Sage Journals: Discover world-class research

Abstract

This study conducted a 2 (chatbot expertise cues: high vs. low) × 2 (correction sidedness: one-sided vs. two-sided) between-subjects experiment to investigate the effects of chatbot features and output framing on misinformation correction and public perceptions of cultivated meat. Results highlighted the importance of chatbot expertise cues in shaping credibility perceptions of misinformation correction, with two-sided corrections from high-expertise chatbots perceived as more credible than those from low-expertise chatbots. Credibility perceptions mediated the interaction effect between chatbot expertise cues and correction sidedness on individuals’ attitudes and consumption intentions towards cultivated meat. Both theoretical and practical implications of these findings were discussed.

Keywords

AI chatbot misinformation correction message sidedness expertise cues cultivated meat

The evolution of artificial intelligence (AI) has significantly impacted the creation and dissemination of information, introducing both challenges and opportunities within the information environment (Wolfe & Mitra, 2024). This development has been particularly influential in the field of science communication, where AI exhibits a dual role. On one hand, the misuse of AI exacerbates the spread of scientifically relevant misinformation by simplifying and accelerating the production of deceptive content, a phenomenon prominently observed during the COVID-19 pandemic (Monteith et al., 2024). On the other hand, AI offers advanced algorithmic solutions and computational tools that significantly enhance fact-checking initiatives and reduce the spread of scientific misinformation (Demartini et al., 2020; Moon et al., 2023; Sun & Pan, 2024). This dual role of AI—its capacity to both hinder and aid science communication—necessitates a deeper investigation into how AI tools can be optimized to enhance the accuracy and impact of science-related messaging.

Research has demonstrated the efficacy of AI tools, such as conversational chatbots, in achieving persuasive objectives across various contexts in science communication (e.g., Wang & Peng, 2023). For instance, Altay et al. (2022) indicated that AI-enabled chatbots could foster positive attitudes toward novel food technologies (e.g., genetically modified [GM] foods) by enhancing human-AI interactive argumentation. Similarly, Moon et al. (2023) observed that fact-checking COVID-related misinformation using AI sources led to less motivated reasoning than fact-checking attributed to human experts. Despite the evident potential of AI in persuasive science communication, its effectiveness in combating science-related misinformation remains unclear. Further examination is needed to understand how the characteristics of AI agents and the features of the misinformation corrections they deliver might affect people’s resistance to scientifically relevant misinformation and influence their attitudes and behavioral intentions toward scientific issues and technological innovations.

Recent studies on persuasion have highlighted the significant impact of AI features on individuals’ attitudes and behaviors (e.g., Cheung & Ho, 2024; Ho & Cheung, 2024; Wu et al., 2024). For example, Wu et al. (2024) found that the specialization or expertise of AI agents substantially affects persuasion outcomes, with specialized applications more likely to engender greater trust and behavioral intentions than the generalist counterparts. Furthermore, the framing of AI-generated content shapes individuals’ attitudinal and behavioral responses; for instance, products recommended by AI as vices (negatively framed) elicit stronger adverse reactions than those framed as virtues (Yang et al., 2024). While the individual effects of AI expertise and content framing are well-documented, the interaction between these two aspects of factors in AI-mediated communication remains underexplored. Therefore, when deploying AI chatbots to address science-related misinformation, it is essential to examine how chatbot characteristics (e.g., expertise) and the construction of the misinformation corrections it provides (e.g., correction sidedness) jointly influence individuals’ cognitive, attitudinal, and behavioral responses. This calls for further integration of the theoretical frameworks from human-computer interaction and media framing to enhance the effectiveness of AI-facilitated misinformation correction and foster public understanding of scientific issues.

Informed by the theoretical perspectives of the Computers as Social Actors (CASA) paradigm (Nass & Moon, 2000) and message sidedness theory (e.g., Allen, 1991), this study examines the role of chatbot expertise and the sidedness of misinformation correction provided by the chatbot in combating scientific misinformation and shaping both attitudinal and behavioral responses toward emerging technologies. The CASA paradigm suggests that individuals interact with computers similarly to how they interact with humans, applying comparable heuristics to both human-computer and human-human interactions (Nass & Moon, 2000). Message sidedness theory posits that a message’s persuasive impact varies by its framing: One-sided messages present a single viewpoint, while two-sided messages include both supporting and opposing viewpoints (Allen, 1991).

Using an experimental design, this study explores how chatbot expertise cues (high vs. low) and correction sidedness (one-sided vs. two-sided) influence individuals’ credibility perceptions of the corrections provided by the chatbot, as well as their attitudes, support, and consumption intentions toward cultivated meat—an emerging and controversial technology particularly susceptible to misinformation. Theoretically, this study advances our understanding of how the source characteristics and output framing of AI chatbots individually and jointly shape public perceptions of scientifically relevant issues, contributing to the literature on human-AI interaction, science communication, and misinformation correction. Practically, these findings could inform strategies for employing advanced communication technologies, such as AI chatbots, to effectively counter misinformation in scientific domains.

Literature Review

Study Context: Scientific Misinformation on Cultivated Meat

This study evaluates the effectiveness of AI-powered chatbots in correcting misinformation and shaping public attitudes and behavioral intentions toward cultivated meat—a novel food technology involving the cultivation of animal cells outside an animal’s body (Bryant & Barnett, 2020). Singapore, the first country to commercialize cultivated meat in 2019 to strengthen food security (Ives, 2020), serves as the context for this research. Despite the potential of cultivated meat to provide a sustainable alternative to conventional livestock farming, addressing environmental, health, and ethical concerns (Treich, 2021), low familiarity among the Singaporean public increases its susceptibility to misinformation (Bryant & Barnett, 2020; Ou & Ho, 2024a, 2024b). Prevailing misinformation on cultivated meat includes unfounded claims that cultivated meat is derived from cancerous cells (AP News, n.d.), lacks proper labeling when sold (Mueller, n.d.), and is developed with hidden agendas (Reuters, n.d.). Without effective interventions, such misinformation may persistently skew public perception. This study therefore explores how AI chatbots can be strategically deployed to correct such misinformation and enhance public understanding, ultimately supporting informed discourse on this innovative food technology.

AI Chatbot for Misinformation Correction

Scientific misinformation, defined as information that is misleading or false relative to the best scientific evidence and contradicts statements from entities adhering to scientific principles (Southwell et al., 2022), remains a persistent challenge. Numerous institutions struggle with employing fact-checkers to identify and correct problematic online content, often facing issues related to scalability, quality, and timeliness of corrections (Zhou et al., 2024). In addition, laypersons frequently encounter difficulties in finding reliable sources for fact-checking. A recent study by Aslett et al. (2024) indicated that individuals attempting to fact-check recent misinformation via search engines often failed to locate credible sources, frequently encountering a lack of reliable information to debunk false claims. Consequently, there is a critical need for robust, technology-powered platforms capable of scaling up and accelerating misinformation correction.

In science communication, AI tools such as conversational chatbots have become essential for increasing public engagement and understanding of complex topics (Altay et al., 2022; Wang & Peng, 2023). These tools employ advanced algorithms to not only present information but also to facilitate interactive dialogues, enhancing the dynamism of communication. AI chatbots also serve as effective platforms for information verification and fact-checking, providing accessible and convenient solutions (Zhao et al., 2023). For instance, a fact-checking chatbot developed by Zhou et al. (2024) uses a large language model to efficiently produce high-quality debunking messages, substantially reducing users’ misconceptions. A study by Costello et al. (2024) found that dialogs with AI chatbots could reduce belief in conspiracy theories. During the COVID-19 pandemic, various chatbots were deployed globally for fact-checking and misinformation correction, demonstrating their efficacy in combating misinformation (e.g., Amiri & Karahanna, 2022). With features like instant fact-checking, personalization, and interactivity, conversational AI chatbots show considerable potential for addressing scientific misinformation more effectively and efficiently.

Recent studies have shown how chatbot design elements affect their persuasive efficacy in science communication. Specifically, integrating social cues that signal a chatbot’s competence or human-like qualities significantly enhances users’ trust and influences their attitudes and behavioral intentions regarding issues such as biodiversity conservation (Wang & Peng, 2023) and healthcare consultations (Liu et al., 2022). While extensive research has explored the impact of these social cues as source factors influencing user responses, the role of content factors—such as the structure and presentation of chatbot-generated content—remains underexplored. Previous research has demonstrated significant interactions between source factors (e.g., source expertise) and content factors (e.g., message sidedness) in driving attitudinal and behavioral changes in the digital media context (Flanagin et al., 2020), yet their interplay in AI-driven communication remains understudied. Thus, this study aims to explore how chatbot source characteristics and content features of their corrections may jointly influence individuals’ attitudinal and behavioral responses to scientific issues, with a particular focus on chatbot-mediated corrections of scientific misinformation about cultivated meat.

Chatbot Expertise Cues

Expertise cues in a chatbot refer to elements signaling its specialized knowledge, skills, or abilities in a particular domain (Liew & Tan, 2021). These cues are designed to enhance the chatbot’s credibility and trustworthiness by conveying competence, which is crucial for increasing persuasive effectiveness in human-computer interaction (Sundar, 2008). The CASA paradigm suggests that individuals apply similar social heuristics when interacting with computers as they would with humans (Nass & Moon, 2000). According to CASA, users engaging with chatbots for misinformation correction expect the chatbot to demonstrate expertise in the relevant field, interpreting expertise cues as social signals that reflect the chatbot’s competence and specialization (Liew & Tan, 2021). In the context of virtual agents, Liew and Tan (2021) identified various cues that convey agent expertise, including appearance, social prestige, communication style, and specialization labels. For example, specialization cues in web agents that highlight their specific (vs. generalized) functions have been shown to foster greater trust and positive attitudes toward both the agents and the information they present (Koh & Sundar, 2010).

Extensive reviews and empirical studies consistently show that perceived source expertise is a pivotal factor in judging information credibility (Go et al., 2014; Metzger et al., 2003; Ou & Ho, 2024a). In science-related misinformation correction, source expertise is essential for effectively addressing misinformation on complex topics, such as GM foods (Wang, 2021) and the Zika virus (Vraga & Bode, 2017). For instance, Vraga and Bode (2017) found that health misinformation corrections from high-expertise sources, like the Centers for Disease Control and Prevention, were considered more credible than those from laypersons. Similarly, Wang (2024) demonstrated that individuals were more likely to accept a corrective message from an expert source than from a social peer.

High-expertise cues also positively influence user attitudes and behavioral intentions by enhancing the credibility and professionalism of the correction source (e.g., Vraga & Bode, 2017; Yuan et al., 2019). Research shows that perceived expertise directly impacts attitude formation, particularly when users engage with novel or scientifically complex information (Wang & Peng, 2023). Given that cultivated meat is a relatively new and sometimes controversial topic, individuals are likely to form more favorable attitudes if the correction comes from a source perceived as knowledgeable and professional. For example, studies in science communication and food technology (e.g., Yuan et al., 2019) have shown that expert-authored content on novel foods, such as GM foods, is more persuasive, leading to more positive public attitudes.

In addition, support for novel technologies, especially in food science, often hinges on the perceived reliability of the information source (e.g., Chambers et al., 2023). Expertise cues not only imply knowledge but also suggest that the chatbot’s corrections and recommendations are trustworthy and well-informed. This can foster greater confidence in such technologies, thereby increasing support for cultivated meat. Behavioral intentions, such as the intention to consume novel food products, could also be strongly influenced by the perceived expertise of the correction source. When users perceive the source as highly knowledgeable, they are more likely to trust its guidance, leading to increased openness to recommendations (Go et al., 2014). In the case of cultivated meat, users who see the chatbot as an expert in correcting related misinformation will be more likely to consider consuming it, trusting that the corrections are based on accurate, research-backed insights. Studies in health communication (Vraga & Bode, 2017) and food science (Yuan et al., 2019) have demonstrated that high-expertise sources significantly increase individuals’ willingness to adopt behaviors aligned with expert recommendations.

Taken together, given the significant impact of source expertise on credibility perceptions, as well as on attitudinal and behavioral responses to emerging technologies, we propose the following hypothesis:

Hypothesis 1 (H1): An AI chatbot with higher expertise cues will elicit (a) greater credibility perceptions of the misinformation correction it delivers, (b) more positive attitudes toward cultivated meat, (c) increased support for cultivated meat, and (d) greater intentions to consume cultivated meat than a chatbot with lower expertise cues.

Effects of Correction Sidedness

Message sidedness refers to the framing of a message based on the number of perspectives it includes, which significantly influences its persuasiveness (Allen, 1991). Specifically, a two-sided message presents arguments both for and against an issue, whereas a one-sided message supports only one perspective (Winter & Krämer, 2012). Two-sided messages can be further classified based on whether opposing arguments are refuted: two-sided non-refutational messages present both positive and negative information without countering the negative, while two-sided refutational messages provide both sides but immediately counter the negative argument to strengthen the positive perspective (Allen, 1991). This study focuses on the two-sided refutational approach, where misinformation is initially acknowledged but quickly countered with corrective information, thereby enhancing the effectiveness of the correction.

Originally explored in advertising and marketing, message sidedness has also been found to influence the selection of science-related articles (Winter & Krämer, 2012) and to enhance the credibility and persuasiveness of misinformation corrections (Wang & Huang, 2021; Xiao & Su, 2021). Two-sided refutational messages, in particular, are considered more persuasive and credible for correcting misinformation, as supported by inoculation theory (McGuire, 1964). According to this theory, presenting a weaker version of an opposing argument within a two-sided message strengthens the persuasiveness of the primary argument by preemptively addressing potential misinformation. This approach not only alerts individuals to misleading information but also bolsters the resilience of their understanding through proactive refutation (Lewandowsky & Van Der Linden, 2021). Research has demonstrated the effectiveness of two-sided correction messages in addressing misinformation on topics such as vaping (Wang & Huang, 2021) and human papillomavirus (HPV) vaccination (Xiao & Su, 2021). Given these findings, we expect that two-sided misinformation correction is more likely to enhance people’s trust in correction messages than one-sided correction.

Two-sided refutational corrective messages can also contribute to positive attitudinal and behavioral shifts toward the subject of correction. When users see that a chatbot presents and refutes opposing views on cultivated meat, they are more likely to view the chatbot as competent, fair, and objective. This perception fosters trust in the message, leading to more positive attitudes toward cultivated meat. Prior research in science communication (Lin & Kim, 2022; Xiao & Su, 2021) has shown that individuals tend to develop more favorable attitudes toward scientific topics when exposed to two-sided messages. Two-sided corrections against misinformation about cultivated meat also foster greater support by acknowledging and addressing users’ potential doubts or misconceptions, thereby enhancing their confidence in this novel food. Studies on controversial topics, such as GM foods and HPV vaccination, indicate that two-sided refutational messages increase support by reinforcing the credibility of the message (Wang & Huang, 2021; Xiao & Su, 2021).

Furthermore, the balanced nature of a two-sided correction can positively influence behavioral intentions, such as intentions to consume cultivated meat. When an AI chatbot openly discusses and counters misinformation through a two-sided correction, users are more likely to trust the information provided, which increases the likelihood of adopting the recommended behavior. Inoculation theory supports this effect, positing that refutational messages can preemptively inoculate individuals against misinformation, strengthening their intention to act in accordance with the corrected information (Lewandowsky & Van Der Linden, 2021). Empirical studies in health and food science (Chambers et al., 2023; Vraga & Bode, 2017) show that individuals are more inclined to follow recommendations from credible, two-sided sources, suggesting that a two-sided correction from a chatbot could significantly enhance intentions to try cultivated meat. Therefore, we propose the following hypothesis:

Hypothesis 2 (H2): A two-sided correction from an AI chatbot will lead to (a) higher credibility perceptions of the misinformation correction it delivers, (b) more positive attitudes toward cultivated meat, (c) increased support for cultivated meat, and (d) greater intentions to consume cultivated meat, compared to a one-sided correction.

The Interaction Between Chatbot Expertise Cues and Correction Sidedness

The relationship between source expertise and message sidedness influencing credibility perceptions is well-documented in traditional digital environments, indicating an additive effect of these two factors on credibility perceptions. For example, Chen (2016) found that two-sided online product reviews written by experts are perceived as more useful than those authored by novices. Similarly, Flanagin et al. (2020) showed that two-sided messages on online encyclopedia sites enhance users’ perceptions of content credibility, particularly on platforms that blend expert and user-generated content. However, the interplay between expertise cues and sidedness framing in shaping credibility perceptions within AI-mediated misinformation correction has not been thoroughly investigated, highlighting the need for a more detailed exploration of how these factors interact in the context of AI-driven communication.

Moreover, the interplay between message sidedness and source expertise has been shown to significantly influence attitudes and behavioral intentions, albeit not specifically within AI-mediated contexts. Uribe et al. (2016) found that two-sided consumer reviews from high-expertise sources significantly boost purchase intentions compared to one-sided reviews from low-expertise sources. Kamins et al. (1989) found that two-sided advertisements with celebrity endorsements more effectively foster positive attitudes and increase purchase intentions than one-sided advertisements from non-celebrity endorsers. Collectively, these studies highlight the crucial role of message sidedness and communicator expertise in boosting message credibility and shaping recipient responses, suggesting a fertile area for exploration in AI-driven communication strategies.

Building upon established research demonstrating the synergistic interaction effects between source expertise and message sidedness on credibility perceptions, attitudes, and behaviors, this study posits that within AI-mediated communication, the impact of a chatbot’s expertise level may be significantly enhanced by two-sided messaging compared to one-sided messaging. This hypothesis is supported by research indicating that information perceived as unbalanced or from less reputable sources is generally considered of lower quality (e.g., Flanagin et al., 2020). Accordingly, we contend that misinformation corrections delivered by chatbots with higher (vs. lower) expertise cues are likely to be more effective in shaping credibility perception, attitudes, and behavioral intention, especially when incorporating two-sided (vs. one-sided) corrections. Therefore, we propose:

Hypothesis 3 (H3): An AI chatbot’s expertise cues and the sidedness of its correction will interact. Specifically, the influence of the chatbot’s expertise cues on (a) credibility perceptions of the misinformation correction, (b) attitudes toward cultivated meat, (c) support for cultivated meat, and (d) intentions to consume cultivated meat will be stronger when the correction is two-sided (vs. one-sided).

The Mediating Role of Correction Credibility

Existing research has established that perceptions of message credibility serve as a vital mediator between information source or message attributes and changes in individuals’ attitudes and behaviors across various communication contexts, including advertising (Jäger &Weber, 2020) and science communication (Wang & Huang, 2021). For example, Jäger and Weber (2020) demonstrated that the way green product benefits are framed—through high or low construal levels—affects purchase intentions by influencing perceived message credibility. Similarly, Wang and Huang (2021) showed that the perceived believability of a message positively mediates the relationship between message sidedness and support for prohibiting e-cigarette use, particularly when the message is presented narratively rather than as factual information. These studies highlight that attitudinal and subsequent behavioral changes are contingent upon the credibility of the information presented.

In the context of emerging scientific issues like cultivated meat, where the product is not widely known or accessible and claims about it are difficult to verify, perceptions of relevant message credibility play a pivotal role in shaping public attitudes and behavioral intentions (Jäger & Weber, 2020). As such, credibility perceptions serve as a vital link between chatbot-delivered messages about cultivated meat and public responses to this novel food. Consequently, we argue that the impact of chatbot expertise (source characteristics) and correction sidedness (message features) on people’s attitudes and consumption intentions toward cultivated meat is mediated by their perceptions of credibility. Given the aforementioned reasoning, corrective messages that present two-sided arguments from a high-expertise chatbot are likely to be perceived as more credible than one-sided corrections from a general expertise chatbot, leading to more favorable attitudinal and behavioral outcomes. Accordingly, we posit:

Hypothesis 4 (H4): Credibility perceptions of the corrections delivered by the AI chatbot will mediate the interaction effects of the chatbot’s expertise cues and correction sidedness on (a) attitudes toward cultivated meat, (b) support for cultivated meat, and (c) intentions to consume cultivated meat.

Method

After receiving ethical approval (IRB No.: IRB-2021-998) from Nanyang Technological University, we conducted a 2 (chatbot expertise cues: high vs. low) × 2 (correction sidedness: one-sided vs. two-sided) between-subjects online experiment. The experiment was conducted from 3 April to 23 April 2024. We engaged a market research firm to recruit participants who were Singaporeans or permanent residents aged 21 years or older.

Participants

Among the 261 completed responses, seven were excluded due to straight-lining and speeding, leaving a final sample of 254 participants. The average age was 41.30 years (SD = 11.63). The sample was almost evenly split by gender, with 48.8% male and 51.2% female participants. Ethnically, 69.7% were Chinese, 25.2% Malay, 1.2% Indian, 2.4% Eurasian, and 1.8% from other ethnic groups. The most common education level was a bachelor’s degree or equivalent. The median household income ranged from SGD 11,000 to SGD 11,999.

Experiment Materials and Manipulations

The experimental materials include an introductory description of an AI-powered chatbot and a screenshot showing a conversation between the chatbot and a user, illustrating the process in which the AI chatbot corrects a piece of misinformation about cultivated meat.

Material Selection

To select the most appropriate piece of misinformation about cultivated meat for our main experiment, we conducted a pilot test involving 99 participants from a local university, including both students and staff. Participants were asked to evaluate seven circulating misinformation statements about cultivated meat on their familiarity and credibility, using a scale from 1 to 5. These false statements were sourced from reputable fact-checking sites and news websites (e.g., Reuters and The Associated Press). The statement claiming that “cultivated meat production contains a high quantity of antibiotics” was deemed the least familiar (M = 2.22) but most credible (M = 3.12). Therefore, it was chosen for further investigation in the main experiment.

The correction message designed to counter the selected misinformation about cultivated meat was based on research from Hanga et al. (2020) and Saied et al. (2023). These sources suggest that, unlike conventional meat, cultivated meat does not use large amounts of antibiotics. The message refuted the claim that “cultivated meat production contains a high quantity of antibiotics” by emphasizing three key points: (a) cultivated meat reduces the risk of antibiotic misuse prevalent in traditional farming, (b) the production process maintains antibiotic-free environments through strict sterility and control measures, and (c) cultivated meat is subject to thorough regulatory oversight to ensure adherence to stringent food safety standards and prevent antibiotic misuse. To ensure the scientific accuracy of the correction, two experts in bioengineering and cultivated meat were consulted to review, refine, and confirm the content’s accuracy.

Experimental Manipulations

We manipulated the chatbot’s expertise cues by altering its introductory description, following the approach of previous research (e.g., Wu et al., 2024). In the high-expertise cues condition, the chatbot was introduced as a fact-checking specialist focused on debunking misinformation about novel foods, developed with the expertise of the Singapore Food Agency and a local university. It sourced information from professional organizations, such as the World Health Organization, the Food and Drug Administration, the Centers for Disease Control and Prevention, and established fact-checking sites like Snopes.com. In contrast, the low-expertise cues condition presented the chatbot as a general AI tool for various tasks, including food recommendations, weather updates, and news, drawing information from a range of online sources, including social media platforms (e.g., Facebook, Twitter, TikTok), personal blogs, commercial sites, and search engines.

We also built upon existing studies (e.g., Wang & Huang, 2021; Xiao & Su, 2021) to manipulate the sidedness of the corrections. In the two-sided correction condition, the chatbot first acknowledged and then refuted the misinformation claim. It started by acknowledging the assertion that cultivated meat requires certain quantities of antibiotics for uncontaminated cell growth due to the lack of an immune system in the cells extracted for cultivating meat. This was followed by a strong refutation, arguing that such claims are false and misleading, and using the aforementioned three key arguments to support this refutation. In contrast, in the one-sided correction condition, the chatbot directly refuted the misinformation without acknowledging any aspect of the misleading claim, providing three supportive points in its rebuttal. To ensure consistency and avoid confounding variables, the introductory descriptions and correction messages were standardized in length across conditions, with introductions set at 102 words, and correction messages between 402 and 404 words. Further details on the materials used are available in the Appendix.

Procedure

In the main experiment, all participants who accessed our online study were first directed to read the study information sheet and provide electronic consent. They then reviewed brief definitions of cultivated meat, antibiotics, and AI-powered chatbots to ensure a basic understanding of the terms used in the experiment. Next, participants completed pre-test questions assessing their knowledge of AI, chatbots, and cultivated meat, as well as their trust in and prior attitudes toward AI and chatbots. Following this, they were randomly assigned to one of four conditions, each varying in chatbot expertise cues and correction sidedness. Each condition presented a static screenshot of an interaction between a user and the chatbot, demonstrating how the chatbot addresses a piece of misinformation about cultivated meat. Participants then completed attention-check questions, manipulation-check questions, and measures of the dependent variables. Finally, they provided demographic information and were debriefed upon completing the experiment.

Measures

Unless otherwise stated, all variables were measured using a scale from 1 (strongly disagree) to 7 (strongly agree). All measurements were adapted from previous research. Minor adjustments were made to tailor the items specifically for the context of this study. All items used for manipulation checks, dependent variables, and control variables, along with their references and descriptive statistics, are listed in Table 1.

Table 1.

Wording and Descriptive Statistics of All Items.

Categories	Variables	Items	M(SD)	Cronbach’s alpha	References
Manipulation checks	Chatbot expertise	(a) This chatbot is specially designed to correct novel food-related misinformation	5.00(1.42)	0.91	Koh & Sundar (2010)
		(b) This chatbot is specially engineered to debunk novel food-related misinformation
		(c) This chatbot is an expert in correcting novel food-related misinformation
		(d) This chatbot specializes in correcting novel food-related misinformation
	Correction sidedness	(a) The correction from the chatbot only gives opinions against the claim provided by the user (reverse-coded)	4.72(1.52)	0.76	De Veirman & Hudders (2020)
		(b) The correction from the chatbot gives opinions both in agreement and disagreement with the claim provided by the user
		(c) The correction from the chatbot is balanced
		(d) The correction from the chatbot mentions arguments in line with the claim provided by the user
Dependent variables	Perceived credibility of misinformation correction	The misinformation correction provided by the chatbot is . . .	5.12(1.34)	0.88	Besalú et al. (2021)
		(a) Accurate
		(b) Authentic
		(c) Believable
		(d) Objective
	Attitudes toward cultivated meat	To me, cultivated meat is . . .	5.05(1.59)	0.95	Ho et al. (2024)
		(a) Unhealthy/healthy
		(b) Disgusting/tasty
		(c) Unsafe/safe
		(d) Unnatural/natural
		(e) Unethical/ethical
		(f) Unappealing/appealing
	Support for cultivated meat	(a) I support the expansion of the cultivated meat market in Singapore	4.95(1.31)	0.94	Ho et al. (2024)
		(b) I support the research and development of cultivated meat in Singapore
		(c) I support the sale of cultivated meat in Singapore
		(d) I support government funding to develop cultivated meat in Singapore
		(e) I support funding by private companies to develop cultivated meat in Singapore
	Intentions to consume cultivated meat	(a) I intend to consume cultivated meat in the future	4.89(1.56)	0.94	Menozzi et al. (2020)
		(b) I will make an effort to consume cultivated meat in the future
		(c) I plan to consume cultivated meat in the future
Covariates	Perceived knowledge about AI	(a) I am well-informed about AI technology	4.51(1.79)	0.94	Yu et al. (2023)
		(b) Among my circle of friends, I am considered an expert on AI technology
		(c) Compared to most other people, I have more knowledge about AI technology
	Perceived knowledge about chatbots	(a) I am well-informed about chatbot	4.52(1.81)	0.94
		(b) Among my circle of friends, I am considered an expert on chatbot
		(c) Compared to most other people, I have more knowledge about chatbot
	Perceived knowledge about cultivated meat	(a) I am well-informed about cultivated meat	4.23(1.92)	0.96
		(b) Among my circle of friends, I am considered an expert on cultivated meat
		(c) Compared to most other people, I have more knowledge about cultivated meat
	Trust in AI	(a) AI technologies care about our well-being	4.91(1.93)	0.96	Choung et al. (2023)
		(b) AI technologies are sincerely concerned about addressing the problems of human users
		(c) AI technologies are to be helpful and do not operate out of selfish interest
		(d) AI technologies are truthful in their dealings
		(e) AI technologies keep their commitments and deliver on their promises
		(f) AI technologies are honest and do not abuse the information and advantage they have over their users
		(g) AI technologies work well
		(h) AI technologies have the features necessary to complete key tasks
		(i) AI is competent in their area of expertise
		(j) AI technologies are reliable
		(k) AI technologies are dependable
	Attitudes toward AI	I think the use of AI would be . . .	5.43(1.32)	0.94	Choung et al. (2023)
		(a) Good
		(b) Wise
		(c) Favorable
		(d) Beneficial
		(e) Positive

Note. M = mean; SD = standard deviation.

Data Analysis

We conducted a two-way univariate analysis of covariance (ANCOVA) to test H1–H3, examining the main and interaction effects of chatbot expertise and correction sidedness. For H4 which examined the mediation effects of perceived credibility of misinformation correction, we employed a moderated mediation analysis using Model 8 from the PROCESS macro of Hayes et al. (2017) in SPSS version 27.0. This analysis considered chatbot expertise as the independent variable, correction sidedness as the moderator, and perceived credibility of the correction as the mediator. Attitudes toward cultivated meat, support for cultivated meat, and intentions to consume cultivated meat were included as dependent variables. We performed the analysis with 95% confidence intervals and 5,000 bootstrapped samples. In addition, we controlled for age; education; knowledge about AI, chatbots, and cultivated meat; attitudes toward AI; and trust in AI in both the ANCOVA and the moderated mediation analysis.

Results

Manipulation Checks

Results from two independent samples t-tests indicated that chatbots with higher expertise cues were perceived as more professional (M = 5.30, SD = 1.01) in correcting misinformation on cultivated meat than chatbots with lower expertise cues (M = 4.70, SD = 1.39), t (252) = 15.705, p < .001. This indicated that participants exposed to the chatbot described as having greater expertise in correcting novel food-related misinformation perceived it as more professional in this role. In addition, participants who read a two-sided correction felt the correction both agreed with and refuted the argument in the misinformation more (M = 5.10, SD = .95) than those who read a one-sided correction (M = 4.11, SD = 1.62), t (252) = 69.23, p < .001. Therefore, the manipulations of chatbot expertise cues and correction sidedness were successful. Participants received points as compensation that they could redeem in the form of gift vouchers.

Test of Hypotheses

Results from the ANCOVA tests revealed that the main effects of chatbot expertise cues on credibility perceptions of misinformation correction were significant, (F(1, 242) = 32.89, p < .001, η_p² = .12), thereby supporting H1a. The chatbot with higher expertise cues (M = 5.38, SD = 0.88) elicited greater credibility perceptions of the correction than the chatbot with lower expertise cues (M = 4.86, SD = 1.24). However, the main effects of chatbot expertise cues on attitudes toward cultivated meat (M_{low-expertise cues} = 5.15, M_{high-expertise cues} = 4.94, F(1, 242) = .01, p = .98, η_p² = .00), support for cultivated meat (M_{low-expertise cues} = 5.00, M_{high-expertise cues} = 4.90, F(1, 242) = .18, p = .67, η_p² = .00), and intentions to consume cultivated meat (M_{low-expertise cues} = 4.92, M_{high-expertise cues} = 4.84, F(1, 242) = .86, p = .36, η_p² = .00) were not significant. Therefore, H1(b), (c), and (d) were not supported.

In addition, the main effects of correction sidedness on credibility perceptions of misinformation correction (M_one-sided = 5.20, M_two-sided = 5.03, F(1, 242) = .01, p = .93, η_p² = .00), attitudes toward cultivated meat (M_one-sided = 5.25, M_two-sided = 4.84, F(1, 242) = .48, p = .49, η_p² = .00), support for cultivated meat (M_one-sided = 5.12, M_two-sided = 4.77, F(1, 242) = .43, p = .51, η_p² = .00), and intentions to consume cultivated meat (M_one-sided = 4.98, M_two-sided = 4.77, F(1, 242) = 1.09, p = .30, η_p² = .00) were not significant. Therefore, H2(a)–H2(d) were rejected.

Moreover, there is a significant interaction effect between chatbot expertise cues and correction sidedness on credibility perceptions of misinformation correction, (F(1, 242) = 4.30, p = .039, η_p² = .02). A simple-effects analysis suggested a significant difference in credibility perceptions between high- and low-expertise chatbot conditions when the misinformation corrections employed one-sided framing (M_{low-expertise cues} = 5.03, M_{high-expertise cues} = 5.39, F(1, 242) = 6.84, p < .01, η_p² = .03). This difference was further amplified when the corrections featured two-sided framing (M_{high-expertise cues} = 5.37, M_{low-expertise cues} = 4.68, F(1, 242) = 29.18, p < .001, η_p² = .11), supporting H3(a). This synergistic interaction effect demonstrates that while high-expertise cues enhance credibility perceptions of misinformation correction in both one-sided and two-sided conditions, the effect is stronger with two-sided framing. This suggests that the combination of high expertise and two-sided framing is most effective for maximizing credibility perceptions in misinformation-correction contexts.

This significant synergistic interaction effect between chatbot expertise cues and correction sidedness was plotted in Figure 1. However, the interaction effects between chatbot expertise cues and correction sidedness on attitudes toward cultivated meat (F(1, 242) = .03, p = .86, η_p² = .00), support for cultivated meat (F(1, 242) = .01, p = .94, η_p² = .00), and intentions to consume cultivated meat (M_one-sided = 4.98, M_two-sided = 4.77, F(1, 242) = .02, p = .88, η_p² = .00) were not significant, addressing H3(b) and (c).

Figure 1.

The Significant Interaction Between Chatbot Expertise and Message Sidedness on Perceived Credibility of Misinformation Correction.

Finally, the moderated mediated analysis demonstrated that the moderated mediation on attitudes (β = .11, BootSE = .06, 95% CI [.01, .24]) and consumption intentions (β = .15. BootSE = .09, 95% CI [.01, .35]) through credibility perceptions is statistically significant. Specifically, credibility perceptions of correction mediated the influence of chatbot expertise cues on attitudes (β = .08, BootSE = .02, 95% CI [.02, .16]) and consumption intentions (β = .11, BootSE = .06, 95% CI [.01, .25]) when the correction was one-sided. This mediation effect was stronger for attitudes (β = .16, SE = .04, 95% CI [.02, .29]) and consumption intentions (β = .25, BootSE = .11, 95% CI [.06, .50]) when the correction was two-sided. Hence, H4(a) and H4(c) were supported. However, the moderated mediation effect on support for cultivated meat through credibility perception of correction was not significant (β = .05, BootSE = .05, 95% CI [−.02, .17]), thereby rejecting H4(b). Table 2 shows the results of the moderated mediation analysis.

Table 2.

Results of Moderated Mediation Effects.

Mediator	Outcome variables	Moderated mediation effects
Mediator	Outcome variables	β (BootSE)	95% CI
Perceived credibility of correction	Attitude toward cultivated meat	.11 (.06)	[.01, .24]
	Support for cultivated meat	.05 (.05)	[−.02, .17]
	Intention to consume cultivated meat	.15 (.09)	[.01, .35]

Note. β = unstandardized regression coefficients.

Discussion

The widespread misinformation requires robust technology-driven tools for effective debunking. Recent advancements in AI have significantly improved AI-powered chatbots, making them viable tools for information verification and fact-checking (Xiao et al., 2023). Informed by the CASA paradigm and message sidedness theory, this study examines how chatbot features like expertise cues and the framing of outputs (i.e., correction sidedness) influence the effectiveness of misinformation correction and individuals’ attitudinal and behavioral response to emerging technologies such as cultivated meat. We found that chatbot expertise cues, more than sidedness of the corrective content, significantly enhance credibility perceptions of misinformation correction. In addition, the interplay between chatbot expertise cues and correction sidedness affects credibility perceptions and, in turn, shapes attitudes and consumption intentions toward cultivated meat.

Corrections from chatbots with high-expertise cues in addressing novel-food misinformation were perceived as more credible than those from chatbots with low-expertise cues, aligning with research on source expertise and credibility (Flanagin et al., 2020; Ou & Ho, 2024a). However, unlike prior studies linking expertise to attitudinal and behavioral changes (Rashedi & Seyed Siahi, 2020; Yoon et al., 1998), chatbot expertise did not significantly affect attitudes or intentions to consume cultivated meat. This discrepancy may stem from the complexity and novelty of cultivated meat, prompting individuals to rely more on pre-existing beliefs or skepticism. Factors like personal food safety beliefs (Bryant & Barnett, 2020), perceptions of naturalness (Fidder & Graça, 2023), and media habits (Ho et al., 2024) likely play stronger roles in shaping attitudes and intentions.

The findings show that correction sidedness—whether one-sided or two-sided—did not significantly impact participants’ perceptions of chatbot credibility, attitudes, or intentions regarding cultivated meat. This aligns with Wang and Huang’s (2021) findings in the context of e-cigarettes. The lack of effect suggests that participants may prioritize source expertise over message structure, particularly for novel or complex topics where they lack domain knowledge (Lucassen et al., 2013; Wang & Huang, 2021). For misinformation correction on such topics, framing corrections as two-sided alone is insufficient to enhance credibility; the chatbot’s perceived expertise remains critical. Users also need the assurance of the chatbot’s expertise to view the two-sided correction as trustworthy. The complexity of two-sided corrections, which present both sides of an issue, may lead to cognitive overload, requiring substantial mental effort to process (Deck & Jahedi, 2015; Sweller, 2023). For less invested or knowledgeable participants, the absence of expertise cues can hinder efficient information processing, reducing engagement and perceived credibility. This highlights that the cognitive load from two-sided corrections can limit their effectiveness, emphasizing the need for both message framing and clear expertise cues to enhance correction credibility.

Our findings reveal a synergistic interaction effect between correction sidedness and chatbot expertise cues on the credibility perceptions of corrections provided by the chatbot. A significant difference in credibility perceptions was observed between high- and low-chatbot expertise conditions when corrections employed one-sided framing. This difference was further amplified when the corrections featured two-sided framing. This suggests that the high congruence between source expertise and the comprehensive nature of the message content (i.e., two-sidedness) can amplify credibility perceptions in human-AI interactions. This may be because users typically expect expert sources to provide comprehensive and balanced information (e.g., Manninen, 2020). Chatbots with low-expertise cues might not be expected to handle complex, two-sided arguments as effectively, which can lead to reduced credibility in both one-sided and two-sided conditions. This aligns with previous findings that people place greater trust in two-sided messages provided by experts than in one-sided messages from novices (Flanagin et al., 2020).

Credibility perceptions of misinformation corrections delivered by chatbots act as a crucial mediator between chatbot expertise cues and correction sidedness and people’s attitudes and consumption intentions toward cultivated meat. This effect can be explained by cognitive dissonance theory, which suggests that individuals strive for consistency in their beliefs and attitudes (Cooper, 2012). Consequently, individuals may adjust their attitudes and behaviors to better align with their credibility perceptions of information provided by the chatbot, potentially enhancing their view of and willingness to consume cultivated meat (Singh & Banerjee, 2018). This indicates that effective AI communication, which presents expert sources and balanced messages, can significantly impact individuals’ attitudes and behaviors through people’s credibility perceptions.

Theoretical Implications

This study integrates the CASA paradigm and message sidedness theory to examine how AI features and output framing shape cognitive, attitudinal, and behavioral responses, bridging science communication, human-AI interaction, and misinformation correction. CASA highlights the role of chatbot expertise in enhancing credibility, while the message sidedness theory explains the persuasive value of two-sided messages. The findings reveal that combining expertise cues with two-sided corrections enhances credibility perceptions and indirectly influences attitudes and behaviors toward cultivated meat. This underscores how AI chatbots can emulate effective human communication strategies to establish credibility, especially in misinformation correction.

We offered new insights into how the features of misinformation correction and AI agency attributes influence persuasive outcomes through credibility judgments, rather than direct impact. By revealing that credibility judgments mediate the relationship between chatbot features (source attributes and message structure) and audience responses, we highlight the complex role that credibility perceptions play in the persuasion process. This contributes to existing literature by suggesting that the effectiveness of misinformation correction is not just about the direct transmission of facts but also about how these facts are framed and by whom. Existing research predominantly focuses on how corrections reduce belief in and sharing of misinformation (e.g., Walter et al., 2020). Our study broadens this by exploring how AI-facilitated misinformation correction affects persuasion, specifically impacting people’s attitudes and behavioral intentions through trust in AI-provided corrections. This enhances the literature on misinformation correction, AI-mediated communication, and persuasion, highlighting the importance of designing AI-driven communication tools that effectively convey information and boost perceived credibility to significantly influence audience attitudes and behaviors.

The results also highlight that message sidedness amplifies the impact of expertise cues on credibility perceptions in human-AI interactions. Nonetheless, message sidedness alone does not substantially affect credibility judgments or persuasive outcomes. While traditionally effective in enhancing credibility and persuasion, applying message sidedness directly to AI communications proves challenging, highlighting the complexities of adapting this theory from human-human to human-AI interactions. Moreover, unlike prior studies like Cheng et al. (2024), which relied on surveys to assess attitudes toward COVID-19 misinformation correction by chatbots, our experimental approach enhances robustness, providing a stronger foundation for causal conclusions regarding the effects of chatbot features on the effectiveness of misinformation correction.

Practical Implications

This study offers practical insights for designing AI chatbots to combat scientific misinformation. Enhancing a chatbot’s specialization and sourcing information from credible, authoritative sources can increase the effectiveness of misinformation correction. For example, including clickable sources for easy verification and clarifying the origins of the correction can strengthen AI-driven debunking efforts. The observed synergistic effects between chatbot expertise and correction sidedness suggest that two-sided corrections from specialized chatbots foster higher trust, more favorable attitudes, and positive behavioral intentions toward scientific topics. Therefore, misinformation-correction chatbots could benefit from integrating expertise cues with two-sided correction framing in a complementary manner. Specifically, the chatbot should clearly communicate its expertise in a field related to the misinformation topic while simultaneously implementing two-sided correction framing. For instance, when addressing misinformation about cultivated meat, the chatbot could introduce itself as an expert developed in collaboration with reputable food science organizations or research institutions. It should then present the misinformation or common misconceptions first, signaling its understanding of opposing perspectives, followed by a balanced, research-based correction.

Limitations and Future Directions

This study has several limitations that suggest directions for future research. First, it focused solely on chatbot expertise cues and correction sidedness within the context of cultivated meat. Future research could extend to other areas of science affected by misinformation to determine if similar effects occur. Second, despite providing an introduction to cultivated meat and controlling for the effects of knowledge and attitudes toward it, individual responses to misinformation and corrections may vary, especially between those who have and have not tried cultivated meat. Further studies should investigate these differences and adjust for their influence. Third, our method involved using static screenshots of interactions between a user and the chatbot, which did not allow participants to interact with the chatbot in real time and may affect the generalizability of our findings to actual interactions. Building on prior research examining the effectiveness of misinformation correction (Aikin et al., 2017; Paquin et al., 2022), as well as its impact on user engagement (Giombi et al., 2022) and behavioral responses (e.g., Aikin et al., 2019), future studies could incorporate live interactions to enhance external validity. Finally, this study selected cultivated meat as a representative case of emerging food technology, given its significant public impact in Singapore and susceptibility to misinformation. However, this focus may limit the broader applicability of the findings. Future research could extend this work to other emerging technologies, such as plant-based proteins, GM foods, or other novel food innovations.

Footnotes

Appendix

Acknowledgements

The authors would like to express our appreciation to Dr. Larry Loo and Mr. Yuxuan Tang for their professional suggestions and insightful comments on our misinformation correction materials.

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This study was supported by the Ministry of Education Academic Research Fund Tier 1 Grant (grant number: RT16/20)

ORCID iDs

Mengxue Ou

Shirley S. Ho

Author Biographies

Mengxue Ou (MA, Nanyang Technological University, Singapore) is currently a PhD candidate in the Wee Kim Wee School of Communication and Information at Nanyang Technological University, Singapore. Her research focuses on science and health communication, with a particular interest in investigating how individuals process misinformation in the context of emerging controversial scientific issues.

Shirley S. Ho (PhD, University of Wisconsin–Madison, USA) is the President’s Chair Professor in Communication Studies in the Wee Kim Wee School of Communication and Information, as well as the Associate Vice President (Humanities, Social Sciences, & Research Communication) in the President’s Office at Nanyang Technological University, Singapore. Her research focuses on science communication, in which she investigates cross-cultural public opinion dynamics related to science and technology, with potential health or environmental impacts. She is an elected fellow of the International Communication Association.

Stanley A. Wijaya (MA, National University of Singapore, Singapore) is a Research Associate at the Wee Kim Wee School of Communication and Information at Nanyang Technological University, Singapore. His research interests are in cross-cultural practices in new media.

References

Aikin

K. J.

Betts

K. R.

Southwell

B. G.

Boudewyns

Johnson

Breslav

A. D. S.

(2019). Consumer reactions to price comparison and disclosure information in prescription drug print advertising. Journal of Consumer Affairs, 53(3), 1255–1272. https://doi.org/10.1111/joca.12227

Aikin

K. J.

Southwell

B. G.

Paquin

R. S.

Rupert

D. J.

O’Donoghue

A. C.

Betts

K. R.

Lee

P. K.

(2017). Correction of misleading information in prescription drug television advertising: The roles of advertisement similarity and time delay. Research in Social and Administrative Pharmacy, 13(2), 378–388. https://doi.org/10.1016/j.sapharm.2016.04.004

Allen

(1991). Meta-analysis comparing the persuasiveness of one-sided and two-sided messages. Western Journal of Speech Communication, 55(4), 390–404. https://doi.org/10.1080/10570319109374395

Altay

Schwartz

Hacquin

A.-S.

Allard

Blancke

Mercier

(2022). Scaling up interactive argumentation by providing counterarguments with a chatbot. Nature Human Behaviour, 6(4), 579–592. https://doi.org/10.1038/s41562-021-01271-w

Amiri

Karahanna

(2022). Chatbot use cases in the Covid-19 public health response. Journal of the American Medical Informatics Association, 29(5), 1000–1010. https://doi.org/10.1093/jamia/ocac014

AP News. (n.d.). Animal cells used to create lab-grown meat are not cancerous, experts say. Retrieved February 1, 2025, from https://apnews.com/article/fact-check-lab-meat-cancer-animal-cells-449786524119

Aslett

Sanderson

Godel

Persily

Nagler

Tucker

J. A.

(2024). Online searches to evaluate misinformation can increase its perceived veracity. Nature, 625(7995), 548–556. https://doi.org/10.1038/s41586-023-06883-y

Besalú

Pont-Sorribes

Martí

(2021). Perceived credibility of tweets by opinion leaders during the COVID-19 pandemic in Spain. International Journal of Communication, 15(2021), 5158–5185.

Bryant

Barnett

(2020). Consumer acceptance of cultured meat: An updated review (2018–2020). Applied Sciences, 10(15), Article 5201. https://doi.org/10.3390/app10155201

10.

Chambers

A. V.

Baker

M. T.

Leggette

H. R.

Osburn

W. N.

(2023). Effects of message framing and information source on consumers’ attitudes toward an amino acid-based alternative meat curing system. Foods, 12(7), Article 1535. https://doi.org/10.3390/foods12071535

11.

Chen

M.-Y.

(2016). Can two-sided messages increase the helpfulness of online reviews? Online Information Review, 40(3), 316–332. https://doi.org/10.1108/OIR-07-2015-0225

12.

Cheng

Wang

Lee

(2024). Using a chatbot to combat misinformation: exploring gratifications, chatbot satisfaction and engagement, and relationship quality. International Journal of Human–Computer Interaction. Advance online publication. https://doi.org/10.1080/10447318.2024.2344149

13.

Cheung

J. C.

S. S.

(2024). Explainable AI and trust: How news media shapes public support for AI-powered autonomous passenger drones. Public Understanding of Science. Advance online publication. https://doi.org/10.1177/09636625241291192

14.

Choung

David

Ross

(2023). Trust in AI and Its Role in the Acceptance of AI Technologies. International Journal of Human–Computer Interaction, 39(9), 1727–1739. https://doi.org/10.1080/10447318.2022.2050543

15.

Cooper

(2012). Cognitive dissonance theory. In Van Lange

Kruglanski

Higgins

(Eds.), Handbook of theories of social psychology: Volume, 1 (pp. 377–397). Sage. https://doi.org/10.4135/9781446249215.n19

16.

Costello

T. H.

Pennycook

Rand

D. G.

(2024). Durably reducing conspiracy beliefs through dialogues with AI. Science, 385(6714), eadq1814. https://doi.org/10.1126/science.adq1814

17.

Deck

Jahedi

(2015). The effect of cognitive load on economic decision making: A survey and new experiments. European Economic Review, 78, 97–119. https://doi.org/10.1016/j.euroecorev.2015.05.004

18.

Demartini

Mizzaro

Spina

(2020). Human-in-the-loop artificial intelligence for fighting online misinformation: Challenges and opportunities—UQ eSpace. Retrieved July 6, 2024, from https://espace.library.uq.edu.au/view/UQ:9b74702

19.

De Veirman

Hudders

(2020). Disclosing sponsored Instagram posts: The role of material connection with the brand and message-sidedness when disclosing covert advertising. International Journal of Advertising, 39(1), 94–130. https://doi.org/10.1080/02650487.2019.1575108

20.

Fidder

Graça

(2023). Aligning cultivated meat with conventional meat consumption practices increases expected tastefulness, naturalness, and familiarity. Food Quality and Preference, 109, Article 104911. https://doi.org/10.1016/j.foodqual.2023.104911

21.

Flanagin

A. J.

Winter

Metzger

M. J.

(2020). Making sense of credibility in complex information environments: The role of message sidedness, information source, and thinking styles in credibility evaluation online. Information, Communication & Society, 23(7), 1038–1056. https://doi.org/10.1080/1369118X.2018.1547411

22.

Giombi

Viator

Hoover

Tzeng

Sullivan

H. W.

O’Donoghue

A. C.

Southwell

B. G.

Kahwati

L. C.

(2022). The impact of interactive advertising on consumer engagement, recall, and understanding: A scoping systematic review for informing regulatory science. PLOS ONE, 17(2), Article e0263339. https://doi.org/10.1371/journal.pone.0263339

23.

Jung

E. H.

(2014). The effects of source cues on online news perception. Computers in Human Behavior, 38, 358–367. https://doi.org/10.1016/j.chb.2014.05.044

24.

Hanga

M. P.

Ali

Moutsatsou

De La Raga

F. A.

Hewitt

C. J.

Nienow

Wall

(2020). Bioprocess development for scalable production of cultivated meat. Biotechnology and Bioengineering, 117(10), 3029–3039. https://doi.org/10.1002/bit.27469

25.

Hayes

A. F.

Montoya

A. K.

Rockwood

N. J.

(2017). The analysis of mechanisms and their contingencies: PROCESS versus structural equation modeling. Australasian Marketing Journal, 25(1), 76–81. https://doi.org/10.1016/j.ausmj.2017.02.001

26.

S. S.

Cheung

J. C.

(2024). Trust in artificial intelligence, trust in engineers, and news media: Factors shaping public perceptions of autonomous drones through UTAUT2. Technology in Society, 77, Article 102533. https://doi.org/10.1016/j.techsoc.2024.102533

27.

S. S.

Ong

Z. T.

(2024). Examining public perceptions of cultivated meat in Singapore: Food neophobia and neophilia as precursors to the influence of presumed media influence model. Asian Journal of Communication, 34, 457–478. https://doi.org/10.1080/01292986.2024.2342827

28.

Ives

(2020, December 2). Singapore approves a lab-grown meat product, a global first. The New York Times. https://www.nytimes.com/2020/12/02/business/singapore-lab-meat.html

29.

Jäger

A.-K.

Weber

(2020). Can you believe it? The effects of benefit type versus construal level on advertisement credibility and purchase intention for organic food. Journal of Cleaner Production, 257, Article 120543. https://doi.org/10.1016/j.jclepro.2020.120543

30.

Kamins

M. A.

Brand

M. J.

Hoeke

S. A.

Moe

J. C.

(1989). Two-sided versus one-sided celebrity endorsements: The impact on advertising effectiveness and credibility. Journal of Advertising, 18(2), 4–10. https://doi.org/10.1080/00913367.1989.10673146

31.

Koh

Y. J.

Sundar

S. S.

(2010). Heuristic versus systematic processing of specialist versus generalist sources in online media. Human Communication Research, 36(2), 103–124. https://doi.org/10.1111/j.1468-2958.2010.01370.x

32.

Lewandowsky

Van Der Linden

(2021). Countering misinformation and fake news through inoculation and prebunking. European Review of Social Psychology, 32(2), 348–384. https://doi.org/10.1080/10463283.2021.1876983

33.

Liew

T. W.

Tan

S.-M.

(2021). Social cues and implications for designing expert and competent artificial agents: A systematic review. Telematics and Informatics, 65, Article 101721. https://doi.org/10.1016/j.tele.2021.101721

34.

Lin

Kim

(2022). The effects of gain-loss framing and message sidedness on GMO acceptance: The mediating role of psychological reactance. Asian Communication Research, 19(2), 70–91. https://doi.org/10.20879/acr.2022.19.2.70

35.

Liu

Yan

Lai

Y. L.

(2022). Effects of personalization and source expertise on users’ health beliefs and usage intention toward health chatbots: Evidence from an online experiment. Digital Health, 8, Article 205520762211297. https://doi.org/10.1177/20552076221129718

36.

Lucassen

Muilwijk

Noordzij

M. L.

Schraagen

J. M.

(2013). Topic familiarity and information skills in online credibility evaluation. Journal of the American Society for Information Science and Technology, 64(2), 254–264. https://doi.org/10.1002/asi.22743

37.

Manninen

V. J. E.

(2020). If only they knew: Audience expectations and actual sourcing practices in online journalism. Journalism Practice, 14(4), 412–428. https://doi.org/10.1080/17512786.2019.1625719

38.

McGuire

W. J.

(1964). Some contemporary approaches. In Berkowitz

(Ed.), Advances in experimental social psychology (Vol, 1) (pp. 191–229). Elsevier. https://doi.org/10.1016/S0065-2601(08)60052-0

39.

Menozzi

Nguyen

T. T.

Sogari

Taskov

Lucas

Castro-Rial

J. L. S.

Mora

(2020). Consumers’ Preferences and Willingness to Pay for Fish Products with Health and Environmental Labels: Evidence from Five European Countries. Nutrients, 12(9), 2650. https://doi.org/10.3390/nu12092650

40.

Metzger

M. J.

Flanagin

A. J.

Eyal

Lemus

D. R.

Mccann

R. M.

(2003). Credibility for the 21st century: Integrating perspectives on source, message, and media credibility in the contemporary media environment. Annals of the International Communication Association, 27(1), 293–335. https://doi.org/10.1080/23808985.2003.11679029

41.

Monteith

Glenn

Geddes

J. R.

Whybrow

P. C.

Achtyes

Bauer

(2024). Artificial intelligence and increasing misinformation. The British Journal of Psychiatry, 224(2), 33–35. https://doi.org/10.1192/bjp.2023.136

42.

Moon

W.-K.

Chung

Jones-Jang

S. M.

(2023). How can we fight partisan biases in the COVID-19 pandemic? AI source labels on fact-checking messages reduce motivated reasoning. Mass Communication and Society, 26(4), 646–670. https://doi.org/10.1080/15205436.2022.2097926

43.

Mueller

(n.d.). Lab—grown meat being sold in US will be labeled “cell—cultivated”—Fact check. USA TODAY. Retrieved July 6, 2024, from https://www.usatoday.com/story/news/factcheck/2023/07/18/false-claim-lab-grown-meat-can-be-sold-without-a-label-fact-check/70414781007/

44.

Nass

Moon

(2000). Machines and mindlessness: Social responses to computers. Journal of Social Issues, 56(1), 81–103. https://doi.org/10.1111/0022-4537.00153

45.

S. S.

(2024a). Factors associated with information credibility perceptions: A meta-analysis. Journalism & Mass Communication Quarterly, 101(2), 346–372. https://doi.org/10.1177/10776990231222556

46.

S. S.

(2024b). Does knowledge make a difference? Understanding how the lay public and experts assess the credibility of information on novel foods. Public Understanding of Science, 33(2), 241–259. https://doi.org/10.1177/096366252311913

47.

Paquin

R. S.

Boudewyns

Betts

K. R.

Johnson

O’Donoghue

A. C.

Southwell

B. G.

(2022). An empirical procedure to evaluate misinformation rejection and deception in mediated communication contexts. Communication Theory, 32(1), 25–47. https://doi.org/10.1093/ct/qtab011

48.

Rashedi

Seyed Siahi

S. F.

(2020, January 2). Role of source trustworthiness, experience sources and source expertise in the formation of WOM and purchase intention (case study of Tehran mobile market customers). SSRN Electronic Journal. https://doi.org/10.2139/ssrn.3513095

49.

Reuters. (n.d.). Fact check: Video creator’s food art mislabeled as Bill Gates’ lab-grown meat. Retrieved July 6, 2024, from https://www.reuters.com/fact-check/video-creators-food-art-mislabeled-bill-gates-lab-grown-meat-2023-11-28/

50.

Saied

A. A.

Chandran

Chopra

Dey

Emran

T. B.

Dhama

(2023). Cultivated meat could aid in reducing global antimicrobial resistance burden—producing meat without antibiotics as a safer food system for the future. International Journal of Surgery, 109(2), 189–190. https://doi.org/10.1097/JS9.0000000000000199

51.

Singh

R. P.

Banerjee

(2018). Exploring the influence of celebrity credibility on brand attitude, advertisement attitude and purchase intention. Global Business Review, 19(6), 1622–1639. https://doi.org/10.1177/0972150918794974

52.

Southwell

B. G.

Brennen

J. S. B.

Paquin

Boudewyns

Zeng

(2022). Defining and measuring scientific misinformation. The Annals of the American Academy of Political and Social Science, 700(1), 98–111. https://doi.org/10.1177/00027162221084709

53.

Sun

Pan

(2024). Self-correction or other-correction: The effects of source consistency and ways of correction on sharing intention of health misinformation correction. Health Communication. Advance online publication. https://doi.org/10.1080/10410236.2024.2346674

54.

Sundar

S. S.

(2008). The MAIN model: A heuristic approach to understanding technology effects on credibility. In Metzger

M. J.

Flanagin

A. J.

(Eds.), Digital media, youth, and credibility (pp. 73–100). MIT Press.

55.

Sweller

(2023). The development of cognitive load theory: Replication crises and incorporation of other theories can lead to theory expansion. Educational Psychology Review, 35(4), Article 95. https://doi.org/10.1007/s10648-023-09817-2

56.

Treich

(2021). Cultured meat: Promises and challenges. Environmental and Resource Economics, 79(1), 33–61. https://doi.org/10.1007/s10640-021-00551-3

57.

Uribe

Buzeta

Velásquez

(2016). Sidedness, commercial intent and expertise in blog advertising. Journal of Business Research, 69(10), 4403–4410. https://doi.org/10.1016/j.jbusres.2016.04.102

58.

Vraga

E. K.

Bode

(2017). Using expert sources to correct health misinformation in social media. Science Communication, 39(5), 621–645. https://doi.org/10.1177/1075547017731776

59.

Walter

Cohen

Holbert

R. L.

Morag

(2020). Fact-Checking: A Meta-Analysis of What Works and for Whom. Political Communication, 37(3), 350–375. https://doi.org/10.1080/10584609.2019.1668894

60.

Wang

Peng

(2023). Striking an emotional chord: Effects of emotional appeals and chatbot anthropomorphism on persuasive science communication. Science Communication, 45(4), 485–511. https://doi.org/10.1177/10755470231194583

61.

Wang

Huang

(2021). Countering the “Harmless E-Cigarette” myth: The interplay of message format, message sidedness, and prior experience with E-cigarette use in misinformation correction. Science Communication, 43(2), 170–198. https://doi.org/10.1177/1075547020974384

62.

Wang

(2021). Debunking misinformation about genetically modified food safety on social media: Can heuristic cues mitigate biased assimilation? Science Communication, 43(4), 460–485. https://doi.org/10.1177/10755470211022024

63.

Wang

(2024). Engaging with misinformation and misinformation corrective messages on social media: Examining the role of source cues, social endorsement cues, and prior attitudes. Atlantic Journal of Communication, 32(4), 499–511. https://doi.org/10.1080/15456870.2023.2291200

64.

Winter

Krämer

N. C.

(2012). Selecting science information in Web 2.0: How source cues, message sidedness, and need for cognition influence users’ exposure to blog posts. Journal of Computer-Mediated Communication, 18(1), 80–96. https://doi.org/10.1111/j.1083-6101.2012.01596.x

65.

Wolfe

Mitra

(2024, June 3–6). The impact and opportunities of generative AI in fact-checking [Conference session]. The 2024 ACM Conference on Fairness, Accountability, and Transparency, Rio de Janeiro, Brazil, pp. 1531–1543. https://doi.org/10.1145/3630106.3658987

66.

Kim

K. J.

Mou

(2024). Minority social influence and moral decision-making in human–AI interaction: The effects of identity and specialization cues. New Media & Society, 26, 5619–5637. https://doi.org/10.1177/14614448221138072

67.

Xiao

(2021). Integrating reasoned action approach and message sidedness in the era of misinformation: The case of HPV vaccination promotion. Journal of Health Communication, 26, 371–380. https://doi.org/10.1080/10810730.2021.1950873

68.

Xiao

Liao

Q. V.

Zhou

Grandison

(2023, March 27–31). Powering an AI chatbot with expert sourcing to support credible health information access [Conference session]. Proceedings of the 28th International Conference on Intelligent User Interfaces, Sydney, NSW, Australia, pp. 2–18. https://doi.org/10.1145/3581641.3584031

69.

Yang

Zheng

Qiu

Wang

(2024). The role of recommendation sources and attribute framing in online product recommendations. Journal of Business Research, 174, Article 114498. https://doi.org/10.1016/j.jbusres.2024.114498

70.

Yoon

Kim

C. H.

Kim

M.-S.

(1998). A cross-cultural comparison of the effects of source credibility on attitudes and behavioral intentions. Mass Communication and Society, 1(3–4), 153–173. https://doi.org/10.1080/15205436.1998.9677854

71.

Yuan

Besley

J. C.

(2019). Should scientists talk about GMOs nicely? Exploring the effects of communication styles, source expertise, and preexisting attitude. Science Communication, 41(3), 267–290. https://doi.org/10.1177/1075547019837623

72.

Fan

(2023). Employees’ Appraisals and Trust of Artificial Intelligences’ Transparency and Opacity. Behavioral Sciences, 13(4), 344. https://doi.org/10.3390/bs13040344

73.

Zhao

Chen

Jin

Zhang

(2023). Comparing button-based chatbots with webpages for presenting fact-checking results: A case study of health information. Information Processing & Management, 60(2), Article 103203. https://doi.org/10.1016/j.ipm.2022.103203

74.

Zhou

Sharma

Zhang

A. X.

Althoff

(2024). Correcting misinformation on social media with a large language model (Version 3).arXiv. https://doi.org/10.48550/ARXIV.2403.11169

Harnessing AI to Address Misinformation on Cultivated Meat: The Impact of Chatbot Expertise and Correction Sidedness

Abstract

Keywords

Literature Review

Study Context: Scientific Misinformation on Cultivated Meat

AI Chatbot for Misinformation Correction

Chatbot Expertise Cues

Effects of Correction Sidedness

The Interaction Between Chatbot Expertise Cues and Correction Sidedness

The Mediating Role of Correction Credibility

Method

Participants

Experiment Materials and Manipulations

Material Selection

Experimental Manipulations

Procedure

Measures

Data Analysis

Results

Manipulation Checks

Test of Hypotheses

Discussion

Theoretical Implications

Practical Implications

Limitations and Future Directions

Footnotes

Appendix

Acknowledgements

Declaration of Conflicting Interests

Funding

ORCID iDs

Author Biographies

References