1. Introduction
As corporate social responsibility (CSR) becomes increasingly important, cause-related marketing (CRM) has gained popularity among businesses [
1]. CRM is an innovative strategy that links business activities with social causes, and it is defined as a “profit-driven philanthropic act” [
2]. This unique model of corporate giving is based on consumer purchases. Varadarajan and Menon describe CRM as “the process of formulating and implementing marketing activities that link a firm’s contributions to a designated social cause with consumer transactions of specific products or services, thereby achieving organizational and individual objectives” [
3]. By engaging in CRM, companies not only meet their social responsibilities [
4] and enhance their corporate image and reputation [
5] but also improve consumer attitudes [
6]. This creates a positive cycle of financial and social benefits. For consumers, buying products linked to CRM not only satisfies their material needs but also allows them to contribute to societal welfare through corporate donations. This behavior fulfills their altruistic motivations and reinforces their social values [
7]. With the continuous rise in consumers’ expectations regarding corporate social responsibility [
8,
9], CRM has evolved from a mere marketing tool into a critical nexus bridging business value and social value.
While traditional CRM has established its foundational value in connecting commercial and social objectives, the rapid development of artificial intelligence (AI) is fundamentally reshaping how these strategies are conceived and implemented. This technological revolution presents both unprecedented opportunities and novel challenges for CRM practitioners and researchers alike. Artificial intelligence-generated content (AIGC) technology has advanced rapidly in the digital era, gaining widespread adoption in commercial settings due to its exceptional content-generation capabilities and user-friendly features [
10,
11]. An increasing number of enterprises now utilize AI to automate content creation, providing novel tools and frameworks for implementing CRM strategies. This integration enables tighter alignment between consumer purchasing intentions and CRM initiatives through intelligent means. The convergence of AI and CRM has emerged as a prominent marketing trend, enabling companies to create more targeted, personalized, and cost-effective campaigns. AI’s capacity for rapid content generation allows organizations to develop CRM advertisements at scale while maintaining consistency in messaging and brand alignment. This technological integration appears to offer a pathway for companies to amplify their social impact while optimizing operational efficiency.
However, this technological advancement introduces novel complexities that challenge conventional CRM dynamics and raise critical questions about consumer perceptions. The integration of AI into CRM contexts, while promising operational benefits, simultaneously creates unprecedented challenges that remain largely unexplored in existing literature. When AI is applied to CRM activities—particularly in CRM advertisement creation—a critical question arises: Does consumers’ awareness of AI’s involvement alter their responses to CRM campaigns? This question becomes particularly complex given CRM’s dual identity, where advertisements simultaneously convey social welfare messages and brand promotion objectives. While AI enables rapid content generation, consumers may fear that companies could exploit this technology to produce false or misleading materials promoting their charitable initiatives [
12], thereby fostering skepticism toward CRM communications [
13].
Existing studies have explored consumer reactions to AI when it serves as a predictive tool via data mining [
14,
15] but have largely overlooked the unique impacts of generative AI in CRM contexts [
16]. Research on AI-generated advertisements primarily focuses on performance comparisons between AI- and human-created content [
17,
18], paying limited attention to differences in effectiveness within CRM-specific scenarios or to the impact of AI disclosure on consumers’ emotional and behavioral responses. Prior research demonstrates that consumers exhibit distinct attitudes and behaviors toward AI-generated content compared to human-created materials [
19,
20], correlating with cognitive biases, emotional responses, and AI aversion [
21,
22].
Amid global calls for the disclosure of AI-generated content, current legal frameworks establish general requirements for AI disclosure [
23]. Yet, industry practitioners remain divided regarding whether and how to disclose AI involvement to consumers. Mixed findings—from negative consumer reactions [
24] to neutral [
25] or even positive outcomes [
26]—highlight contextual dependencies. Although employing AI-generated advertisements in CRM may enhance operational efficiency in the advertising design and reduce production costs, it comes at the potential expense of reduced trust and less positive ad attitudes when disclosing the use of AI technology in ads [
27]. However, the existing literature inadequately addresses consumers’ unique sensitivity to the coexistence of altruistic and commercial motives in CRM contexts, creating a critical knowledge gap that hinders both theoretical understanding and practical implementation of AI-driven CRM strategies.
These unresolved conflicts underscore the urgent need for a systematic investigation. Current literature inadequately addresses how AI disclosure interacts with CRM’s dual-identity nature or how consumers reconcile algorithmic involvement in socially conscious messaging. Furthermore, the mechanisms driving attitude shifts—particularly the mediating role of skepticism and the moderating influence of AI aversion—remain unexamined in CRM contexts. Consequently, this study proposes three research questions:
- (1)
Does AI disclosure affect consumer purchase intention differently between CRM and non-CRM contexts?
- (2)
How does disclosing AI involvement in CRM impact consumers’ purchase intentions?
- (3)
What mechanistic role does AI aversion play in this process?
To address these questions, we propose a conceptual framework (see
Figure 1) grounded in the PKM. The PKM offers a critical perspective for understanding consumer reactions to AI disclosure, positing that consumers scrutinize persuaders’ motives when persuasive intent is perceived [
28]. In CRM contexts, AI disclosure may lead consumers to question the firm’s motives for using AI-generated content, activating persuasion knowledge [
29]. AI aversion amplifies this skepticism by highlighting the incongruity of AI’s role in CRM. We empirically test this framework through two experiments. First, we compare consumer responses to AI-generated versus human-designed CRM advertisements and examine whether such differences persist in non-CRM contexts. Second, we analyze the mechanisms underlying the impact of AI disclosure on purchase intentions in CRM scenarios, focusing on the mediating roles of consumer skepticism and advertisement attitudes, as well as the moderating effect of AI aversion. Our findings aim to elucidate the implications of AI disclosure in CRM advertisements and provide actionable recommendations for firms to balance AI transparency with consumer engagement in CRM initiatives.
4. Study 2
Study 2 aimed to examine the impact of AI disclosure on consumer purchase intentions in CRM contexts and explore the underlying mechanisms and boundary conditions of this effect. We sought to validate the chain mediation role of consumer skepticism and advertisement attitudes and test whether AI aversion moderates the influence of AI disclosure on purchase intentions.
4.1. Participants and Design
A single-factor, three-level (AI disclosure: disclosing AI-generated, disclosing human-created, no disclosure) between-subjects design was employed. For this study, 240 participants were recruited from Credamo, exceeding the minimum sample size of 159 calculated by G*Power 3.1 (effect size = 0.25, α = 0.05, power = 0.8, groups = 3). Each participant received a compensation of CNY 2 after completing all survey items. After excluding 20 invalid responses (due to failed attention-check questions or uniform responses across measures), 220 valid responses remained (50.9% female). Participants had a mean age of 32.59 years (SD = 9.64), with 77.3% holding a bachelor’s degree or higher and a median monthly income of CNY 5000–10,000 (see
Appendix A for details of sample characteristics).
Participants subsequently completed scales measuring consumer skepticism, advertisement attitudes, purchase intentions, and AI aversion.
4.2. Development of Stimulus Materials
To test the research hypotheses, we designed a stimulus scenario. The experimental stimulus material was bottled mineral water, and all advertisements were presented in a print ad format containing both text and images. To eliminate the potential influence of consumer brand familiarity on the experimental results, we selected a fictitious brand name, “Green Life,” for testing. This study selected the “Public Welfare Tree-Planting” initiative as the CRM context, using the slogan “Green Life Natural Mineral Water, sourced from deep glaciers, pure in every sip. For every bottle sold, Green Life donates 0.1 yuan to desert tree-planting” across all CRM advertisements. In the AI-disclosure group, a label stating, “This advertisement was generated by AI” was placed in the bottom-right corner of the ad; the human-disclosure group featured the label “This advertisement was created by humans,” while the no-disclosure group omitted any creator information. Detailed experimental materials are provided in
Appendix B Figure A2.
4.3. Procedure and Measures
The procedure and measures were the same as in Study 1. Participants were randomly assigned to one of three experimental conditions: the AI disclosure group, the no-disclosure group, or the human disclosure group. All three groups viewed the same advertisement. In the disclosure conditions, participants were informed that the ad was either AI-generated or human-created, while the no-disclosure group received no such information. After viewing the stimulus, participants were asked to report their consumer skepticism, advertisement attitudes, AI aversion, purchase intention, and other measures, including attention checks, manipulation checks, and demographic information (such as age, gender, and education level).
Consumer suspicion was measured using a four-item scale adapted from Do et al. (2012) [
73], with items such as “The disclosure of the ad’s image source makes me feel that the advertisement is deceptive”; “The disclosure of the ad’s image source makes me doubt the product quality”; “The disclosure of the ad’s image source makes me feel the advertisement contains misleading information and has an obvious persuasive intent”; and “The disclosure of the ad’s image source makes me suspicious of the ad’s marketing motives and feel that the ad is untrustworthy” (Cronbach’s α = 0.871). Attitude toward the advertisement was measured using three items adapted from Baek et al. (2024) and Song et al. (2024) [
29,
72]: “I would consider accepting this advertisement”; “I think this advertisement is good”; “My attitude toward this advertisement is positive” (Cronbach’s α = 0.904). AI aversion was measured using a four-item scale adapted from Grassini (2023) [
70]: “I believe AI will not improve my quality of life”; “I believe AI will not enhance my work efficiency”; “I do not think I will use AI technology in the future”; “I believe AI technology has no positive significance for humanity” (Cronbach’s α = 0.919). Consumer purchase intention (Cronbach’s α = 0.844) was measured using the same items as in Study 1. All items were measured using a seven-point Likert scale (1 = strongly disagree, 7 = strongly agree).
For manipulation checks, we tested the effectiveness of the disclosure labels by asking participants, “Did the advertisement you just viewed mention the creator of its content?” (1 = Yes, explicitly stating it was created by a human; 2 = Yes, explicitly stating it was generated by AI; 3 = No, no creator information was mentioned; 4 = Not sure/did not notice).
4.4. Results and Discussion
4.4.1. Convergent and Divergent Validity of the Measurements
To assess the reliability, convergent validity, and discriminant validity of the constructs and their scales, a confirmatory factor analysis (CFA) was conducted. Results indicated good model fit (CMIN/DF = 1.12, RMSEA = 0.03, IFI = 0.993, TLI = 0.992, CFI = 0.993). All factor loadings exceeded 0.5 [
74]. The average variance extracted (AVE) values surpassed the minimum threshold of 0.5, and composite reliability (CR) values were well above 0.7, collectively indicating good convergent validity and composite reliability across all dimensions [
75] (see
Table 3).
4.4.2. Manipulation Check
The validity of experimental conditions was verified using Fisher’s exact test. Participants accurately identified the three disclosure contexts—human disclosure, AI disclosure, and none disclosure—with correct identification rates of 93.3%, 91.8%, and 90.3%, respectively. Despite low expected cell frequencies in some categories, Monte Carlo-simulated Fisher’s exact test results (p < 0.001) confirmed the reliability of the experimental manipulations, indicating successful manipulation.
4.4.3. Hypotheses Check
Main Effects. A univariate analysis with purchase intention as the dependent variable and AI disclosure as the independent variable revealed significant differences across AI disclosure conditions (F(2, 217) = 43.167,
p < 0.001, η
2 = 0.285; see
Figure 3). Consumers in the AI-disclosure group reported significantly lower purchase intentions (MAI = 4.11, SDAI = 1.1) compared to both the human-disclosure group (Mhuman = 5.20, SDhuman = 0.93) and the none-disclosure group (M = 5.12, SD = 0.64). No significant difference emerged between the human-disclosure and none-disclosure groups (
p = 0.73; see
Table 4). Thus, AI disclosure negatively impacts purchase intentions. These effects remained robust after controlling for demographic variables (gender, age, education), reconfirming H1.
Chain Mediation Analysis of Consumer Skepticism and Advertisement Attitudes. Based on the main effects, we conducted a mediation effect analysis focusing on the AI-disclosure and no-disclosure groups. Using Model 6 in SPSS 24.0’s PROCESS 4.1 plugin [
76] with 5000 bootstrapping samples, we tested the chain mediation effect with AI disclosure as the independent variable, consumer skepticism and advertisement attitudes as sequential mediators, and purchase intention as the dependent variable. Results (see
Table 5 and
Table 6) showed that after introducing mediators, AI disclosure no longer had a significant direct effect on purchase intentions (b = −0.022, SE = 0.041, t = −0.529, 95% CI [−0.104, 0.060]). However, the sequential mediation pathway (AI disclosure → consumer skepticism → advertisement attitudes → purchase intentions) was significant (Ind3 effect = −0.1469, SE = 0.0475, 95% CI [−0.2426, −0.0559]), with specific path coefficients illustrated in
Figure 4. These results remained consistent after controlling for demographics (gender, age, education), confirming H3.
Interaction Effect of AI Aversion and AI Disclosure. To mitigate multicollinearity, we mean-centered AI aversion and constructed an interaction term (AI disclosure × AI aversion). A moderated regression analysis was then conducted to examine the effects of AI aversion, AI disclosure, and their interaction on consumer skepticism. As shown in
Table 7, the interaction term exerted a significant positive effect on consumer skepticism (b = 0.131, t = 2.905, 95% CI [0.042, 0.220]), indicating that AI aversion strengthens the impact of AI disclosure on skepticism. Thus, H4 was supported.
Moderated Chain Mediation Analysis. A moderated chain mediation analysis was conducted using PROCESS Model 83 with 5000 bootstrapping samples. AI disclosure served as the independent variable, consumer skepticism and advertisement attitudes as sequential mediators, AI aversion as the moderator, and purchase intention as the dependent variable. Results revealed a significant moderated chain mediation effect (effect = −0.647, SE = 0.032, 95% CI [−0.131, −0.005]). Specifically (see
Table 8), under high AI aversion (+1 SD), the chain mediation effect of consumer skepticism and advertisement attitudes was significant (indirect effect = −0.135, SE = 0.044, 95% CI [−0.222, −0.052]), whereas under low AI aversion (−1 SD), the effect was nonsignificant (indirect effect = 0.002, SE = 0.04, 95% CI [−0.076, 0.085]). Path coefficients are detailed in
Figure 5. These results remained robust after controlling for demographic variables (gender, age, and education level). Thus, H5 was supported.
Study 2 revealed the mechanisms and boundary conditions of AI disclosure’s impact on purchase intentions in CRM contexts. The findings demonstrate that consumer skepticism and advertisement attitudes sequentially mediate the effect of AI disclosure on purchase intentions. Furthermore, AI aversion negatively moderates this relationship, strengthening the detrimental influence of AI disclosure.
5. General Discussion
Through two sequential experiments, this study systematically investigated the impact and mechanisms of AI-generated advertisement disclosure on consumer behavioral intentions in CRM contexts. Study 1 employed a 2 × 2 between-subjects design. By comparing CRM and non-CRM scenarios, it revealed that while consumers exhibited higher overall purchase intentions in CRM contexts, the negative effect of AI disclosure was significant only in CRM contexts, highlighting the highly context-dependent nature of AI disclosure effects. Study 2 focused on CRM contexts using a single-factor, three-level design. It further validated the unique inhibitory effect of AI disclosure on consumer behavioral intentions in CRM scenarios, uncovering the chain mediation role of consumer skepticism and advertisement attitudes. Additionally, Study 2 proposed a dual-stage moderation model of AI aversion: AI aversion strengthens AI disclosure’s triggering of consumer skepticism and intensifies the negative transmission from skepticism to behavioral intentions via cognitive-affective pathways. Collectively, the findings demonstrate that when marketing activities involve moral value transmission, the instrumental rationality of AI technology conflicts with the ethical attributes of CRM, activating consumers’ persuasion knowledge defense mechanisms.
5.1. Theoretical Contributions
First, this study extends the existing literature on CRM. Current CRM research predominantly focuses on analyzing CRM campaigns themselves, such as the impact of consumers’ familiarity with charitable initiatives [
77], donation amounts [
78], and cause-brand fit [
79,
80] on CRM effectiveness. Few studies have explored the role of AI technology in CRM contexts. By integrating the technical attributes of generative AI into the activation framework of the PKM, our findings reveal a cognitive process distinct from traditional PKM research centered on advertiser intent inference [
12]: AI disclosure in CRM triggers consumer skepticism, which negatively influences advertisement attitudes and subsequently suppresses purchase intentions. Crucially, we emphasize that AI disclosure in morally charged CRM advertisements activates consumers’ “moral hypocrisy attribution,” intensifying skepticism toward CRM campaigns and deepening the theoretical explanatory power of the PKM in CRM contexts.
Second, our research significantly contributes to the literature on consumer responses to AIGC. While technological advancements have expanded AIGC’s marketing applications [
19], studies examining consumer attitudes toward AI-generated advertisements remain limited. Although Gu et al. (2024) [
17] investigated consumer acceptance of AI-generated ads from a content effectiveness perspective, they overlooked AI’s technological instrumentality. By examining AI disclosure in high-morality CRM contexts, this study uncovers the negative impact of AI’s instrumental attributes on purchase intentions. The finding contrasts with prior work suggesting AI disclosure either does not affect brand attitudes [
25] or enhances consumer engagement through increased transparency [
81]. Responding to Baek et al.’s (2024) [
29] call, we compare AI disclosure effects across CRM and non-CRM contexts, revealing consumers’ unique sensitivity to AI involvement in CRM. This study offers fresh insights into how consumers process AI-disclosed advertisements, highlighting the context-dependent and complex nature of AI disclosure effects.
Finally, grounded in consumers’ skepticism toward AI technology [
82], we find that AI disclosure labels negatively affect consumer responses to AI-generated advertisements. This conclusion aligns with Arango et al. (2023) [
24], who observed reduced donation intentions for AI-generated charity ads. However, our research extends this by incorporating the role of consumers’ subjective attitudes toward AI [
83] and empirically validating the moderating effect of AI aversion. We demonstrate that high AI aversion not only directly intensifies the initial processing of disclosed information—thereby strengthening consumer skepticism—but also indirectly exacerbates resistance to the purchase intention of AI-generated CRM advertisements through cognitive-affective transmission pathways. These findings suggest that the negative effects of AI disclosure seem to be driven by humans’ subjective attitudes toward AI, offering novel theoretical insights into understanding the mechanisms underlying the negative impact of AI disclosure and expanding our understanding of consumer responses to AI-involved advertisement creation in CRM contexts.
5.2. Managerial Implications
Beyond theoretical contributions, this study offers practical insights for marketing practitioners, particularly firms and advertisers employing AIGC in CRM. While AIGC provides cost reduction and design efficiency benefits [
34], firms must balance these advantages against potential negative consumer responses [
24]. Advertisers should strategically align AIGC’s efficiency gains with advertising effectiveness, adapting disclosure practices to avoid over-transparency that triggers adverse reactions.
This study found that disclosing AI’s involvement in CRM advertisement creation significantly reduces consumers’ purchase intentions. Given the increasing legal and regulatory requirements for AI-generated content disclosure [
23,
25], firms and advertisers need to find a balance between marketing effectiveness and transparency. First, strengthening consumer education on AI through awareness campaigns that explain the role of AI in CRM can help mitigate consumer aversion to AI. Second, given the significant disparity in consumer purchase intentions between disclosing human involvement versus AI involvement, it is advisable to adopt a compromise approach by gradually disclosing AI’s role (e.g., initially emphasizing human participation or selectively disclosing AI’s use in non-critical design elements) [
26,
27,
54]. This phased disclosure may allow consumers to adapt to AI’s evolving role, thereby reducing the negative impact of AI disclosure on purchase intentions. Additionally, firms can refine consumer profiles based on AI aversion levels and develop personalized disclosure strategies to alleviate negative impacts and improve advertisement acceptance and effectiveness.
Furthermore, these findings provide implications for policymakers and legislators. Mandatory disclosure of AIGC in advertisements may negatively affect performance. Our research demonstrates that in CRM contexts, disclosing AI-generated ads heightens consumer skepticism, diminishing ad appeal and efficacy. Thus, policymakers should incorporate this potential adverse effect into regulatory design, advocating for flexible disclosure frameworks to avoid overregulation and maintain an equilibrium between transparency and advertising effectiveness. Especially in CRM scenarios, we recommend establishing tiered disclosure systems and exemption clauses based on advertisements’ philanthropic attributes to ensure both regulatory transparency and the practical impact of CRM campaigns.
5.3. Limitations and Future Directions
While this study contributes to theory and managerial practice, several limitations warrant attention and offer avenues for future research. First, the current research focused solely on text-image advertisements for CRM, though CRM campaigns employ diverse formats (e.g., video, audio, and interactive ads) [
84]. Future research could employ eye-tracking experiments to observe consumers’ visual attention toward AI-generated content or utilize electroencephalography (EEG) to capture users’ neural responses during consumer–AI content interactions. Such investigations would systematically compare how AI involvement affects consumer responses across these formats, thereby contrasting AI’s impact in CRM campaigns under different advertising modalities. Second, while we examined the presence or absence of AI disclosure, we did not distinguish whether advertisements were solely AI-generated, AI–human collaborative creations, or exclusively human-generated [
19]. Future studies could examine the degree of human–AI collaboration in CRM, specifically investigating how disclosing varying degrees of AI involvement in CRM advertisements (e.g., “AI generation + human review” or “human-led + AI optimization”) influences consumer behavior. This will help to deepen both theoretical and practical insights. Finally, this study employed scenario-based experimental methods with simulated advertisements, capturing only consumers’ immediate reactions. The long-term effects of AI disclosure on behavior remain unexplored. Future research could collaborate with firms engaged in CRM campaigns to collect longitudinal tracking data in real-world settings, observe how AI disclosure in genuine good cause marketing scenarios affects consumer purchasing behavior, and analyze whether consumer responses to AI disclosure evolve with technological advancements or information dissemination over extended periods.