Adoption decision-making of AI digital humans for agricultural regional public brands in short videos

1 Introduction

The Plan for Accelerating the Construction of a Strong Agricultural Country (2024–2035) explicitly proposes to “cultivate several regional public brands with excellent quality and strong competitiveness” and emphasizes the role of agricultural digital transformation in brand building and market access. Agricultural regional public brands face inherent challenges in digital promotion. Low profitability limits large-scale marketing investment, while smallholders and non-professional operators often lack the resources and capabilities needed for effective promotion on platforms such as WeChat, Douyin, and Instagram. At the same time, agricultural product marketing is still affected by excessive intermediaries, uneven market access, and unstable supply chain conditions. These challenges make digital marketing tools increasingly important for improving brand communication and expanding market participation.

With the rapid development of digital information and artificial intelligence, AI digital humans (AI-DHs) have emerged as an efficient marketing tool for agricultural brand communication and short-video promotion (Farivar and Wang, 2022; Jiang et al., 2024; Shamim et al., 2024). Numerous agricultural regional public brands have already adopted AI-DHs for promotion, such as Tianmu Shanbao's “Xiaoshanbao”, Anhui Dingyuan's “Dingxingguan”, and Lanxian Potato's “Lan DouDou”. A notable example is Zunyi City, Guizhou Province, where the introduction of Baidu's e-commerce digital human live streaming helped “pink crystal honeydew”—a product previously constrained by limited sales channels and weak promotion—expand market exposure and improve short-term sales performance. Such cases suggest that AI-DHs may help agricultural regional public brands broaden digital communication channels, reduce some information barriers between producers and consumers, and create more opportunities for market participation (Lyu et al., 2025).

In short video sales, traditional human sales can establish emotional connections with consumers but face limitations such as high management costs, time-space constraints, and inconsistent information dissemination. AI-DHs present several advantages in agricultural brand promotion: compared with traditional human promoters, they can provide more stable content presentation, continuous service, and standardized information delivery, thereby reducing uncertainty in product communication (Miao et al., 2022; Hong and Pavlou, 2014; Arrow, 1978). For search-based agricultural products, consumers rely heavily on rational information and credible product signals (Jiménez and Mendoza, 2013). Therefore, AI-DHs may support agricultural regional public brands by presenting product attributes, origin information, and brand credibility in a more visual and interactive way.

However, AI-DH adoption also entails potential trade-offs. First, the high initial investment in technology development and maintenance may widen the digital divide between large brands and smallholder-led regional brands. Second, over-reliance on short video platforms may increase brands' exposure to platform fees, algorithmic changes, and traffic allocation rules. Third, the environmental footprint of AI content production may create new challenges for green digital transformation. These issues indicate that AI-DH adoption should be analyzed not only as a marketing decision, but also as a digital governance issue related to inclusive participation and responsible technology use.

Recent research in service sectors suggests that organizational conditions can influence whether AI-related technologies are effectively translated into practical performance (Wang and Zhang, 2025). In agricultural regional public brands, this issue may be more salient because many smallholder-oriented organizations have limited technological capacity, marketing resources, and operational experience. The practical value of AI-DHs also depends on brands' digital readiness, content production capacity, and ability to coordinate AI tools with agricultural brand communication strategies (Xi, 2025). These conditions may shape the implementation quality of AI-DH adoption and its subsequent communication outcomes.

Existing studies have examined the development and marketing value of AI-DHs, especially their effects on consumer perception, engagement, trust, and marketing performance in general commercial contexts. However, the adoption of AI-DHs by agricultural regional public brands remains less examined. In agricultural short-video promotion, AI-DH adoption is not only a matter of consumer response; it also requires regional brands to weigh production costs, consumer preference, platform recommendation, and peer behavior before choosing between AI-DH and real-human promotion. Existing technology adoption studies provide useful explanations of adoption behavior, yet they offer limited insight into how AI-DH adoption spreads among heterogeneous regional brands through platform exposure and inter-brand learning. To capture this process, this study constructs a complex network evolutionary game model that links AI-DH costs, real-human promotion costs, consumer preference, recommendation priority, and network size to adoption dynamics in short-video promotion. By bringing relative promotion costs, consumer acceptance, platform recommendation, and inter-brand learning into the same analytical framework, AI-DH research is extended from consumer-side responses to the adoption decisions of agricultural regional public brands.

2 Literature review

Agricultural regional public brands refer to collective agricultural product brands built on a specific region's ecological environment, historical background, and cultural resources, and are usually managed by relevant organizations and shared by multiple agricultural producers. As their value is strongly tied to regional identity and origin credibility, these brands depend on stable consumer recognition and trust to transform local agricultural resources into market value (Li et al., 2025). The development of such brands is still constrained by high promotion costs and the need to maintain consumer trust. Building a stable brand reputation requires quality certification, continuous communication, and sustained market exposure, which many smallholder-oriented brands cannot easily afford. Consumer trust therefore becomes central to the value formation of agricultural regional public brands; it is shaped by product quality, brand influence, operating entities, and consumers' awareness, evaluation, and behavioral intention toward the brand (Alamsyah and Febriani, 2020; Yongji and Yutong, 2016). In this context, short-video platforms provide an important channel for specialty agricultural product promotion, as platform traffic, visual presentation, and embedded advertising can increase exposure, attract consumer attention, and influence purchase decisions (Yin et al., 2024). Digital sales tools, hybrid sales structures, and virtual spokespersons may further improve communication efficiency and reduce repetitive promotional tasks in brand marketing (Guenzi and Habel, 2020; Lapoule and Colla, 2016; Singh et al., 2019; Thaichon et al., 2018). However, existing studies mainly explain the communication value of short-video platforms and digital sales tools after they are adopted. They offer less explanation of how agricultural regional public brands choose between AI-DH promotion and real-human promotion under cost pressure, trust-building demands, and platform competition.

In marketing contexts, digital humans have gradually evolved from virtual images and digital spokespersons to AI-driven agents capable of interactive communication and flexible content production. Early applications appeared mainly as virtual idols or brand endorsers in entertainment and fashion contexts, where visual novelty and symbolic appeal were central (Silva and Bonetti, 2021). With the rise of social media, virtual influencers have developed follower networks, interactive capabilities, and influence mechanisms similar to those of human influencers (Arsenyan and Mirowska, 2021; Schouten et al., 2021; Franke et al., 2023). More recently, AI-DHs have been used as anthropomorphic digital agents that support stable information delivery, personalized interaction, and adaptive marketing communication (Miao et al., 2022; Wang and Qiu, 2024). Digital human avatars can also be incorporated into marketing strategies to enhance user engagement and brand communication (Oliveira et al., 2025). For agricultural regional public brands, these functions are particularly relevant to short-video promotion, where product display needs to be combined with origin reputation, product trust, regional culture, and collective brand identity. Existing studies have also shown that consumer cognition and emotion play important roles in agricultural short-video live streaming (Zeng et al., 2023). AI-DHs may therefore provide a visual and interactive means of presenting agricultural product information and regional cultural symbols in short-video communication.

Building on agricultural technology adoption research, AI-DH adoption in short-video promotion can be viewed as a strategic technology-use decision shaped by expected returns, market demand, peer competition, and network interaction. Studies on agricultural technology adoption have emphasized that adoption decisions evolve through learning, adjustment, and expected benefits (Conley and Udry, 2010; Lambrecht et al., 2014). Adoption may also develop through observation, imitation, and diffusion among heterogeneous actors, which is consistent with the logic of innovation diffusion (Rogers et al., 2014). In complex systems, peer competition and network structure can further influence how strategies spread among decision-makers (Shi et al., 2021). For agricultural regional public brands, this diffusion process is closely related to consumer-side market responses, because agricultural products often involve information asymmetry and consumers cannot immediately evaluate product quality, origin credibility, or production processes. Short-video promotion can work as a signaling mechanism by transmitting visual and interactive information about product quality, regional characteristics, and brand credibility, thereby affecting consumer trust and purchase responses (Spence, 1978). This theoretical logic suggests that AI-DH adoption in agricultural short-video promotion is a cost-based technology-use judgment by brands, while its diffusion is shaped by consumer market signals and inter-brand observation.

Research on digital humans has mainly explained their communication effects after adoption, while studies on agricultural short-video marketing have focused more on exposure, consumer response, and marketing performance. Technology adoption research provides explanations for why actors adopt new technologies, but usually pays less attention to how adoption decisions evolve through platform exposure and inter-brand learning. These research streams therefore offer only a partial explanation of how agricultural regional public brands compare AI-DH promotion with real-human promotion when cost conditions, market response, and peer behavior change simultaneously. A complex network evolutionary game model is used here to capture this decision process by linking payoff comparison with platform-mediated interaction among brands. The model further examines how AI-DH costs, real-human promotion costs, consumer preference, recommendation priority, and network size influence the diffusion of AI-DH adoption.

3 An Evolutionary game model of AI digital humans adoption

This study adopts a simulation-based computational modeling approach to examine AI-DH adoption among agricultural regional public brands. It combines evolutionary game modeling with scale-free network simulation to describe how brands choose between AI-DH promotion and real-human promotion, learn from connected brands, and adjust their adoption strategies.

3.1 Problem description and underlying assumptions

In short-video promotion of agricultural regional public brands, brands may choose either AI-DHs or real humans as promotional agents. For these brands, choosing AI-DHs is not merely a replacement of promotional agents, but also a judgment about whether expected market returns can support the required cost input and operational adaptation. This choice also carries signaling implications, because different agents present product quality, origin credibility, and brand image to consumers in different ways. The decision is not made in a fully isolated environment. Agricultural regional public brands often operate within shared platform channels and regional marketing networks, as well as local competition and public brand governance systems. The promotional performance of one brand can be observed by other brands through platform exposure, market feedback, consumer interaction, and local demonstration effects. Under such conditions, brands may adjust their adoption decisions by comparing expected payoffs, observing peer strategies, and learning from successful cases.

The evolutionary game framework is therefore used to describe AI-DH adoption as a dynamic decision-making process under bounded rationality. The complex network setting captures unequal brand connections. Brands with stronger market visibility or platform support may influence more peers, while small and medium-sized brands tend to rely more on observed market feedback and gradual strategy adjustment. The Fermi updating rule is adopted to represent this probabilistic learning process, in which a brand with a lower payoff is more likely, but not certain, to adjust its strategy by referring to a neighboring brand with a higher payoff. Based on this logic, the model links brand strategy choice with cost input, market response, platform exposure, and short-video communication feedback. The two promotion strategies, AI-DH promotion and real-human promotion, represent alternative digital marketing choices faced by agricultural regional public brands. In the model, cost inputs capture the resource requirements of different promotion modes, consumer preference reflects market-side acceptance of AI-DH promotion, recommendation priority represents platform exposure conditions, and short-video interaction describes the additional communication value generated by exposure, likes, comments, and sharing. On this basis, the following assumptions are specified.

Assumption 1: enterprises and farmers in the region collaborate to cultivate an agricultural regional public brand with good quality and distinctive features. Two homogeneous agricultural regional public brands, brand i and brand j, participate in the evolutionary game. During repeated interactions, their strategy choices gradually evolve toward relatively stable states. In agricultural settings, adoption decisions are often shaped by information exchange, peer learning, and feedback from observed outcomes (Benyishay and Mobarak, 2019; Lasdun et al., 2025).

Assumption 2: both regional public brands i and j have two pure strategies. One is adopting AI-DHs for short video promotion of agricultural products; the other is adopting real humans for short video promotion. The probabilities corresponding to the two strategies for brand i are x and 1−x, and the probabilities corresponding to the two strategies for brand j are y and 1−y, xy∈[0, 1], both as a function of time t. AI-powered virtual influencers and digital human avatars have been increasingly used as marketing communication agents, offering brands an alternative to human-based promotion in digital marketing contexts (Allal-Chérif et al., 2024; Oliveira et al., 2025).

Assumption 3: brand decision-making behavior is influenced by short video promotion cost, online sales, and market acceptance. The price of agricultural products is P, regardless of production costs. The cost of AI-DHs short video promotion includes the cost of AI-DHs generation b and the cost of short video advertisement production c₁; the cost of real humans short video advertisement production is c₂. A recommendation priority parameter αis introduced for AI-DH-related content on short-video platforms, and a higher α indicates greater platform exposure (α∈[0, 1]). Let d₁ and d₂ denote the baseline sales volumes driven by AI-DH promotion and real-human promotion, respectively. The effect of recommendation priority is incorporated into the payoff structure through the exposure-related sales terms specified in the subsequent game model. In AI-based digital marketing, production and design inputs are closely related to advertising engagement, sales outcomes, cost savings, and productivity gains, while algorithmic recommendation mechanisms may further shape firms' strategic decisions and market returns (Kapoor and Kumar, 2025; Liu et al., 2025; Garcia et al., 2026).

Assumption 4: a is the market demand for agricultural regional public brand, θ is the degree of consumer preference for AI-DHs promotion, the larger θ is, the more consumers are willing to buy AI-DHs promotion products (θ∈[0, 1]). In AI-based marketing contexts, consumer responses to digital agents and AI-generated advertising content are not uniform. Differences in perceived attractiveness, psychological distance, emotional perception, social presence, and recommendation acceptance may shape consumers' preference for AI-DH promotion (Arango et al., 2023; Liu et al., 2025; Khodakarami et al., 2026).

Assumption 5: agricultural regional public brands may obtain additional brand communication benefits by adopting AI-DH promotion. Based on the characteristics of short-video promotion, interactive behaviors such as user likes, comments, and shares reflect the social potential of short-video content. The interaction effect is denoted by η, where η∈[0, 1] (Wang, 2020), and the social benefit is introduced as f′ = f(1+η), reflecting the feedback effect generated by short-video interaction. In social media marketing, likes, comments, and shares do more than reflect audience reactions; they also contribute to brand engagement, social sharing value, advertising effectiveness, and potential conversion outcomes (Simon and Tossan, 2018; Lin et al., 2018; Osei-Frimpong et al., 2022).

The model parameters and their meanings are summarized in Table 1.

Table 1. Summary of model parameters.

3.2 Modeling and model solving

Based on the above assumptions and parameter settings, the game matrix between agricultural regional public brands is obtained as shown in Table 2.

Table 2. Game matrix.

The revenues for brands i and j are described as follows: when brands i and j both choose to use AI-DHs promotion, the brand revenue is P(aθ+d₁(1+α+η))+f(1+η) and the promotion cost is T₁. When brands i and j both choose to use real humans for promotion, the brand revenue is P(a(1−θ)+d₂) and the promotion cost is T₂.

The expected benefits for brands i using AI-DHs for promotion are: The expected benefits for brands i using real humans promotion are: The expected benefits for brands j using AI-DHs for promotion are: The expected benefits for brands j using real humans promotion are: Based on the multi-round game, the replicator dynamic equations for agricultural regional public brands i and j are expressed as follows: For clarity, let A = P[aθ+d₁(1+α)]+f(1+η)−T₁−P(a(1−θ)+d₂)+T₂+Pηd₂) and B = P(ηd₁−αd₁−2ηd₂) where T₁ = c₁+b denotes the total cost of AI-DH promotion and T₂ = c₂ denotes the cost of real-human promotion. The replicator dynamic equations can therefore be rewritten as F(x) = x(1−x)(A+yB), F(y) = y(1−y)(A+xB).

Ei1=y(P(aθ+d1(1+α+η))+f(1+η)-T1) +(1-y)(P(aθ+d1(1+α)+ηd2)+f(1+η)-T1)

Ei2=y(P(a(1-θ)+d2+αd1+ηd2)-T2)+(1-y)(P(a(1-θ)+d2)-T2)

Ej1=x(P(aθ+d1(1+α+η))+f(1+η)-T1) +(1-x)(P(aθ+d1(1+α)+ηd2)+f(1+η)-T1)

Ej2=x(P(a(1-θ)+d2+αd1+ηd2)-T2)+(1-x)(P(a(1-θ)+d2)-T2)

F(x)=x(1-x)[P(aθ+d1(1+α))+f(1+η)-T1-P(a(1-θ)+d2)+T2+Pηd2)+y(P(ηd1-αd1-2ηd2))]

F(y)=y(1-y)[P(aθ+d1(1+α))+f(1+η)-T1-P(a(1-θ)+d2)+T2+Pηd2)+x(P(ηd1-αd1-2ηd2))]

Setting F(x) = 0 and F(y) = 0yields four boundary equilibrium points: E₁(0, 0), E₂(0, 1), E₃(1, 0), E₄(1, 1). In addition, when B≠0, the system may have an interior equilibrium point: E5(x,y)=(-,AB-AB). The interior equilibrium is feasible only when0<-AB<1. Therefore, the stability analysis should consider both the four boundary equilibria and the feasible interior equilibrium. When B = 0, no isolated interior equilibrium exists unless A = 0; in the degenerate case A = B = 0, every point in the strategy space is stationary.

The Jacobian matrix of the system is:

The local stability of each equilibrium can be determined from the eigenvalues of the Jacobian matrix. An equilibrium is asymptotically stable when all eigenvalues have negative real parts. The stability conditions of the equilibrium points are summarized in Table 3.

J=[(1-2x)(A+yB) x(1-x)By(1-y)B (1-2y)(A+xB)].

Table 3. Stability conditions of equilibrium points.

The stability conditions show that the long-run adoption outcome depends on the relative payoff between AI-DH promotion and real-human promotion. When A < 0, E₁(0, 0) is stable, indicating that both brands tend toward real-human promotion. When A+B>0, E₄(1, 1) is stable, indicating that both brands tend toward AI-DH promotion. Under the condition A>0 and A+B < 0, the asymmetric equilibria E₂(0, 1) and E₃(1, 0) are locally stable, implying that different brands may adopt different promotion strategies. When the feasible interior equilibrium E₅ exists, its two eigenvalues have opposite signs, indicating that it is a saddle point rather than an asymptotically stable equilibrium.

The formal model also provides a clear directional implication for the cost parameters. Since T₁ = c₁+b enters the relative payoff of AI-DH promotion negatively, an increase in either the AI-DH short-video production cost c₁ or the AI-DH generation cost b reduces the relative payoff of AI-DH promotion, all else being equal. By contrast, because T₂ = c₂ represents the cost of real-human promotion, an increase in c₂ strengthens the relative payoff advantage of AI-DH promotion. The network simulation should therefore preserve these payoff relationships.

The bilateral evolutionary game abstracts from differences in the interaction structure among agricultural regional public brands. In practice, brands may be connected through platform exposure, regional competition, information exchange, and marketing relationships, creating heterogeneous opportunities for inter-brand observation and interaction. To represent this networked decision environment, the payoff structure developed above is further embedded in a complex network. Each node represents an agricultural regional public brand occupying a particular network position, while network connections determine the scope of interaction and strategy observation among brands.

3.3 Linking the evolutionary game model with complex network simulation

The bilateral evolutionary game developed above provides the payoff structure for the complex network simulation. In the network setting, the same payoff relationships are applied to interactions between connected brands, while network topology determines the scope of interaction and the Fermi rule governs strategy adjustment. The parameter definitions of the formal model are retained throughout the simulation so that the theoretical model and network evolution follow the same decision logic.

3.3.1 Network adaptation of payoff functions: expansion from bilateral game to multilateral interaction

The 2 × 2 payoff matrix in Table 2 is retained as the basis for calculating brand payoffs in the network simulation. Each brand interacts with its connected neighbors, and the pairwise payoffs generated from these interactions are used to calculate the node-level payoff. Because nodes in a scale-free network may differ substantially in the number of connected neighbors, node payoffs are adjusted for degree differences before strategy comparison.

• Node Strategy and Neighbor Matching: In the complex network, each node corresponds to Brand i in the 2x2 game, and its network neighbor nodes correspond to Brand j. The strategy combination of node i and each neighbor node j directly maps to the four game scenarios in Table 2.
• Degree-Normalized Payoff Calculation: In a scale-free network, nodes differ in the number of connected neighbors. If pairwise payoffs are directly summed, highly connected nodes may obtain larger payoff values simply because they participate in more interactions, making payoff comparisons across nodes dependent on network degree. To reduce this degree-induced scale effect, the payoff of node i is calculated as the average payoff obtained from interactions with all of its directly connected neighbors: Ūi=1∣Ni∣∑j∈Niπ(Si,Sj), where N_*i* denotes the set of neighbors of node i, ∣N_*i*∣ denotes the number of its connected neighbors, and π(S_*i*, S_*j*) represents the pairwise payoff corresponding to the strategy combination (S_*i*, S_*j*) in Table 2. This treatment makes node payoffs comparable across different network positions while retaining the payoff structure of the formal evolutionary game.
• Consistency of Cost and Payoff Parameters: Core parameters in Table 1 retain the same definitions and directional effects in the network simulation as in the formal evolutionary game, ensuring consistency between the payoff model and the simulation.

3.3.2 Strategy update via fermi rule: payoff scaling and probabilistic strategy learning

As the core mechanism for strategy updating in the network, the Fermi rule converts payoff differences between neighboring brands into probabilistic strategy-learning behavior. Because the payoff parameters are expressed on a monetary scale, the degree-normalized node payoffs are rescaled using a common payoff scale before entering the Fermi function.

Specifically, the scaled payoff of node i is defined as

where S = Pa is used as the common payoff scale, corresponding to the baseline market-revenue magnitude in the formal payoff model. The same value of S is applied across all nodes, iterations, and parameter scenarios. This treatment controls the numerical scale of the payoff difference while preserving the relative magnitude of payoff changes across sensitivity analyses.

Ũi=UiS,

In each iteration, node i randomly selects one directly connected neighbor j. The probability that node i adopts the strategy of node j is defined as

When Ũ_*j*>Ũ_*i*, node i is more likely to adopt the strategy of node j, whereas a lower-payoff strategy can still be adopted with a smaller probability because of decision noise. The parameter K controls the sensitivity of strategy updating to payoff differences. A smaller K produces more payoff-sensitive updating, while a larger K increases decision noise. The benchmark value of K is set to 0.10. Additional analyses under alternative Fermi noise levels are reported in the Supplementary material.

P(Si←Sj)=11+exp[(Ũi-Ũj)/K].

3.3.3 Parameter consistency between the formal model and network simulation

The parameters used in the network simulation retain the definitions established in the formal model. For example, θ continues to represent consumer preference for AI-DH promotion, while α represents the recommendation weight assigned to AI-DH-related content on short-video platforms. Parameter values are assigned according to the model assumptions and available industry information. Homogeneous game agents in the formal model are extended to nodes occupying heterogeneous positions in the network. Node degree reflects differences in the number of inter-brand connections and determines the range of observable interactions available to each brand. Strategy adjustment remains payoff-dependent, while the scope of interaction is determined by network topology.

3.3.4 Connection of evolutionary processes: from replicator dynamics to network-level evolution

Replicator dynamics describe changes in the proportion of strategies at the group level, whereas the network simulation represents the evolutionary process through repeated interactions and strategy updates among individual brand nodes. Initial node strategies correspond to the initial strategy proportions in the evolutionary game. During each iteration, brands calculate payoffs from interactions with neighboring nodes and update their strategies according to the Fermi rule. Changes in the overall proportion of AI-DH adopters then emerge from these repeated node-level adjustments.

4 Complex network-based evolutionary algorithm for AI digital humans adoption

4.1 Network infrastructure

Agricultural regional public brands are embedded in a short-video promotion network shaped by platform exposure, regional competition, information sharing, and marketing interaction. Within this network, brands differ in the number of connections they maintain with other brands, creating heterogeneous opportunities for observation and strategic learning. Some brands occupy more highly connected positions and can interact with a larger set of peers, whereas others have more limited network reach. A scale-free network is therefore used to represent this uneven connectivity and to examine how AI-DH adoption evolves through inter-brand interaction and strategy adjustment. The model does not attempt to reproduce every feature of actual platform or brand relationships; instead, it captures the heterogeneous connection structure most relevant to the adoption mechanism examined in the payoff model.

The network of agricultural regional public brands is denoted by G = (V, E), where V is the set of brand nodes and E is the set of connections between brands. The network is represented by an adjacency matrix A = (a_*ij*). If a_*ij* = 1, a direct connection exists between brands i and j; if a_*ij* = 0, no direct connection exists. Each node represents an agricultural regional public brand, and connected nodes can interact and observe one another's strategies during the simulation.

The scale-free network is generated using a preferential attachment mechanism. The network starts with m₀ mutually connected nodes. At each subsequent step, one new node is added and connected to m existing nodes, where m ≤ m₀. The probability that the newly added node connects to an existing node i is proportional to the current degree of that node:

where k_*i* denotes the degree of node i. This preferential attachment mechanism gives nodes with higher degrees a greater probability of receiving new connections and thereby generates a heterogeneous network structure. The process continues until the network reaches the predetermined size N. The resulting network is undirected, with no self-loops or repeated edges. Figure 1 illustrates a scale-free network with N = 100 agricultural regional public brand nodes. Once the network is generated, its topology remains fixed during the evolutionary process, while node strategies and payoffs change over successive iterations.

Π(i)=ki∑j∈Vkj

4.2 Evolutionary rules

In the scale-free network, each agricultural regional public brand updates its promotional strategy through repeated interactions with neighboring brands. At initialization, each node is assigned one of the two pure strategies, AI-DH promotion or real-human promotion. During each iteration, brands interact with their directly connected neighbors, calculate their degree-normalized payoffs based on the payoff matrix, and divide these payoffs by the common payoff scale S = Pa before strategy comparison. Each node then randomly selects one neighboring node as a potential learning target and updates its strategy according to the Fermi rule.

The strategy-updating probability depends on the difference between the scaled payoffs of the focal brand and the selected neighboring brand. A neighboring strategy with a higher scaled payoff is more likely to be adopted, whereas a lower-payoff strategy may still be adopted with a smaller probability because of decision noise. The network topology remains unchanged throughout the evolutionary process, while node strategies and payoffs are updated iteratively.

At each iteration t, the proportion of nodes adopting AI-DH promotion is recorded as the AI-DH adoption rate:

where NtAI denotes the number of nodes adopting AI-DH promotion at iteration t, and N is the total number of brand nodes in the network. The evolution of R_*t* over the predetermined simulation horizon is used to characterize the diffusion dynamics of AI-DH adoption and to assess whether the adoption rate exhibits convergence or persistent variation over the simulation horizon.

Rt=NtAIN,

5 Simulation process and results

5.1 Simulation steps

Following the above simulation-based computational modeling design, the complex network evolutionary algorithm is implemented through the following steps.

Step 1: Construct the scale-free network and initialize node strategies. A scale-free network with Nagricultural regional public brand nodes is generated according to the preferential attachment mechanism described above. At the beginning of each simulation run, each node is independently assigned AI-DH promotion with probability p₀ = 0.5 and real-human promotion with probability 1−p₀ = 0.5. The remaining model parameters are initialized according to the benchmark settings.

Step 2: Calculate node payoffs. Each brand interacts with all of its directly connected neighbors, and the pairwise payoff for each interaction is determined according to the corresponding strategy combination in Table 2. The node payoff is calculated as the average payoff across all neighboring interactions to control for differences in node degree. Before strategy comparison, the degree-normalized payoff is divided by the common payoff scale S = Pa, which is held constant across nodes, iterations, and parameter scenarios.

Step 3: Update node strategies. Each node randomly selects one directly connected neighbor as a potential learning target. Based on the difference between their scaled payoffs, the focal node updates its strategy according to the Fermi rule. The updating process is probabilistic, allowing higher-payoff strategies to diffuse more readily while retaining decision noise. All nodes determine their next strategies using payoffs from the current iteration, and the strategy vector is updated synchronously after all node-level decisions have been completed.

Step 4: Record the AI-DH adoption rate. At the end of each iteration t, the proportion of nodes adopting AI-DH promotion is calculated as Rt=NtAI/N. The network topology remains fixed within each simulation run, while node strategies and payoffs evolve over successive iterations.

Step 5: Repeat the simulation and summarize stochastic outcomes. Steps 2–4 are repeated for T = 100 iterations. Because network generation, initial strategy assignment, and Fermi-based strategy updating contain stochastic components, 100 stochastic simulation runs are conducted for each parameter scenario using a reproducible sequence of random seeds. At each iteration, the mean AI-DH adoption rate across repeated runs is calculated, and the 2.5th and 97.5th percentiles are reported as the 95% simulation interval.

This repeated-simulation design reduces the influence of individual stochastic realizations and allows the reported adoption trajectories to reflect the central tendency and uncertainty of the evolutionary process.

5.2 Agricultural regional public brand AI digital humans adoption simulation results

The benchmark settings and corresponding adjustment ranges for the main model parameters are summarized in Table 4. Additional simulation settings are reported in Supplementary Table S1.

Table 4. Benchmark settings for the main model parameters.

Consumer preference for AI-DH promotion is not directly observable in the current agricultural short-video context. As a benchmark, θ is set to 0.38 with reference to industry survey evidence showing that approximately 38% of consumers reported a positive attitude toward AI-related advertising (Yahoo Publicis Media, 2024). This value represents a relatively cautious level of consumer acceptance in the baseline scenario rather than a precise empirical estimate of preference for agricultural AI-DH promotion. The sensitivity analysis therefore varies θ from 0.2 to 1.0 to examine whether the adoption dynamics change across different levels of consumer preference.

For the cost-related benchmark parameters, the AI-DH generation cost is set to b = 5, 000 RMB, the AI-DH short-video production cost to c₁ = 6, 000 RMB, and the real-human short-video production cost to c₂ = 30, 000 RMB based on available industry information. The remaining market and payoff parameters, together with their benchmark values and sensitivity ranges, are reported in Table 4. These cost values are used as scenario-based benchmark settings rather than precise estimates of actual operating costs for agricultural regional public brands.

To calibrate the sales-volume parameters, the sales volume generated by real-human short-video promotion is linked to the assumed advertising conversion range of 3%−7%. The midpoint of this range, 5%, is used for the benchmark scenario, giving d₂ = 0.05a = 500. Following the model assumption that AI-DH promotion has a stronger potential sales-driving effect, d₁ is set to three times d₂, resulting in d₁ = 1500. Additional scenarios using conversion rates of 3% and 7%, while maintaining the relationship d₁ = 3d₂, are reported in the Supplementary material.

5.2.1 The impact of short video production costs on AI digital humans adoption

To examine the effect of AI-DH short-video production cost on adoption, c₁ is varied from 2,000 to 10,000 RMB in increments of 2,000 RMB. As shown in Figure 2, higher production costs consistently reduce the AI-DH adoption rate during the evolutionary process. At iteration 20, the mean adoption rate decreases from 0.1504 when c₁ = 2, 000 RMB to 0.0521 when c₁ = 10, 000 RMB. The differences are most visible during the transitional stage, while under the benchmark parameter combination the adoption rate in all cost scenarios eventually approaches zero by the end of the simulation. The 95% simulation intervals reported in Supplementary Figure S1 are relatively wide in the early and middle stages, indicating stochastic variation across independent network realizations, but gradually narrow as the system approaches the low-adoption state. These results are consistent with the payoff specification in which c₁ enters the AI-DH payoff as a cost term: increasing the short-video production cost weakens the relative payoff of AI-DH promotion and accelerates the decline in its adoption.

The production cost of real-human short-video promotion, c₂, is then varied from 10,000 to 50,000 RMB. Figure 3 shows a clear positive relationship between c₂ and the relative diffusion of AI-DH promotion. At iteration 20, the mean AI-DH adoption rate increases from 0.0040 at c₂ = 10, 000 RMB to 0.5760 at c₂ = 50, 000 RMB. By iteration 100, the mean adoption rates for c₂ = 10, 000, 20,000, and 30,000 RMB have converged to approximately zero, whereas the corresponding values are 0.0384 for c₂ = 40, 000 RMB and 0.6626 for c₂ = 50, 000 RMB. The 95% simulation interval remains wide under the highest-cost scenario, indicating that different stochastic realizations may evolve toward different adoption outcomes near the transition region. Because c₂ reduces the payoff of real-human promotion, an increase in this cost improves the relative payoff position of AI-DH promotion and makes its diffusion more likely.

5.2.2 The impact of AI-DH generation cost on adoption

The AI-DH generation cost b is varied from 3,000 to 7,000 RMB to examine its influence on adoption dynamics. As shown in Figure 4, higher generation costs are associated with lower AI-DH adoption trajectories. At iteration 20, the mean adoption rate decreases from 0.1240 when b = 3, 000 RMB to 0.0733 when b = 7, 000 RMB. As the evolutionary process continues, the adoption rates under all five cost scenarios gradually approach a low level; by iteration 100, the mean adoption rates are approximately zero. The 95% simulation intervals are wider during the transitional stage and narrow as the system approaches the low-adoption state. This pattern is consistent with the payoff specification, as a higher generation cost bincreases the total AI-DH promotion cost T₁ and reduces the relative payoff of AI-DH promotion.

5.2.3 The impact of consumer preference on AI-DH adoption

To examine the effect of consumer preference on AI-DH adoption, θ is set to 0.2, 0.4, 0.6, 0.8, and 1.0. As shown in Figure 5, the evolutionary outcome changes substantially as consumer preference increases. When θ = 0.2, the AI-DH adoption rate declines rapidly and eventually approaches zero. At θ = 0.4, the mean adoption rate remains substantially higher and reaches 0.6626 at iteration 100, although the relatively wide 95% simulation interval indicates considerable variation across stochastic realizations. Under the tested settings of θ = 0.6, 0.8, and 1.0, AI-DH promotion develops a clear relative payoff advantage, and the mean adoption rate converges to approximately 1.

These results indicate that consumer preference plays an important role in determining the relative attractiveness of AI-DH promotion. At low preference levels, the market-side returns generated by AI-DH promotion are insufficient to sustain its diffusion. As consumer preference increases, the relative payoff improves and the network shifts toward higher AI-DH adoption. The pronounced change across the tested preference levels further shows that the evolutionary response to consumer preference is not uniform across the parameter range.

5.2.4 The impact of recommendation priority on AI-DH adoption

To examine the effect of recommendation priority on AI-DH adoption, α is set to 0.1, 0.3, 0.5, 0.7, and 0.9. As shown in Figure 6, higher recommendation priority is associated with a higher AI-DH adoption trajectory throughout the evolutionary process. At iteration 20, the mean adoption rate increases from 0.0048 when α = 0.1 to 0.3782 when α = 0.9. By iteration 100, the mean adoption rates under α = 0.1, 0.3, and 0.5 have approached zero, whereas the corresponding values remain at 0.0268 and 0.3059 for α = 0.7 and 0.9, respectively. The 95% simulation interval under the highest recommendation-priority scenario remains relatively wide, indicating substantial variation across stochastic network realizations.

These results suggest that recommendation priority can strengthen the relative attractiveness of AI-DH promotion and sustain its diffusion for a longer period. However, under the benchmark payoff conditions, a higher recommendation priority alone does not necessarily lead to complete AI-DH adoption. Its influence is reflected more clearly in the evolutionary trajectory and persistence of adoption than in a uniform shift toward full diffusion.

5.2.5 The impact of network size on AI-DH adoption

To examine whether network size affects AI-DH adoption dynamics, N is set to 100, 300, and 500, respectively. As shown in Figure 7, the mean adoption trajectories under the three network sizes are broadly similar. Under the benchmark payoff conditions, AI-DH adoption gradually declines across all three scenarios, and the mean adoption rates at iteration 100 are close to zero. Increasing network size therefore does not produce a systematic increase in the average level of AI-DH adoption.

The main difference appears in the stochastic variation of the evolutionary process. Compared with the smaller network, larger networks exhibit narrower simulation intervals and smoother mean trajectories, indicating that the aggregate adoption outcome becomes less sensitive to individual stochastic realizations as the number of interacting brands increases. Network size therefore affects the variability of the diffusion process more clearly than its long-run directional outcome under the benchmark parameter setting.

Overall, the simulation results show that AI-DH adoption is shaped primarily by the relative payoff between AI-DH promotion and real-human promotion. Higher AI-DH-related costs, including short-video production cost c₁ and generation cost b, weaken the relative payoff of AI-DH promotion and lead to lower adoption trajectories, whereas a higher real-human promotion cost c₂ improves the relative payoff position of AI-DH promotion. Consumer preference θ produces the most pronounced change in the evolutionary outcome: low preference leads the network toward a low-adoption state, while the higher tested preference levels shift the system toward widespread AI-DH adoption. Recommendation priority α also supports AI-DH diffusion by maintaining higher adoption levels over the evolutionary process, although its increase alone does not necessarily produce complete adoption under the benchmark conditions. By contrast, changes in network size have little systematic effect on the average long-run adoption level under the benchmark payoff setting, and their influence is reflected more clearly in the variability of the evolutionary process, with larger networks exhibiting smoother mean trajectories and narrower simulation intervals. AI-DH diffusion therefore depends more strongly on payoff-related market conditions than on network size itself, while stochastic network interactions mainly affect the transitional path and uncertainty of the evolutionary process. Additional analyses under alternative sales-conversion conditions, Fermi noise levels, and initial adoption shares are provided in the Supplementary material.

6 Conclusions and recommendations

6.1 Discussion

Existing studies on digital humans and virtual agents have mainly examined their communication effects after adoption, including consumer engagement, brand evaluation, and purchase response. In agricultural short-video promotion, however, the marketing value of AI-DHs first depends on whether agricultural regional public brands are willing and able to adopt them. This decision is shaped not only by consumer response, but also by the expected payoff of different promotion modes and the observable behavior of other brands within platform-mediated networks. The simulation results show that AI-DH adoption evolves through payoff comparison and inter-brand learning, rather than through a one-time technology choice made by isolated brands.

AI-DH adoption in the brand network does not arise from isolated decisions, but develops through repeated observation, comparison, and strategy adjustment among connected brands. A brand may retain or imitate a promotion strategy according to the payoff information available from its neighbors, allowing local strategic choices to accumulate into a broader diffusion pattern. When AI-DH promotion offers a relative payoff advantage, this strategy is more likely to persist and spread through such interactions; when the advantage disappears, the network can instead move toward real-human promotion. Platform recommendation enters this process by reshaping the payoff environment surrounding AI-DH promotion, whereas simply enlarging the network does not systematically raise its long-run adoption level. The network therefore matters less because of its size than because it provides the relational structure through which brands observe one another and continuously revise their promotional choices.

The cost-related results suggest that the economic attractiveness of AI-DH promotion depends on how its costs compare with those of real-human promotion. Higher AI-DH short-video production cost c₁ and generation cost breduce the relative payoff of AI-DH promotion and lead to lower adoption trajectories, whereas a higher real-human promotion cost c₂ improves the relative attractiveness of AI-DH promotion. This finding indicates that the economic value of AI-DHs should be evaluated comparatively rather than on the basis of their absolute cost alone. For agricultural regional public brands, the adoption decision depends on whether the expected market returns from AI-DH promotion are sufficient to offset its generation and content-production costs relative to real-human promotion. Cost advantage therefore represents an important condition shaping AI-DH adoption, rather than an inherent advantage of the technology itself.

Consumer preference changes the market environment in which brands evaluate AI-DH promotion. When preference for AI-DHs is low, the returns generated by this promotion mode are insufficient to sustain its diffusion, and adoption gradually moves toward a low level. As consumer preference increases, the payoff position of AI-DH promotion improves markedly, making the strategy more likely to persist and spread across the network. The pronounced differences observed across the tested preference levels show that the evolutionary response to consumer preference is not uniform. Changes in consumer acceptance can be associated with markedly different adoption outcomes as the payoff balance between the two promotion modes changes. This finding is consistent with research on virtual influencers and digital endorsers showing that consumer acceptance is an important basis for the market effectiveness of virtual agents (Franke et al., 2023; Volles et al., 2024). In the context of agricultural regional public brands, consumer preference therefore functions as a key demand-side condition that can reshape the relative attractiveness of AI-DH promotion.

Platform recommendation operates through a different channel by altering the visibility and payoff potential of AI-DH-related content. A higher recommendation priority gives such content greater exposure, which can strengthen the relative return of AI-DH promotion and help sustain its adoption over the evolutionary process. However, the simulation does not indicate that increasing recommendation priority alone is sufficient to drive the network toward complete AI-DH adoption under the benchmark conditions. Its role is therefore better understood as reinforcing the persistence of AI-DH strategies rather than independently determining the final evolutionary state. This interpretation is consistent with studies showing that platform traffic allocation and algorithmic recommendation influence consumer attention and purchase behavior in short-video environments (Yin et al., 2024; Zeng et al., 2023). For agricultural regional public brands, recommendation priority can improve the market visibility of AI-DH promotion, but its effect remains conditional on the broader payoff structure faced by brands.

Changing the size of the brand network does not substantially alter the long-run direction of AI-DH adoption within the tested range. The main difference lies in the evolutionary process: larger networks produce smoother mean trajectories and narrower simulation intervals, suggesting that network expansion mainly dampens stochastic fluctuations rather than creating stronger incentives for AI-DH adoption. Peer observation and learning may still shape how strategies spread among connected brands (Conley and Udry, 2010; Lambrecht et al., 2014), but a broader interaction network should not be interpreted as inherently more favorable to AI-DH adoption. For agricultural regional public brands, the network primarily provides the structure through which promotional strategies are observed and adjusted, while the eventual adoption level remains more closely tied to the payoff conditions of AI-DH and real-human promotion.

Emerging applications of AI-DHs in agricultural digital marketing provide a practical context for these evolutionary mechanisms. Cases such as “Xiaoshanbao” illustrate how digital humans are being used to present agricultural products, regional cultural elements, and brand images through short-video and livestreaming channels. In practice, the attractiveness of such applications is closely connected with production costs, platform exposure, and consumer response—factors that also enter the payoff environment considered in this study. From this perspective, AI-DH adoption by agricultural regional public brands can be understood as an ongoing strategic choice shaped by changing market returns and platform conditions, rather than as a one-off decision to introduce a new digital tool.

Beyond the adoption process itself, AI-DHs may also have broader implications for agricultural brand communication and digital participation. By supporting product presentation, origin-related storytelling, and repeated information delivery, AI-DHs may provide small and medium-sized agricultural brands with additional opportunities to reach consumers through short-video platforms. These potential benefits, however, should not be understood as direct outcomes of the present model. Whether smaller brands can translate AI-DH adoption into broader market participation depends on their digital capabilities, implementation resources, and access to platform support. Without sufficient training or shared digital services, the diffusion of AI-DHs may instead reinforce the advantages of resource-rich brands and widen existing participation gaps.

Much of the digital-human marketing literature begins after the technology has already been adopted, focusing on how digital agents influence consumer responses and brand communication. The present study shifts attention one step earlier by examining how agricultural regional public brands adjust their AI-DH promotion strategies over time. In the model, adoption evolves through comparisons between the payoffs of AI-DH and real-human promotion, while the network provides the structure through which brands observe neighboring strategies and revise their own choices. This perspective complements existing research on peer learning in agricultural technology adoption by placing such learning within a platform-mediated brand network rather than treating it only as local observation among producers or organizations. It also extends digital-human marketing research by treating AI-DH use as an evolving strategic choice rather than a fixed technological input. The contribution therefore lies in explaining how relative payoffs, platform conditions, and inter-brand learning jointly shape the adoption process.

6.2 Conclusions

To examine the adoption of AI digital humans (AI-DHs) in agricultural regional public brand promotion, this study constructs a complex network evolutionary game model with scale-free networks as the carrier and agricultural regional public brands as nodes occupying heterogeneous network positions. The model focuses on how AI-DH short-video production costs, AI-DH generation costs, real-human promotion costs, consumer preference, recommendation priority, and network size affect AI-DH adoption in short-video promotion. The main conclusions are as follows.

(1) The cost comparison between AI-DH and real-human promotion plays an important role in adoption decisions. Higher AI-DH short-video production cost c₁ and generation cost b weaken the relative payoff of AI-DH promotion and suppress its adoption, whereas a higher real-human promotion cost c₂ makes AI-DH promotion comparatively more attractive. This means that brands do not evaluate AI-DHs on the basis of their absolute cost alone; what matters is whether AI-DH promotion provides a more favorable economic return than the available human-promotion alternative.

(2) Consumer preference can substantially reshape the evolutionary outcome of AI-DH adoption. When preference for AI-DHs is low, AI-DH promotion fails to maintain a sufficient payoff advantage and adoption gradually declines toward a low level. As consumer preference increases, the relative payoff of AI-DH promotion improves and adoption rises markedly; under the higher preference settings examined, the network moves toward widespread adoption. The pronounced differences across the tested preference levels indicate that consumer acceptance can alter the direction of strategy evolution rather than merely changing the speed of diffusion.

(3) Recommendation priority mainly affects the persistence and transitional level of AI-DH adoption. Higher recommendation priority improves the payoff position of AI-DH promotion and helps maintain higher adoption levels during the evolutionary process. However, even under relatively high recommendation settings, widespread adoption does not necessarily emerge under the benchmark conditions. Platform recommendation therefore acts more as a reinforcing condition for AI-DH diffusion than as an independent determinant of the final adoption state.

(4) Network size has limited influence on the long-run level of AI-DH adoption within the tested range. Expanding the number of brand nodes does not systematically increase the final adoption rate, although larger networks produce smoother mean trajectories and narrower simulation intervals. The main role of network size is therefore reflected in the variability of the evolutionary process rather than in the direction of the final adoption outcome. This suggests that the spread of AI-DH strategies depends more on the payoff conditions faced by brands than on network expansion itself.

Several limitations should be noted. This study relies on a simulation-based evolutionary game and therefore does not estimate actual adoption probabilities or causal effects from observed brand behavior. Although the benchmark parameters are calibrated using theoretical assumptions and available industry information, some values remain scenario-based and may vary across platforms, product categories, and regional brand contexts. In addition, the model represents inter-brand interactions through a fixed scale-free network, whereas actual brand networks may evolve over time and exhibit different structural characteristics. Finally, the model focuses on AI-DH adoption decisions and does not directly evaluate downstream outcomes such as sales performance, rural income, digital inclusion, or environmental impacts. Future research can combine platform or brand-level data with simulation modeling, compare alternative network structures, and examine these broader consequences when appropriate empirical measures become available.

6.3 Recommendations

Based on the above findings, this study proposes the following recommendations for promoting the effective and sustainable application of AI-DHs in agricultural regional public brands.

(1) Reduce cost barriers and strengthen governance support for AI-DH adoption. Because AI-DH production and generation costs can weaken adoption when they erode the relative payoff of AI-DH promotion, policy support should focus on reducing avoidable implementation costs, particularly for small and medium-sized agricultural brands with limited digital resources. Governments and local agricultural authorities can provide technical training, shared AI-DH service platforms, and targeted financial support to lower initial adoption barriers. Short-video platforms should also improve the transparency of service fees, traffic allocation rules, and algorithmic changes so that brands can assess expected returns more clearly. At the same time, environmental standards for AI content production should be incorporated into relevant support measures to ensure that digital marketing innovation remains consistent with green development goals.

(2) Develop differentiated AI-DH strategies according to brand capabilities. Agricultural regional public brands should avoid adopting AI-DHs through a uniform model. Brands with stronger financial and technological capabilities can develop more mature AI-DH applications, establish stable content-production systems, and build brand-specific digital spokespersons. Small and medium-sized brands, by contrast, can adopt AI-DHs gradually through shared service platforms, cooperation with local e-commerce platforms, or participation in regional digital marketing programs. Such differentiated approaches can help brands control adoption costs and choose implementation paths that better match their available resources and expected returns.

(3) Strengthen credible and context-appropriate AI-DH content design. In agricultural regional public brand promotion, the practical use of AI-DHs should extend beyond visual representation to the clear communication of product information, origin characteristics, and brand identity. Brands can use AI-DHs to present production processes, quality certification, regional culture, and product-use scenarios in ways that are consistent with the characteristics of agricultural products. At the same time, excessive entertainment and highly standardized content should be avoided when they weaken the distinctiveness or credibility of regional brands. AI-DH content should therefore balance informational clarity with appropriate cultural and interactive expression so that technological adoption remains aligned with the communication needs of agricultural regional public brands.

(4) Strengthen inter-brand learning and resource-sharing mechanisms. Expanding the number of participating brands alone is unlikely to increase AI-DH adoption, but effective interaction among brands can still support information exchange and strategy adjustment. Agricultural regional public brands can therefore develop cooperative mechanisms with peer brands, short-video platforms, technology providers, and research institutions to share implementation experience, technical support, and market information. Regional AI-DH service centers or shared digital marketing platforms may further provide content resources, training, and technical services for brands with limited digital capabilities. The emphasis should be placed on improving the quality of inter-brand learning and access to shared resources rather than simply enlarging the scale of the network.

(5) Establish ongoing monitoring and evaluation mechanisms for AI-DH implementation. Adoption rate alone is insufficient for judging whether AI-DH promotion creates practical value for agricultural regional public brands. Brands and policymakers should also monitor implementation costs, content exposure, consumer response, and sales conversion, while considering longer-term issues such as platform dependence, unequal access to digital resources, and the environmental burden of AI content production. These indicators extend beyond the outcomes directly examined in the present model, but they are important for evaluating whether AI-DH applications remain economically viable and socially sustainable in practice. Continuous monitoring can also provide a basis for adjusting promotion strategies and refining future policy and platform governance.

AI-DH promotion can create new possibilities for agricultural regional public brands, but its practical viability depends on whether brands can achieve a favorable balance between expected returns, implementation costs, platform conditions, and their own digital capabilities.

Generative AI statement

The author(s) declared that Generative AI was not used in the creation of this manuscript.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher's note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

Supplementary material

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fsufs.2026.1879083/full#supplementary-material

Supplementary material Supplementary material at the source

Supplementary material Supplementary material at the source

Footnotes

Edited by: Binyao Ning, Shaoguan University, China

Reviewed by: Suci Sandi Wachyuni, Politeknik Sahid, Indonesia

Marcelo De Rezende Pinto, Pontifical Catholic University of Minas Gerais, Brazil