbach et al. (2024) warned that the proliferation of low-priced GenAI outputs may disrupt traditional market structures and devalue higher-priced offerings created without GenAI. These developments imply that consumers may increasingly expect GenAI gigs to be priced lower than non-GenAI gigs. However, a growing body of research indicates that creators may conversely price GenAI-assisted works at a premium to signal enhanced creativity or quality (Wessel et al., 2025; Zhang et al., 2025). Aligning with this stream of research, our observations of supply-driven crowdsourcing platforms reveal a similar trend. We noted that some solvers price GenAI gigs higher than non-GenAI gigs to signal superior quality. Price strategies could determine how consumers interpret the same price signal differently.

Signaling theory, first proposed by Spence (1973), is fundamentally concerned with reducing information asymmetry between the signaler and the receiver(s). It seeks to explain how a signal affects the behavior of the parties involved, where receivers may have access to different information (Cheung et al., 2014). It has been widely applied across online settings to examine information asymmetry (Piazza et al., 2022). Signaling theory is particularly relevant to supply-driven crowdsourcing platforms, where consumers (receivers) face uncertainty and solvers (signalers) must actively communicate their value. A solver’s adoption of GenAI can serve as a signal of enhanced capability to prospective customers (Boussioux et al., 2024). Additionally, we are interested in how the interpretation of a technological signal is shaped by its interaction with traditional signals, such as price, which is underexplored in existing literature. Figure 1 presents the research model and the hypotheses, which are discussed next.

Previous studies suggested that adopting new technologies can improve the popularity of service providers on digital platforms by signaling advanced capabilities to potential customers (Marchiori et al., 2022). GenAI represents state-of-the-art technology capable of autonomously producing creative outputs (Brynjolfsson et al., 2023; Yuan and Chen, 2023). Supply-driven crowdsourcing platforms have reshaped how creative work is sourced and delivered in online labor markets (Ge et al., 2015), and the innovative nature of GenAI makes it particularly well-suited for creative tasks. On supply-driven crowdsourcing platforms, where solvers proactively advertise their services, and customers face information asymmetry regarding service quality, signaling theory provides a useful lens to understand the role of GenAI adoption (Brunzel, 2024; Piazza et al., 2022).

Recent studies have shown that GenAI can be leveraged to improve the quality and novelty of creative works (Boussioux et al., 2024; Stokel-Walker and Van Noorden, 2023). For example, Boussioux et al. (2024) highlighted GenAI’s ability to solve complex creative problems, implying that solvers using GenAI can deliver superior results. When solvers publicly indicate that their gigs are GenAI-assisted, they send a message to clients that they are technologically adept and capable of delivering higher-quality and more innovative work. By adopting GenAI, solvers reduce uncertainty for clients, signaling superior ability and service value relative to competitors who do not adopt GenAI. In supply-driven crowdsourcing, solver popularity is often manifested through customer engagement metrics such as the number of comments or favorites on their service (Brunzel, 2024; Gao et al., 2024; Huang et al., 2019). Perceiving the solver as an expert with advanced tools, consumers are more likely to visit the solver’s page, inquire about services, and ultimately place orders. In sum, through the lens of signaling theory, the adoption of GenAI functions as a credible signal within the context of supply-driven crowdsourcing platforms, as it helps reduce information asymmetry and reassures clients of the solver’s value, leading to greater solver popularity. We thus propose that solvers who adopt GenAI are more likely to receive more comments and likes, increasing their popularity on the platform.

Based on our literature review, price can produce either a “boosting” or “boomerang” effect depending on the context (Cai et al., 2016; Hsiao et al., 2024; Jang and Chung, 2021). We draw on signaling theory to explain the mechanism underlying these outcomes. In markets with information asymmetry, price acts as a potent signal, and its complexity arises from its dual effects on consumers: an information effect and a sacrifice effect (Jang and Chung, 2021; Völckner, 2008). From an information perspective, price influences consumers’ decisions in three ways. First, higher prices are often associated with better quality, thus affecting purchase choices (Ba and Pavlou, 2002; Li and Hitt, 2010). Second, price can signal prestige, thus increasing purchase intent (Lichtenstein et al., 1993; Yang et al., 2022). Third, expensive products may enhance consumers’ enjoyment and excitement, thereby further boosting their purchase likelihood (Hirschman and Holbrook, 1982; Völckner, 2008). Collectively, informational cues from price help reduce consumer uncertainty. However, from a sacrifice perspective, higher prices impose a financial burden on consumers, thereby limiting their budget and increasing their opportunity costs (Jang and Chung, 2021; Völckner, 2008).

A solver’s adoption of GenAI sends a powerful signal. When this signal is combined with a price signal, consumers interpret them jointly. The difference in prices between GenAI gigs and non-GenAI gigs creates two distinct signaling contexts shaping how consumers interpret the combined signals of technology adoption and price. On one hand, GenAI is known to reduce production costs and increase efficiency for content creation (Frey and Osborne, 2023; Wamba et al., 2023). As a result, consumers may expect GenAI gigs to be priced lower than comparable manual services, since the solver’s cost to deliver the work is presumably lower. On the other hand, some solvers deviate from this expectation and price their GenAI gigs higher than their non-GenAI gigs to signal enhanced creativity or quality. Each scenario (solver’s GenAI gig priced above vs. below non-GenAI gigs) sends a different composite signal to potential customers, thereby setting the stage for either a boosting or boomerang effect in the impact of GenAI adoption on solver popularity.

When a solver prices GenAI gigs higher than non-GenAI gigs, it creates a distinctive context that frames GenAI adoption as a value-added offering. By charging a premium for GenAI gigs, the solver’s price sends a bold signal that the GenAI gigs offer unique value or superior quality. It frames GenAI adoption within a cognitive schema of enhanced creativity and superior quality rather than cost-saving efficiency. In this scenario, a high average price creates signal consistency, leading the consumer to resolve the initial ambiguity in favor of a positive interpretation. In other words, the solver’s price strengthens the positive effect of GenAI adoption on the solver’s popularity (boosting effect). Therefore, we argue that for solvers who price GenAI gigs higher than non-GenAI gigs, price enhances the positive relationship between GenAI adoption and solver popularity.

Conversely, when solvers price their GenAI gigs lower than non-GenAI gigs, they align with consumers’ default expectation that GenAI services should be more cost-efficient (Frey and Osborne, 2023; Wamba et al., 2023). However, this low-price context may create signal inconsistency when interpreted alongside the solver’s absolute price positioning. Specifically, if a solver who adopts GenAI maintains high overall prices positioning but instead sets a lower price for their GenAI gigs, this inconsistency can trigger negative consumer inferences. From the signaling perspective, the information effect suggests that consumers may assume that lower-priced GenAI gigs reflect lower quality (Piazza et al., 2022; Spence, 1973). The sacrifice effect suggests that consumers might question the lower-priced GenAI gigs as a solver’s gig prices get higher. Hence, the price signal backfires. The lower price of GenAI gigs may be interpreted as indicating lower quality, greater reliance on automation over personalization, or reduced solver involvement (Frey and Osborne, 2023; Wamba et al., 2023). The higher the solver’s prices, the stronger this dissonance, as the low-priced GenAI offerings appear out of place relative to the solver’s price signals. This phenomenon leads to a boomerang effect, whereby the higher the solver’s prices of gigs, the weaker the positive impact of GenAI adoption on solver popularity. We thus propose the following hypothesis:

We collected the data from Fiverr (hereinafter referred to as Platform-A), one of the world’s largest supply-driven crowdsourcing platforms, with over 4.3 million active buyers from more than 160 countries worldwide (Fiverr, 2023). Platform-A allows solvers to proactively advertise their services, where clients can select these services. Given this supply-driven structure, Platform-A provides an ideal context to examine how technological signals such as GenAI adoption and price affect solver popularity in the context of information asymmetry. To support GenAI-assisted services, Platform-A introduced an “AI Artists” category in January 2023, allowing solvers to explicitly label which gigs are generated using GenAI tools (e.g. Midjourney, Stable Diffusion). Many “AI Artists” solvers previously provided traditional manual services and then expanded their gigs to include GenAI-assisted content services, making it feasible to analyze within-solver price differentiation and signaling strategies. Figure 2 shows two gigs by a solver. The left gig uses GenAI tools, and the right gig offers the traditional service without using GenAI tools.

To understand the impact of solvers’ GenAI adoption on their popularity, we obtained panel data from August 2022 to September 2023. The dataset consisted of 1,618 solvers, including 555 solvers with GenAI gigs and 1,063 solvers without GenAI gigs. Following the approach of Guan et al. (2023), we address the complex heterogeneity effects that arise when solvers post multiple GenAI gigs. To disentangle these effects, for solvers who posted more than one GenAI gig, we only retained data related to their first GenAI gig. To examine how the price context influences the effect of GenAI adoption, we categorized solvers into two distinct groups based on the difference in prices of their GenAI gigs and non-GenAI gigs. Specifically, we compared each solver’s average price for GenAI gigs with their average price for non-GenAI gigs. Solvers who consistently priced their GenAI gigs higher than non-GenAI gigs across all the available months were assigned to Group 1, while those who consistently priced their GenAI gigs lower were assigned to Group 2. A negligible number of solvers with inconsistent pricing strategies (e.g. pricing GenAI gigs higher in some months and lower in others) were excluded from the analysis to avoid potential confounding effects.

We adopted a quasi-experimental approach to test our hypotheses using two unbalanced panel datasets, consisting of a solver profile table and a gig list table. The solver profile table contains personal information about the solvers, including their level, country/region, Fiverr registration date, availability of social media accounts (for identity verification), education history (if provided), list of certifications received (if provided), average response time to client inquiries, and most recent date of delivery. Meanwhile, the gig list table provides details on the gigs posted by solvers, including the number of comments, scores, and favorites and their task descriptions. We linked these tables to the solvers via their worker ID to facilitate the analysis. We excluded those solvers who solely offer GenAI gigs with no traditional manual services (non-GenAI gigs). We removed solvers with both zero comments and zero likes to eliminate dormant accounts that are uninformative for our analysis, as they have no transaction history and no visible market engagement. Table 1 shows the descriptions of the variables in this research.

Following previous studies, we measured the dependent variable, solver popularity, using two indicators, namely, Like_it and Comment_it (Gao et al., 2024). Crowdsourcing platforms provide various means for consumers to express their appreciation and satisfaction with a solver’s work, such as giving likes and leaving comments. We thus measured solver popularity based on the average number of likes and comments their gigs received. Specifically, we measured the average number of likes and comments received by solver i in month t using the variables Like_it and Comment_it, respectively. To ensure the validity of the results, we conducted additional analyses with scores as the dependent variable. While likes are a non-transactional measure of general appeal that can be indicated by any user, scores are a transactional measure of client satisfaction provided only after a purchase has been completed. We measured the average score received by solver i in month t using the variable Score_it.

We used TreatmentGroup_i and AfterTreatment_it as the independent variables. If solver i adopted GenAI on the crowdsourcing platform during the study period, then TreatmentGroup_i equals 1; otherwise, it equals 0. Using the time at which each solver adopts GenAI as the cut-off point, if solver i belongs to the treatment group and adopted GenAI in month t, then AfterTreatment_it in month t and all months thereafter equals 1; otherwise, it equals 0. The core independent variable is the interaction term of TreatmentGroup_i and AfterTreatment_it, with its coefficient in the DID showing the influence of the GenAI adoption on solver popularity.

To assess Price_i, we calculated the average price set by solvers i in the pre-treatment period on the crowdsourcing platform and used the natural logarithm of the average price.

Following previous studies on solver popularity on crowdsourcing platforms and considering data availability, we controlled for several other variables that may affect our dependent variables. First, a solver’s level of knowledge impacts consumers’ perceptions (Huang et al., 2019). We thus used the solver’s language proficiency (Langprof_it), number of skills (Skillsnum_it), and number of certifications (Certification_it) as control variables. Second, customer information displayed by solvers on the platform influences their popularity, leading us to include the number of images in displayed works (Portfolio_it) and authoritative clients (Aclients_it) as control variables. Lastly, the solver’s last delivery time (Avgresday _it) and average response time (Lastdvday_it) are also essential control variables.

We further explored the heterogeneity in the solvers’ GenAI adoption behavior based on their duration of membership on the crowdsourcing platform (Dur_dum_i). Duration of membership is a significant indicator of their experience, which, to a certain extent, reflects their work capability. Generally, solvers who have been platform members for a longer duration have more experience and skills. We then used this duration to indicate the experience level of crowdsourcing platform solvers. To understand the specific heterogeneity of the impact of GenAI adoption on solver popularity, we categorized the solvers based on their experience levels, namely, high (i.e. those above the median duration) and low experience levels (i.e. those below the median duration).

There is no multicollinearity in the sample. We used variance inflation factors (VIF) to check the extent of multicollinearity in our model. The VIFs of all independent variables in our model range between 1.00 and 1.03 (<5), indicating that multicollinearity is not a concern. We then employed PSM to identify similar solvers during the pre-treatment stage to mitigate self-selection bias within the control group.

To reduce potential differences between a solver who has adopted GenAI and one who has never adopted GenAI, we used PSM to match the solvers in the treatment group to those in the control group based on observable characteristics (Durward et al., 2020). First, we calculated the propensity score of each solver. Second, based on the calculated score, we paired the solvers in the treatment group with those in the control group. PSM aims to ensure that the two groups are comparable or exhibit similar trends before the treatment. We calculated the propensity score using logit regression with the indicator of adopting GenAI as a dichotomous outcome and a set of observed characteristics as covariates, including price (Pre_Price), duration of platform membership (Duration), and the solver’s language proficiency (Langprof), number of skills (Skillsnum), number of certifications (Certification), number of displayed works (Portfolio), and number of authoritative clients (Aclients). These covariates are selected to account for observable characteristics that may influence a solver’s likelihood of adopting GenAI. For instance, the price of a solver’s gigs in the first month reflects their market positioning, which may influence their motivation to adopt GenAI either as a tool for cost efficiency or as a means to enhance perceived quality. Similarly, platform membership duration reflects experience and familiarity, which can affect GenAI adoption behavior. Following Nichols (2007), we performed one-to-one nearest neighbor matching to calculate the propensity scores for the two groups using the psmatch2 command in the software STATA. Appendix A presents the summary statistics of the treatment and control groups before and after matching. The standardized bias of the covariates is largely reduced after matching. The t-test results also show that the covariates between the treatment and control groups have a similar value without any significant differences after matching. We also used other matching algorithms as robustness checks.

Before proceeding to our model specifications, we grouped the data and compared the changes in the treatment and control groups to gain direct model-free evidence following Lin et al. (2023). To visualize the effect, we further divided each group into low-price (blue shaded bars) and high-price (red non-shaded bars) sub-groups. These categories are determined by a median split of the average price, classifying solvers into high-price (above the median) and low-price (below the median) groups. As shown in Appendix B, the bars for the Treatment Groups (Group 1 and Group 2) are substantially higher than those for their respective Control Groups. This observation provides preliminary, model-free evidence that the popularity (number of likes and comments) of the solvers is higher after GenAI adoption. Furthermore, for Group 1, the increment for high-priced solvers (red non-shaded bar) is slightly greater than for low-priced solvers (blue shaded bar) within the treatment condition, suggesting a potential boosting effect of price. Conversely, for Group 2, this price effect seems to be reversed for the treatment group. Next, we further apply formal econometric analysis to explore the research questions.

One pivotal assumption of the DID method is the common trend assumption, which requires the treatment and control groups to have parallel variation trends regarding solvers’ popularity before these solvers adopted GenAI (Angrist and Pischke, 2009). To preliminarily test this assumption, we plotted the treatment effects (and their 95% confidence intervals) in the pre- and post-treatment periods (see Appendix C). Appendix C shows that none of the treatment effects in the pre-treatment periods significantly differ from zero, and that significant trend differences only appeared after the treatment. Therefore, the treatment and control groups show similar trends in the absence of any treatment.

The DID regression estimation allowed us to leverage the advantages of panel data while controlling for specific time and solver effects. In this way, we can detect changes not only between solvers in the treatment and control groups but also in their popularity before and after GenAI adoption. The equation of our model is written as follows:

where Y_it serves as the dependent variable and refers to Like_it and Comment_it. If solver i adopted GenAI between August 2022 and September 2023, then TreatmentGroup_i equals 1 and is 0 otherwise. AfterTreatment_it = 1 indicates that solver i adopts GenAI at time t, Z_it represents the control variables, μ_i is the individual-level fixed effect, and λ_t is the time-period fixed effect. We included μ_i and λ_t here to control for the unobserved characteristics that affect all solvers but differ in time. To investigate the impact of price on solvers’ adoption of GenAI, we run regressions by including the price variable into an extended difference framework. The equation of our model is written as follows:

We also distinguished between Group 1 (i.e. solvers who price GenAI gigs higher than non-GenAI gigs) and Group 2 (i.e. solvers who price GenAI gigs lower than non-GenAI gigs), to capture heterogeneous effects across contrasting price contexts.

Table 2 presents the results of the main variables for PSM-DID. We initially tested the effect of GenAI adoption on the popularity of solvers in the first and second groups. (These results are based on our primary analytical sample, which retains only data related to their first GenAI gig; for a comparative analysis using the full sample, see Appendix D.) Results in Table 2 show that solvers’ adoption of GenAI positively affects their popularity. When a solver posts GenAI gigs, he/she will gain more popularity on the crowdsourcing platform, irrespective of his/her group (i.e. for Group 1 (Like_it: β₁ = 1.005, p < 0.01; Comment_it: β₁ = 0.896, p < 0.01) and Group 2 (Like_it: β₁ = 0.641, p < 0.01; Comment_it: β₁ = 0.406, p < 0.01)), thus supporting H1. The analysis above addresses RQ1.

We further examined how the price influences the impact of GenAI adoption. Results in Columns (1) and (2) of Table 2 show that the coefficient of the interaction term TreatmentGroup_i × AfterTreatment_it × Price_i is significantly positive for Group 1 (Like_it: β₂ = 0.104, p < 0.01; Comment_it: β₂ = 0.101, p < 0.01), thereby suggesting that price can strengthen the positive relationship between GenAI adoption and online popularity for solvers in Group 1, which supports H2a. Furthermore, the coefficient of the interaction term TreatmentGroup_i × AfterTreatment_it × Price_i presented in Columns (3) and (4) of Table 2 is significantly negative for Group 2 (Like_it: β₂ = - 0.228, p < 0.01; Comment_it: β₂ = - 0.155, p < 0.01), thereby suggesting that price weakens the positive effect of GenAI adoption on solvers’ online popularity in Group 2 (boomerang effect), which supports H2b. The analysis above addresses RQ2.

As we use the number of comments obtained by solvers as a proxy for their popularity, this measure may be subject to bias because it potentially includes both positive and negative feedback. We conducted additional analyses using ratings as the dependent variable to address this limitation. The results presented in Appendix E are consistent, further supporting the robustness of our findings.

To examine the heterogeneous effects based on a solver’s duration of membership on the platform, we introduced a duration dummy (Dur_dum, where 1 = long duration) and examined its interaction with the key variables. The results for the two groups are presented in Appendix F. For Group 1, the results in Table F1 show that the coefficient of the interaction term TreatmentGroup_i × AfterTreatment_it × Dur_dum_i is positively significant for both Score (β = 0.033, p < 0.01) and Likes (β = 0.079, p < 0.1), which suggests that the positive impact of adoption is of a greater magnitude for solvers with a longer membership duration.

Conversely, for Group 2, the results in Table F2 reveal the opposite effect. The coefficient for the interaction term TreatmentGroup_i × AfterTreatment_it × Dur_dum_i is negatively significant (e.g. for Comment_it: β = - 0.125, p < 0.05), which suggests that the benefit of GenAI adoption is significantly smaller for solvers with a longer duration of membership. Regarding the control variables, a review of our models indicates that most were not significant.

To further confirm that the increase in solver popularity is due to GenAI adoption instead of other unobservable factors, we conducted a placebo test following Cantoni et al. (2017) to check for any missing variables. We randomly selected a treatment group and repeated our sampling 500 times. The logic of this test is that if our main DID estimate for the true treatment effect captures a real phenomenon, then the coefficients from these 500 placebo regressions should, on average, be centered around zero. Results in Appendix G confirm this expectation. The distribution of the 500 estimated placebo coefficients is approximately normal and centered around zero.

The DID model has a critical parallel trend assumption that no pre-treatment trend exists between the treatment and control groups (Angrist and Pischke, 2009). Results in Appendix C show no sign of pre-treatment trends in our study, thereby further confirming the robustness of our findings. Results from the parallel trend assumption test rule out the potential influence caused by time-variant unobservable confounds. We also used a different PSM method to check the robustness of our model, which is the Caliper matching technique, where we matched each treatment solver to the control solvers and used the same covariates as the main results. Appendix H presents the results, which are similar to those presented in Appendix A, thereby further confirming the robustness of our findings.

Informed by signaling theory, this study investigates how the adoption of GenAI serves as a signal that influences solvers’ popularity on crowdsourcing platforms. We also analyze the data by comparing the relative prices of GenAI gigs and non-GenAI gigs. We conducted PSM on an unbalanced panel dataset covering August 2022 to September 2023 to address the potential endogeneity concerns raised by self-selection. We used DID to assess the changes in solvers’ popularity before and after adopting GenAI.

Our results show that solvers who adopt GenAI are more popular on the platform, regardless of their pricing strategy. We analyzed how price interacts with GenAI adoption as contextual signals to unpack the dynamics. The analysis reveals that price moderates the effect of GenAI adoption on solver popularity. Specifically, for solvers who charge GenAI gigs higher than non-GenAI gigs, the price signal enhances the positive effect of GenAI adoption on solvers’ popularity (boosting effect). Conversely, for solvers who set lower prices for GenAI gigs, the price signal weakens the positive effect of GenAI adoption (boomerang effect). Furthermore, our analysis of heterogeneous effects shows that for solvers who price GenAI gigs higher than non-GenAI gigs, the positive effect on adoption is significantly greater for solvers that have been on the platform longer. Conversely, for solvers who set lower prices for GenAI gigs, the positive effect on adoption is significantly lower for solvers that have been on the platform longer. Finally, it is noteworthy that most of our control variables were not significant.

This study contributes to both the GenAI adoption and signaling literature. First, our study advances the literature on the impact of GenAI by exploring its effects on creative service solvers within supply-driven crowdsourcing platforms. While prior research has examined the potential of GenAI in augmenting productivity and content creation, few studies provide empirical evidence on how GenAI adoption affects consumer engagement in the decentralized labor market (Boussioux et al., 2024; Stokel-Walker and Van Noorden, 2023). Consistent with recent work on GenAI-augmented creativity, we find that solvers who adopt GenAI experienced a rise in popularity, which is aligned with evidence that human–GenAI collaboration can match or exceed human-only creative output (Boussioux et al., 2024; Marshall et al., 2024). For example, Boussioux et al. (2024) proposed that human-GenAI solutions demonstrated superior strategic viability, financial and environmental value, and overall quality. Our work extends these findings by showing that GenAI adoption confers popularity benefits (higher likes/comments) on solvers in a decentralized labor market, demonstrating a new pathway through which GenAI complements creative labor. In sum, our study is one of the first to empirically examine the impact of GenAI adoption on supply-driven platforms.

Second, our study extends signaling theory by underscoring the dual signaling role of price which operates through the information effect and the sacrifice effect. Prior studies also showed that the impact of price as a signal is context-dependent (Hsiao et al., 2024; Jang and Chung, 2021). We introduce a contextualized view of signaling in the GenAI context, showing that the effect of price depends on the relative price between GenAI and non-GenAI gigs. When GenAI gigs are priced higher, price acts as a positive signal, thereby strengthening the effect of GenAI adoption (boosting effect). In contrast, when GenAI gigs are priced lower, a misalignment arises, leading consumers to question the quality or effort behind the GenAI service, thereby weakening the positive effect of adoption (boomerang effect). Our study extends signaling theory by bridging GenAI adoption and price signals.

Third, the results of our heterogeneity analysis enrich research on digital inequality. Specifically, we find that for solvers that charge higher prices for GenAI gigs compared to non-GenAI gigs, GenAI adoption benefits solvers that have been on the platform longer. This insight supports previous findings that digital tools often disproportionately benefit more experienced users (Hsieh et al., 2008). However, our results also suggest that for solvers that charge lower prices for GenAI gigs, new users receive greater benefits from adoption than experienced users. This novel finding challenges a unidimensional view of digital inequality and thus enriches the discussion on the differential impacts of technology adoption.

This study offers significant practical implications. First, we offer practical insights to platform managers regarding GenAI adoption. Considering the positive impact of GenAI adoption on solvers’ popularity, platforms could provide the necessary resources, training, or tools to support GenAI adoption. For example, designers with higher pre-existing skills tend to adapt by shifting to more complex tasks using GenAI tools (Demirci et al., 2025), so providing tutorials can accelerate skill growth. Empirical evidence showed platform-sponsored training reduces performance gaps (Niederhoffer et al., 2025). By empowering solvers to enhance their skills to remain competitive, GenAI adoption can benefit the entire contributor base rather than only a few experts.

Second, this study provides a dual-faceted strategic framework for individual solvers seeking to maximize their market attractiveness. Our findings show GenAI adoption increases solvers’ popularity, so solvers should proactively disclose their use. Moreover, our findings offer guidance for solvers regarding their price strategies and emphasize the benefits of strategically pricing GenAI gigs to reflect the added value of GenAI assistance. Our analysis indicates that when GenAI gigs are priced higher (or lower) than non-AI gigs, the solver’s average gig price affects the impact of GenAI adoption on popularity through the boosting (boomerang) effect. Therefore, the optimal strategy requires solvers to carefully align their disclosure and pricing to send a consistent and compelling message about their value to the market. Platforms can also assist by publishing price benchmarks or analytic tools so that solvers can align their price with consumer expectations and quality signals to avoid having the service offerings interpreted as AI slops (Niederhoffer et al., 2025).

Third, crowdsourcing platforms should integrate GenAI capabilities into their system architecture (e.g. via APIs or embedded toolkits) to facilitate responsible adoption by solvers. Policymakers are also encouraged to provide incentives and regulatory support to foster GenAI’s productive and ethical use in digital labor markets (Nah et al., 2023).

Despite providing valuable insights into the impact of GenAI on the dynamics of crowdsourcing platforms, our work has several limitations that open up new directions for future research. First, our investigation focuses on specific crowdsourcing platforms, thereby raising questions about the generalizability of our findings across different supply-driven platforms. The extent to which our results apply to other contexts where GenAI is employed remains an open question and will need further empirical investigations across various platforms.

Second, our study operationalizes solver popularity using engagement metrics, namely the number of likes and the number of comments received. We acknowledge a significant limitation in this approach, particularly concerning the comment count. Our method fails to capture the valence of the comments, treating all engagement equally. A negative comment, while increasing the total count, clearly does not reflect positive popularity and may, in fact, signal service issues or customer dissatisfaction. By not distinguishing between positive and negative feedback, our measurement of popularity may not fully capture a solver’s true market reputation. Future research could employ natural language processing (NLP) to conduct sentiment analysis on comment text, thereby creating a more nuanced dependent variable.

Third, although we explored the complex relationship between the gig price strategies employed by solvers and the consumers’ perceptions, we did not delve into the underlying motivations of solvers in setting prices for gigs with or without GenAI adoption. Understanding the psychological factors that guide these pricing decisions is warranted. Future research may examine this question to generate deeper insights into consumer psychology.

Finally, we did not investigate the long-term implications of GenAI adoption on supply-driven crowdsourcing platforms. As GenAI technologies evolve and become more integrated into creative processes, the nature of creative work in the marketplace may shift. Future research may explore these longitudinal effects with a particular focus on the evolution of skills and the role of human creativity in the GenAI era.

With the widespread adoption of GenAI, the landscape of creative work on supply-driven crowdsourcing platforms is undergoing a major transformation. While GenAI empowers solvers to create content with greater efficiency and quality, its adoption presents a crucial signaling dilemma. It remains unclear whether the signal of using GenAI boosts a solver’s popularity by showcasing technical prowess or diminishes it due to client concerns over originality and authenticity. Our study aims to address this ambiguity by empirically investigating the effect of this signal. Using a large-scale panel dataset on an international crowdsourcing platform, we examined how GenAI adoption can change the popularity of solvers on crowdsourcing platforms. Our findings enrich and expand the literature on GenAI and supply-driven crowdsourcing platforms and provide theoretical guidance for solvers to establish long-term customer relationships. First, we emphasize that GenAI adoption plays a significant role in increasing solver popularity. Second, we combined signaling theory with the dual role of price signals to highlight the differentiated effect of price on the relationship between GenAI adoption and solver popularity (boosting and boomerang effects), thus guiding solvers on how to post and price gigs on supply-driven crowdsourcing platforms.

The supplementary material for this article can be found online.