Unveiling the Impact of AI Technological Anxiety on the Marketers' Intention to Adopt Generative AI

You’re listening to “Unveiling the Impact of AI Technological Anxiety on the Marketers' Intention to Adopt Generative AI,” by M. Tao and colleagues. Published in 2025.

Abstract.

With the rapid development of Generative AI technology, businesses need their marketers to adopt it to assist in completing their job tasks. However, marketers are not entirely optimistic and may feel anxious and concerned about it. This research explores how AI anxiety affects marketers' AI-generated content intention. Based on social cognitive theory, this study empirically examines the impact of anxiety on marketers' intention to use AI-generated content, including the mediating role of AI trust and the moderating effect of AI self-efficacy. Using survey data from 495 marketers, we found that AI privacy, bias, and opacity anxiety negatively correlate with AI trust and weaken marketers' intention to use AI-generated content.

Furthermore, this relationship is moderated by AI self-efficacy, where the negative relationship is significantly mitigated when marketers possess AI self-efficacy. These findings enrich the research on Generative AI in marketing and help organizations focus on marketers' anxious resistance to adopting Generative AI and reduce the adverse effects of AI anxiety.

INTRODUCTION.

With the continuous development of digital technology, artificial intelligence (AI) has become a critical factor in driving marketing innovation, business model innovation, and digital transformation. From intelligent production and automated supply processes at the production end to capturing user data at the marketing end, customizing products and services, and enhancing consumer experience

Volume 33 • Issue 1 • January-December 2025

, AI is playing an increasingly important role. Academic research also indicates that AI is the most important technology in future marketing, not only improving the work efficiency and effectiveness of enterprises and practitioners but also playing a crucial role in the restructuring of business models and enhancing market competitiveness. AI-driven innovation and digital shifts offer businesses both novel chances and challenges. Stanford University's Human-Centered Artificial Intelligence Institute points out in the Artificial Intelligence Index Report 2024 that AI is having an unprecedented impact on society, especially with generative AI receiving significant attention and investment. The report indicates that compared to 2022, the investment in generative AI surged nearly eight times in 2023 (from $3 billion to $25.2 billion).

Generative AI is characterized by its ability to produce creative content that includes text, images, audio, and video, such as ChatGPT and Sora. Unlike traditional AI technologies such as discriminative AI, generative AI leverages deep machine learning and generative adversarial networks, requiring users to be highly involved in the content generation process. Generative AI is a double-edged sword for marketing departments and marketers. On the one hand, generative AI can help companies efficiently screen innovative solutions, significantly expanding the creative boundaries for marketers and thereby enhancing the competitive advantage of enterprises.

On the other hand, the high involvement of marketers in the generative AI creation process can lead to increased role ambiguity, and there are concerns about privacy breaches, biased decisions, algorithm transparency, and accountability related to generative AI. Therefore, marketers might not have an entirely positive attitude and perception toward using generative AI. Their concerns about AI technology when using generative AI can significantly affect their trust in AI and their intention to use AI-generated content (AIGC), which is detrimental to the company's digital and AI strategy.

Previous studies have indicated that people can experience anxiety when using computer technologies. In recent years, scholars have begun to focus on the anxiety felt by humans when interacting with and using AI, categorizing the dimensions of AI anxiety and exploring its negative effects. Although these studies have enriched the literature on AI anxiety, there are still deficiencies. The current literature overlooks the impact of generative AI anxiety on the acceptance intentions of marketers, a distinct group, which is significant. Unlike ordinary users and consumers, marketers first adopt generative AI with different purposes and concerns. They generate advertisements, promotional videos, marketing plans, etc., which have to be fully oriented toward consumers and client groups rather than just personal use.

This means that the generated content must not carry any bias or discrimination, as this could bring unforeseeable disasters to the company. Second, unlike traditional AI technology, the use of generative AI requires a high degree of marketer involvement, necessitating constant iteration and updating of their proposals, and the results generated are closely related to the participants themselves. Khasawneh (2018) found that during this process, their concerns about the leakage of personal and corporate marketing strategy information increased. Additionally, due to the more complex underlying logic and the opacity of the algorithms, marketers have to worry about potential errors, which could lead to trust issues, further affecting acceptance intention.

It is evident that although generative AI is emerging and rapidly developing in marketing work, concerns about AI privacy, bias, transparency, and other technical anxieties among marketers may make its implementation in corporate practice challenging. Understanding and alleviating AI anxiety has become an important issue, and this is precisely what this study focuses on addressing. Therefore, this study aims to address how marketers' anxiety about generative AI affects their AIGC intention, how trust in AI mediates this process, and how AI self-efficacy moderates it. The expected theoretical contributions of this paper are as follows: first, we have extended research on AI marketing by identifying AI anxiety among marketers and studying the mechanisms of its impact on AIGC intention.

Second, we have expanded the research scope of social cognitive theory and the research field and scope of AI anxiety, which has previously focused on general user groups. By targeting marketers, a more purposeful group, we have deepened our understanding of AI anxiety in specific groups. Third, our research contributes to the understanding of AI trust and AI self-efficacy, with a special focus on the mediating role of trust in AI adoption and the moderating role of AI self-efficacy. This deepens our understanding of AIGC in the field of marketing. Last, by combining AI technology with the work of marketers, we have also provided an interdisciplinary research approach to AI marketing. Based upon that, we are proposing the following research questions:

• How does AI technology anxiety, including AI privacy anxiety, AI bias anxiety, and AI opacity anxiety, affect marketers' AIGC intention?

• How does AI trust play a mediating role and how does AI self-efficacy play a moderating role?

LITERATURE REVIEW AND HYPOTHESIS DEVELOPMENT

AI Anxiety, AI Trust, and Marketers' AIGC Intention

Anxiety, as a psychological state and trait disposition, reflects an individual's cognitive assessment and emotional response to threats and risks. Beck (1976) emphasizes the critical role of cognitive patterns in anxiety, noting that anxious individuals tend to perceive situations as threatening. In organizational behavior, workplace anxiety is defined as the emotional response of tension, unease, and stress toward job performance. In the realm of information technology, anxiety is used to describe the emotional experiences of individuals as they adopt and learn new technologies. For instance, computer anxiety describes the discomfort, worry, or fear experienced by individuals when using computers, which may stem from unfamiliarity with the technology, learning difficulties, and concerns about privacy breaches.

With the rise of AI technology, there has been increased attention on AI-related anxiety. AI technology, especially generative AI, is more complex than computer technology, leading to differences in AI anxiety compared to computer anxiety.

First, computer anxiety is more about fears and concerns about technology use. In contrast, AI anxiety reflects fears about the rapid development of AI, the potential loss of control, and concerns about its possible negative impacts. Second, AI involves more algorithmic mechanisms and opaque content generation, leading to concerns not only about general technological anxieties similar to computer anxiety but also about the potential decision-making processes of AI, privacy handling, and a series of social and ethical issues raised by AI. In recent years, scholars have attempted to categorize dimensions of AI anxiety and conduct empirical research. Most studies focus on anxiety's relationship to AI privacy risks, learning, and job replacement.

However, this paper examines the context of marketers using AIGC, which differs from previous studies, as marketers using AIGC are accountable to their companies and must not disclose trade secrets. Additionally, marketers using generative AI interact with consumers or clients rather than using it for personal purposes and thus have a stronger orientation toward interests and objectives. Therefore, the internal decision-making mechanisms and transparency of AI become critical considerations. Consequently, this paper examines the impact of AI privacy anxiety, AI bias anxiety, and AI opacity anxiety on marketers’ AI trust and AIGC intention.

AI Privacy Anxiety, AI Trust, and AIGC Intention

Social cognitive theory suggests that outcome expectancies can affect an individual's psychological state and choice of behavior. In the marketing context, marketers' efforts are closely linked to enterprises' business secrets and marketing strategies, such as launching new products, setting appropriate prices, and market strategy. This underscores the imperative for marketers to prioritize privacy and protection, as the disclosure of confidential business information could result in catastrophic and immeasurable consequences for the enterprise. If marketers believe that generative AI is unreliable in protecting privacy, or if they experience anxiety about privacy risks, their negative outcome expectancies toward generative AI will significantly increase, reducing their trust in AI and consequently diminishing their intention to use AIGC.

On the other hand, AI privacy anxiety reflects an individual's perception of risk, which will influence their behavioral strategies. Marketers' AI privacy anxiety includes not only anxiety about the leakage of their personal information but also concerns about the theft of company business secrets and strategic deployments. Once privacy is breached, competitors might obtain and utilize the company's marketing ideas and plans. Therefore, if marketers' levels of AI privacy anxiety are high, they will find it more challenging to trust generative AI to assist in their work and will choose to reduce their AIGC intention to minimize risk. Based on the above analysis, we propose hypotheses H1a and H2a:

H1a: AI privacy anxiety has a significant negative effect on AI trust. H2a: AI privacy anxiety has a significant negative effect on AIGC intention.

AI Bias Anxiety, AI Trust, and AIGC Intention

Marketers' work is inherently customer-oriented, requiring close engagement with consumers or clients to ensure satisfaction through designing attractive products and services and producing suitable advertisements, promotional videos, or planning strategies. These tasks are closely linked to the company's marketing performance, leading marketers to scrutinize the applicability of the content produced. AI bias occurs when AI systems generate systematically unfair outcomes. Although scholars and AI designers are doing their best to reduce AI algorithm bias, this issue is still unavoidable at present. When marketers have concerns that AIGC may contain biases, such as potential price or gender discrimination, it can severely affect their trust in AI, and this lack of trust along with negative outcome expectations might reduce their intention to use AIGC.

Furthermore, marketers' bias anxiety regarding generative AI reflects their uncertainty about AI risks, which could weaken their propensity to adopt new technologies. Higher levels of AI bias anxiety might lead marketers to perceive the unreliability of AI algorithms as potentially harmful and loss-inducing, thus diminishing their trust in generative AI. If marketers have AI bias anxiety, they may fear that algorithmic biases could lead to dissatisfaction among particular customer groups and unfair marketing strategies, with severe cases possibly leading to legal litigation. This could significantly impact the company's brand and reputation, reducing marketers' willingness to use AIGC. Consequently, marketers will become more cautious when using AI to generate content. Based on the above analysis, we propose the following hypotheses:

H1b: AI bias anxiety has a significant negative impact on AI trust. H2b: AI bias anxiety has a significant negative impact on AIGC intention.

AI Opacity Anxiety, AI Trust, and AIGC Intention

AI opacity anxiety reflects marketers' concerns about the invisible decision-making processes and the difficulty in explaining complexity, which negatively impacts trust from users rather than developers and may influence their intention to use AI to generate content. Through extensive pretraining, generative AI technology learns the complex patterns and structures of language. Although marketers can obtain desired content and results through simple and direct conversations, the internal decision-making mechanisms of generative AI are like a black box to users. Marketers cannot understand how generative AI generates specific decision content or comprehend its working principles, and this opacity anxiety leads marketers to expect unsatisfactory results from dialogue training.

According to the social cognitive theory's interpretation of outcome expectations, we have reason to believe that a lack of transparency will reduce marketers' trust in generative AI and their intention to use AIGC. On the other hand, AI opacity anxiety reflects a negative intrinsic motivation, where marketers feel they may not be able to fully understand the AI's algorithmic mechanisms and control the potential outcomes it may produce. If the decision-making process and mechanisms are opaque, marketers may doubt the accuracy and ethics of the generated content. Once the results generated by generative AI do not match their judgments due to the lack of transparent mechanism explanations, marketers are likely to distrust AI and choose to reject the use of generative AI to avoid potential risks and undesirable outcomes. Based on the above analysis, we propose the following hypotheses:

H1c: AI opacity anxiety has a significant negative impact on AI trust. H2c: AI opacity anxiety has a significant negative impact on AIGC intention. H3: AI trust has a significant positive effect on AIGC intention.

Mediating Role of AI Trust

Trust is considered a key factor in the success of relationships between humans and nonhuman agents, such as AI, influencing the adoption of AI. Building on the above hypotheses, we further propose that AI trust mediates AI anxiety and AIGC intention, given that marketers' concerns about privacy, bias, or opacity can lead to skepticism and distrust toward generative AI. This distrust may cause marketers to question the competence and capabilities of generative AI designers and the technology itself, potentially leading to aversion and rejection, reducing their AIGC intention.

First, from a cognitive perspective, AI anxiety (including AI privacy, AI bias, or AI opacity anxiety) reflects marketers' unclear and negative expectations toward generative AI technology and its designers. In such instances, they perceive generative AI technology as unreliable and lacking credibility, leading marketers to doubt the effectiveness of generative AI in executing and completing marketing tasks, thereby weakening their AIGC intention. Conversely, transparency, along with the clarity of algorithmic and functional logic, may enhance the sense of trust. Second, from an emotional and behavioral perspective, the negative emotions elicited by AI privacy, AI bias, and AI opacity anxieties—such as fear and unease—lead marketers to experience a diminished trust in generative AI.

They develop aversion and rejection toward using generative AI technology even if it offers significant technological advantages. Consequently, marketers may hesitate and reduce their AIGC intention. Therefore, we propose the following hypothesis:

H4: AI trust mediates the relationship between AI anxiety and AIGC intention.

The Moderating Role of AI Self-Efficacy

Self-efficacy is an individual's belief in their capability to complete tasks, and it is an intrinsic trait. People with high self-efficacy have greater motivation and confidence in accomplishing a task and are better at overcoming difficulties and obstacles. In this paper, AI self-efficacy refers to marketers' confidence and perceived ability in learning and using generative AI. We believe that AI self-efficacy can positively regulate the impact of AI trust on the positive effects of AIGC, thereby mitigating the adverse effects of AI anxiety. On the one hand, social cognitive theory suggests that individuals with high self-efficacy possess more substantial confidence and perceived ability. Therefore, under high levels of AI self-efficacy, marketers would feel more confident and capable of solving problems and challenges in the process.

Similarly, they are more likely to perceive the positive aspects brought by AI, which strengthens the positive relationship between AI trust and AIGC intention, as they are more inclined to leverage AI's benefits effectively.

On the other hand, individuals with high AI self-efficacy are more inclined to continuous learning and adaptation. Even if AI anxiety leads to a lower level of AI trust among marketers, affecting their intention toward AIGC, the moderating effect of AI self-efficacy allows for a higher AIGC intention compared to the average situation under lower levels of AI trust. This is because individuals with high AI self-efficacy believe that through their learning and effort, they can adapt to the negative effects of distrust, raising the expectations for positive outcomes. Therefore, we propose the following hypothesis:

H5: AI self-efficacy moderates the relationship between AI trust and AIGC intention.

Based upon our hypotheses, we propose the following conceptual framework.

RESEARCH METHOD

Sample Selection and Data Collection

We tested our hypotheses by surveying marketers' thoughts and feelings on generative AI. It should be noted that the term “marketers” here not only includes sales personnel but also encompasses individuals involved in marketing planning, operations, product development, and testing, among other roles related to marketing. Our sample data is derived from various cities in China. By avoiding the concentration of sample data in specific cities, we can enhance the diversity and representativeness of the sample. Additionally, in our questionnaire, we specifically requested marketing professionals with relevant work experience and knowledge or familiarity with generative AI to ensure that the sample aligns with the research context described in this study.

After determining the scope of sample selection, we performed a preliminary test with a small sample size. We rigorously reviewed the questionnaire with five marketing professors and 10 marketing doctoral students (active research scholars) to ensure the questions aligned with the research context. Subsequently, we conducted a pretest with 50 marketing professionals to examine the questionnaire's clarity, conceptual validity, and reliability. This can ensure consistency in the concepts and semantics of the questions and better control for respondent biases.

Based on the feedback from the pre-survey, we made some minor adjustments to the wording of relevant variable items and formed the formal survey questionnaire. In the formal survey, a professional research company was contracted to collect data. Their extensive data collection experience can help improve our data's reliability and validity. Through a combination of online and offline approaches, we collected 600 questionnaires in April 2024. The data was gathered from various cities across China, including Beijing, Dalian, Shenyang, Guangzhou, and Changsha. This process was conducted using simple random sampling to ensure a representative sample.

To ensure empirical rigor, we removed the invalid data samples based on the following criteria: first, if the respondent did not complete the questionnaire information; second, if the respondents were unfamiliar with the questions or the generative AI mentioned in the questionnaire; third, if the participants had less than one year of relevant work experience; fourth, if the data was suspected to be randomly answered (incorrectly answering the screening questions we set, short questionnaire response time, or providing almost identical answers to most questions). Finally, we screened and obtained 495 valid questionnaires with an effective response rate of 82.5%. To further ensure the data quality, we examined the data from the questionnaires. We found that the average work experience of the respondents was 7.29 years, and 76.1% of them had three years or more of relevant work experience.

In terms of gender composition, 50.5% of the respondents were female, while 49.5% were male. Age-wise, 8.3% of the respondents were below 20 years old, 49.5% were between 20 and 30 years old, 29.7% were between 31 and 40 years old, and 12.5% were over 40 years old. Regarding work experience, 24.0% had less than three years of experience, 23.5% had three to five years, 31.1% had six to 10 years, and 21.4% had more than 10 years of experience. By qualification, 41.2% of the respondents had qualifications below a bachelor degree, 49.1% were college graduates, and 9.7% had a postgraduate degree or higher (as shown in Table 1).

continued on following page

Volume 33 • Issue 1 • January-December 2025

Construct Measurement

In this study, all the multi-item variables were derived from well-established scales in authoritative journals and modified appropriately to the research context. All items were measured using a 5-point Likert scale (1 = Strongly Disagree, 5 = Strongly Agree).

In Table 2, AI privacy anxiety refers to the concerns of marketers about their privacy and creative ideas being monitored while using generative AI to complete tasks (such as designing marketing strategies, advertising campaigns, etc.; Li & Huang, 2020; Smith et al., 1996). AI opacity anxiety refers to the inherent anxiety that marketers feel about the opacity or secrecy of generative AI decision-making mechanisms. AI bias anxiety refers to the concerns about bias in the content generated when using AI to produce the required work tasks. AI trust measures marketers' perceived reliability and confidence in generative AI technology. We integrated and adapted existing scales based on the research context and used a three-item scale to measure it.

AI self-efficacy was referenced from Compeau and Higgins (1995), Rayburn et al. (2021), and Tierney and Farmer (2002) and has been integrated and adapted within the context of this paper to measure marketers' confidence and perceived ability to use generative AI. AIGC intention refers to the behavioral intention of marketers to use generative AI in their work. We used a three-item scale to measure it.

continued on following page

Note. AI = artificial intelligence, α = Cronbach’s alpha, CR = composite reliability, AVE = average variance extracted. Model fit index: CMIN/DF=1.203, RMSEA = 0.020, GFI = 0.964, NFI = 0.960, IFI = 0.993, TLI = 0.991, CFI = 0.993, SRMR = 0.058, GOF = 0.677.

RESULTS ANALYSIS

Reliability and Validity Testing

In our research, we used Cronbach's α to test for reliability, and the results indicated that Cronbach's α for all variables exceeded the threshold of 0.70 (the smallest value is 0.819 for AI self-efficacy), suggesting good internal consistency. Furthermore, the composite reliability also exceeded the recommended minimum of 0.70 (the smallest value is 0.819 for AI self-efficacy), indicating that our scale is reliable. We conducted a confirmatory factor analysis. The factor loadings for each item exceeded 0.7, and the average variances extracted (AVEs) were all over 0.5, indicating that our study has good construct reliability and convergent validity. Previous research indicates that discriminant validity between two constructs is achieved when the square root of the AVE of one construct exceeds the correlation between this construct and any other construct.

The analysis showed that the smallest square root of the AVE is 0.776 (for the AI self-efficacy construct), surpassing the correlation observed between AI self-efficacy and others (as shown in Table 3). We also reported cross-loading (as shown in Table 4). To ensure the rigor of our study, we conducted a heterotrait–monotrait analysis. The heterotrait–monotrait ratio is smaller than 0.9 (the absolute value of the largest heterotrait–monotrait value is 0.537), further confirming the constructs' discriminant validity.

Note. NA = not applicable; SD = standard deviation; AIPA = AI privacy anxiety; AIOA = AI opacity anxiety; AIBA = AI bias anxiety; AIT = AI trust; AISE = AI self-efficacy; AIGCI = artificial intelligence generated content intention. The bold numbers on the diagonal represent the square roots of the AVE for each corresponding variable, the numbers below the diagonal are the zero-order correlation coefficients between variables, and the numbers above the diagonal are the adjusted correlation coefficients accounting for common method bias. Market uncertainty (MC) is a label variable.

Common Method Bias

Our sample data were self-reported by marketing professionals; therefore, common method bias may exist. Based on previous authoritative research, we have adopted multiple methods and procedures to test the likelihood of its occurrence. First, we implemented procedural remedies. Before the formal questionnaire survey process, we pretested with a small sample size. We carefully arranged the order of the questionnaire to ensure that the central questions did not appear in a hypothesized sequence. We also ensured the privacy and clear understanding of the respondents to promote more reliable responses and to control social desirability bias. Second, we conducted a Harman single-factor analysis using IBM SPSS, employing principal component analysis with a factor extraction criterion of eigenvalues greater than 1 to assess common method bias.

By conducting factor analysis on the main items in this study, we obtained six factors that accounted for 74.253% of the variance. The first factor had a variance contribution rate of 37.240%. It can be preliminarily concluded that the common method bias in this study is not severe. Third, following the research conducted by Lindell and Whitney (2001), we selected market uncertainty as a marker variable, which is conceptually independent of the theoretical underpinnings of the model. Then we conducted a partial correlation analysis. The findings indicate that the inclusion of this marker variable did not result in any significant alterations to the observed correlations among the original variables.

Finally, we did a single-factor test and confirmatory factor analysis on the main variables of the model using Amos. All variables showed different factor loadings; the most significant factor variance was 11.20%. The fit of the single-factor model (χ2/df = 12.269, IFI = 0.613, TLI = 0.562, CFI = 0.611, RMSEA = 0.151, GOF = 0.677) was significantly poorer compared to the 10-factor model, indicating that no single factor explained a large portion of the covariance. This suggests that the issue of common method bias is not severe.

,, One-tailed test: p < 0.001, p < 0.01, p < 0.05.

Hypothesis Testing

After standardizing the measurement methods, we used SPSS and Amos to conduct multiple regression models to test the proposed hypotheses empirically. This study first centralized the independent and moderating variables to avoid the potential multicollinearity issues that may arise when exploring the moderating effects. In our analysis, Models 1–5 used AI trust as the dependent variable, with Model 1 including all control variables. Model 5 simultaneously included all three independent variables to examine their impact on AI trust. Models 6–10 used AIGC intention as the dependent variable, with Model 6 including the same control variables. Models 7, 8, and 9 added all independent and mediating variables, respectively, while Model 9 added the moderating variable and checked the regression results of the overall model. The results are revealed in Table 4.

Note. The numbers inside parentheses are t-values.

,, One-tailed test: p < 0.001, p < 0.01, p < 0.05.

H1a–H1c hypothesized that AI privacy anxiety, AI opacity anxiety, and AI bias anxiety would negatively impact AI trust. In Model 2, when all three independent variables were entered together, the coefficients remained significant (βAIPA' = -0.206, p < 0.001; βAIOA' = -0.256, p < 0.001; βAIBA' = -0.221, p < 0.001); thus, H1a–H1c are supported. This indicates that AI bias anxiety, AI opacity anxiety, and AI privacy anxiety all have significant effects on the construct of trust.

Based on the absolute values of the coefficients, the importance of the influence of the three types of anxiety on the construct of trust can be ranked as follows: first, AI opacity anxiety (βAIOA' = -0.256); second, AI bias anxiety (βAIBA' = -0.221); and third, AI privacy anxiety (βAIPA' = -0.206). H2a–H2c predicted a negative effect of AI privacy anxiety, AI opacity anxiety, and AI bias anxiety on AIGCI, and in Model 4, the regression coefficients of these variables were significantly negative (βAIPA'' = -0.244, p < 0.001; βAIOA'' = -0.222, p < 0.001; βAIBA'' = -0.275, p < 0.001); hence, H2a–H2c are supported.

In Model 5, the effect of AI trust on AIGC intention was examined independently, showing a significant positive impact (β = 0.420, p < 0.001). In Model 6, upon adding AI privacy anxiety, AI opacity anxiety, AI bias anxiety, and AI trust, the regression coefficient of AI trust was significantly positive (β = 0.161, p < 0.001). The coefficients of AI privacy anxiety, AI opacity anxiety, and AI bias anxiety decreased significantly, indicating that AI trust has a positive effect on AIGC intention and partially mediates the relationship between AI anxiety variables and AIGC intention. Thus, H3 and H4 are supported.

In Model 7, AI self-efficacy and its interaction term with AI trust were introduced. Following the approach of Edwards and Lambert (2007), we standardized the variables before assessing the moderating effect. The interaction term's coefficient was significantly positive (β = 0.185, p < 0.001), suggesting that AI self-efficacy has a positive moderating effect. To further illustrate the moderating effect, simple slope analysis was applied and a moderation effect diagram was produced, as shown in Fig. 2, indicating that when AI self-efficacy is high, the linear slope between AI trust and AIGC intention is steeper. Therefore, H5 is supported.

Note. AI = artificial intelligence; AIGC = artificial intelligence generated content.

To further clarify the mediating effects, we used the process tool for a bootstrapping analysis of the mediating roles. The bootstrapping sample size was set at 5,000, and a 95% confidence interval (CI) was included in the results, as shown in Table 5. Through AI trust, the indirect effect of AI privacy anxiety on AIGC intention was -0.114, which was statistically significant [95% CI = (-0.155,

-0.076), excluding 0]. Under the mediation of AI trust, the indirect effect of AI opacity anxiety on AIGC intention was -0.124, which was statistically significant [95% CI = (-0.165, -0.084), excluding 0]. Through the mediation of AI trust, the indirect effect of AI bias anxiety on AIGC intention was -0.112, which was statistically significant [95% CI = (-0.151, -0.074), excluding 0].

Note. Independent variable: AIGC intention. CI: 95%. CI = confidence interval; AIPA = AI privacy anxiety; AIOA = AI opacity anxiety; AIBA = AI bias anxiety; AIT = AI trust; AISE = AI self-efficacy; AIGCI = artificial intelligence generated content intention.

DISCUSSION.

Research Findings

This study, adopting the perspective of social cognitive theory, empirically investigates the influence mechanism of AI technology anxiety (AI privacy anxiety, AI bias anxiety, AI opacity anxiety) on marketers' intention to use AIGC. The research findings indicate that AI opacity anxiety, AI bias anxiety, and AI privacy anxiety affect AI trust in descending order. Furthermore, these anxieties ultimately suppress AIGC intention by reducing trust in AI. Moreover, we identified the moderating role of AI self-efficacy, which can positively moderate the promoting effect of AI trust on AIGC intention, thereby alleviating the negative impact of AI anxiety.

Our research focuses on how AI technology anxiety, a negative aspect, affects AIGC intention among marketers. In contrast, previous studies have focused mostly on the positive factors that influence the intention of general users or consumers. Unlike ordinary users or consumers, marketers, due to their work environments and task demands, tend to be guided more by performance and benefits, resulting in differences in technology acceptance and usage. The content that marketers need to generate (such as promotional videos, advertisements, posters, marketing plans, etc.) is not solely for their own needs but must cater to a broad customer and consumer base, placing higher demands on the applicability and accuracy of AI-generated results.

Therefore, they need to consider whether the content generated by AI contains biases, such as whether advertisements or promotional videos implicitly contain controversial elements such as racial or gender discrimination or whether AI possesses transparency—that is, whether they can understand AI's decision-making mechanisms to correct and optimize their required plans continuously. In marketing, the negative results produced by AI could be fatal to a company. Hence, AI technology anxiety can affect AI trust. Simultaneously, in the absence of adequate safeguards, marketers might also reduce their intention to use AIGC to ensure that everything is conducted within their control.

p < 0.001, p < 0.01, p < 0.05.

Theoretical Implications

This study has three significant theoretical contributions. First, we have identified and delved deeply into three types of AI-related anxieties among marketers: AI privacy anxiety, AI bias anxiety, and AI opacity anxiety. Previous research predominantly discussed the facilitators of technology behavior intention through the technology acceptance model and the Unified Theory of Acceptance and Use of Technology model, with these anxiety factors as negative aspects rarely addressed. Our findings, from the perspective of social cognitive theory, demonstrate that these dimensions of AI anxiety can reduce marketers' intention to use AIGC by lowering trust in AI, thereby enriching the research on social cognitive theory and providing a significant supplement to the framework for understanding and predicting marketers' intention to use generative AI technology.

Moreover, research in the marketing context provides deeper insights into AI anxiety within a specific group, enriching existing research on AI anxiety.

Second, we have focused on studying how AI trust is a mediating variable affecting the relationship between AI technological anxiety and AIGC intention. This finding underscores the importance of AI trust in marketers' use of generative AI, providing a more comprehensive perspective for observing and measuring this critical factor and further elucidating the impact of outcome expectations on individual behavioral choices within the framework of social cognitive theory. In this context, it has also expanded the scope of research on AI trust mechanisms.

Third, we have explored the moderating role of AI self-efficacy, which is particularly significant as it reveals how individual confidence levels influence the relationship between marketers' AI trust and AIGC intention, thereby reducing the negative impact of AI technological anxiety. This offers a new angle to understand and promote the positive adoption of generative AI technology. This also extends research on AI self-efficacy and brings new theoretical insights to social cognitive theory and self-efficacy theory research.

These theoretical contributions provide a more profound understanding framework for future research and practice related to AI marketing, promoting the practical application and development of generative AI technology in marketing.

Practical Implications

Our research provides the following recommendations for corporate marketing management practice. First, companies need to pay attention to the impact of marketers' anxiety about generative AI. Although generative AI is gradually becoming a mainstream trend in marketing, marketers harbor various anxieties about this new technology, including issues related to privacy, bias, and opacity. Specifically, they worry that generative AI might infringe on corporate and personal privacy, exhibit algorithmic bias, and lack a transparent decision-making process. These factors significantly reduce their trust in and intention to use generative AI.

Therefore, companies must actively identify and address these sources of anxiety by establishing effective privacy protection policies, algorithmic fairness review mechanisms, and measures to increase technological transparency, thereby enhancing marketers' trust in and acceptance of generative AI.

Second, to effectively address marketers' anxiety about AI technology, companies should provide comprehensive and ongoing support and training. To alleviate privacy anxieties, companies can, in accordance with the California Consumer Privacy Act and the General Data Protection Regulation, formulate clear usage principles and standards, guiding marketers on how to use generative AI safely in their work to ensure data and privacy security. For anxieties about AI bias and opacity, companies should offer systematic education and training to raise employees' awareness of potential biases and data ethics issues.

Regular evaluations and feedback, as well as cross-department collaboration and open discussions, can increase AI technology transparency, helping marketers understand the internal mechanisms and decision-making processes of generative AI, thereby demystifying the technology and reducing their anxiety. Companies that have the resources can set up dedicated AI departments or bring in advanced human–AI collaboration systems. These are very effective methods.

Last, companies should actively cultivate marketers' AI self-efficacy to mitigate the negative impact of AI anxiety and promote the widespread adoption of generative AI technology. Companies can set specific and achievable incentive goals and use performance evaluations and reward mechanisms to encourage marketers to actively try and use generative AI. Additionally, establishing a culture that supports innovation is crucial. This not only requires the management to fundamentally recognize the immense potential and importance of generative AI in marketing work but also involves implementing specific measures such as management support and employee training to enhance employees' technical confidence and application capabilities.

Through these comprehensive measures, companies can effectively promote the widespread adoption of generative AI in marketing, ultimately enhancing overall marketing efficiency and competitiveness.

CONCLUSION.

Based on existing research, we explored the relationship between marketers' AI anxiety and their intention to use AIGC. This exploration helps us understand the negative factors influencing marketers' AI acceptance intentions in the marketing context and aids in identifying key issues surrounding AI anxiety. Furthermore, we discussed the mediating role of AI trust and the moderating role of AI self-efficacy in this relationship. These research findings enrich and enhance the body of knowledge on AI in the marketing field, thereby improving our understanding of the use of generative AI by this particular group of marketers. Generative AI is gradually becoming an important part of marketing work. Therefore, reducing marketers' AI anxiety and enhancing their intention to use such technologies are crucial.

This paper suggests that companies adopt specific strategies, such as establishing unbiased and transparent principles for AI usage and providing AI technology support and training, to reduce marketers' AI anxiety, improve their trust and self-efficacy, and thereby help companies enhance marketers' AIGC intention.

Our study has certain limitations that should be addressed in future research. First, our sample included a broad range of marketing practitioners without distinguishing specific job roles or hierarchical positions, which may lead to differing levels of AI-related anxiety and subsequently affect trust and willingness to adopt AI. Additionally, the sample consisted solely of managerial personnel from China, necessitating caution when generalizing findings to other economies due to China's unique cultural characteristics. We also did not examine potential moderating variables such as country, culture, and industry, which future studies should explore to provide more targeted business recommendations.

Second, we did not categorize or discuss the types of AIGC. Some research in library and information science and information systems has already discussed the types of AIGC. In the context of marketers' work, the willingness for human-centric AIGC may differ from that for AI-centric AIGC in terms of influencing and moderating factors. In future research, we can further differentiate the types of AIGC, the differences in human–machine learning, and the methods of human-intelligence interaction, potentially leading to differentiated conclusions.

Last, our research methodology still has certain limitations. We conducted a questionnaire survey among marketers, and although we controlled for common method bias and performed reliability and validity tests, the impact caused by the inherent limitations of the method cannot be resolved entirely. Future research could consider collecting matched questionnaires from employees and management or combining long-term sequences, scenario experiments, and panel data methods to improve and expand the study.

AUTHOR NOTE

The authors of this publication declare there are no competing interests This research was supported by the National Natural Science Foundation of China (grant number 72072026).

PROCESS DATES