Journalists’ Perceptions of Artificial Intelligence and Disinformation Risks

Academic Editors: Mathias Felipe DeLima-Santos and Adeola Abdulateef Elega Received: 30 June 2025 Revised: 13 August 2025 Accepted: 14 August 2025 Published: 30 August 2025 Citation: Peña-Alonso, U., Peña-Fernández, S., & Meso-Ayerdi, K. (2025). Journalists’ Perceptions of Artificial Intelligence and Disinformation Risks. Journalism and Media,6(3), 133. https://doi.org/ 10.3390/journalmedia6030133 Copyright: © 2025 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https://creativecommons.org/ licenses/by/4.0/). Article Journalists’ Perceptions of Artificial Intelligence and Disinformation Risks Urko Peña-Alonso *, Simón Peña-Fernández and Koldobika Meso-Ayerdi Department of Journalism, University of the Basque Country (UPV/EHU), 48940 Leioa, Spain; [email protected] (S.P.-F.); [email protected] (K.M.-A.) *Correspondence: [email protected] Abstract This study examines journalists’ perceptions of the impact of artificial intelligence (AI) on disinformation, a growing concern in journalism due to the rapid expansion of generative AI and its influence on news production and media organizations. Using a quantitative approach, a structured survey was administered to 504 journalists in the Basque Country, identified through official media directories and with the support of the Basque Association of Journalists. This survey, conducted online and via telephone between May and June 2024, included questions on sociodemographic and professional variables, as well as attitudes toward AI’s impact on journalism. The results indicate that a large majority of journalists (89.88%) believe AI will considerably or significantly increase the risks of disinformation, and this perception is consistent across genders and media types, but more pronounced among those with greater professional experience. Statistical analyses reveal a significant association between years of experience and perceived risk, and between AI use and risk perception. The main risks identified are the difficulty in detecting false content and deepfakes, and the risk of obtaining inaccurate or erroneous data. Co-occurrence analysis shows that these risks are often perceived as interconnected. These findings highlight the complex and multifaceted concerns of journalists regarding AI’s role in the information ecosystem. Keywords: artificial intelligence; risk perception; journalism; disinformation; news media 1. Introduction In the past decade, disinformation has emerged as one of the main concerns in the field of journalism (Cea & Palomo,2021). This phenomenon has gained a relevance comparable to the exponential increase in the strategic dimension of journalistic practice, driven by the expansion of artificial intelligence (Palomo et al.,2023;Verma,2024;Kevin-Alerechi et al.,2025). Thus, the advances in artificial intelligence (AI) have surprised us in recent years due to their speed and their impact on almost every field. Generative AI, which marked a turning point in the development of this technology, is particularly relevant in content creation spaces (Gonzalo,2024;Newman et al.,2024). In the field of journalism, it is increasingly evident that AI is beginning to make inroads transversally in the news production process, as well as in the structure and functioning of media organizations (Túñez López et al.,2021;Ufarte Ruiz et al.,2023;Londoño-Proaño & Buele,2025). The predictions of those who foresaw that AI would ultimately determine the future development of media transformation seem to be coming true (Zheng et al.,2018;Cui,2025). Journal. Media 2025,6, 133 https://doi.org/10.3390/journalmedia6030133 Journal. Media 2025,6, 133 2 of 15 Peña-Fernández et al. (2023) state that, although AI brings benefits to journalism, its adoption entails ethical challenges that must be proactively addressed. Its emergence has rekindled the debate regarding the ambivalent nature of technological developments (Brennen et al.,2018), that is, the need to assess the balance between their potential benefits and the new insecurities they introduce (Vrabiˇc Dežman,2024;Wei et al.,2024;Schaetz et al.,2025). It is evident that the use of artificial intelligence and its application to journalism reveal significant opportunities for improvement in the efficiency of news production and distribution, but also the threats posed by its use to undermine quality journalism (Mateos Abarca & Gamonal Arroyo,2024) or to disseminate disinformation on a large scale (Astobiza,2024). These challenges are compounded by concerns regarding the veracity of content, ethics and transparency in data usage, intrusion into private life, and its contribution to reinforcing existing social divides (Hansen et al.,2017;Brundage et al., 2018;Pihlajarinne & Alén-Savikko,2022). Media professionals themselves are not immune to this debate. On the one hand, AI can assist journalists by freeing them from the most routine and low value-added tasks (Wu et al.,2019;Cools & Diakopoulos,2024;Cools & de Vreese,2025;García de Torres et al.,2025), which could help strengthen the more cognitive aspects of the profession, such as creativity, critical thinking, curiosity, or skepticism (Thurman et al.,2017;Guzman & Lewis,2020;Moravec et al.,2024). For media organizations, AI may serve as the key to accessing business models that improve their economic viability (Lindén & Tuulonen,2019; Kotenidis & Veglis,2021;Oh & Jung,2025) or help them reduce costs (Wilczek et al.,2024). However, the main reservations regarding the implementation of AI come from the workers themselves, who see it more as a threat than as a complement (Mondría Terol,2023; van Dalen,2024). In fact, some studies conclude that media managers adopt automation as a strategy aimed primarily at reducing costs and improving productivity, resulting in a reduction in the number of employees (Kim & Kim,2017;Rick & Hanitzsch,2023; Thäsler-Kordonouri & Barling,2023) or an increase in job insecurity (López Jiménez & Ouariachi,2020;Ananny & Karr,2025). In any case, the aim is to make the journalist’s work more efficient (Papadimitriou, 2016;Dodds et al.,2025), which makes its implementation in the media essential (Autor, 2015;Rostamian & Moradi,2024). At the same time, the adoption of AI demands new professional profiles (Sixto García et al.,2021;Møller et al.,2024), which require not only learning to use certain tools (Murcia Verdú & Lara Ramos,2024), but also focusing their training on three central pillars: foundational knowledge, technical skills, and ethical competencies (Lopezosa et al.,2023). In this regard, various authors argue that AI literacy will be a fundamental requirement for future journalists (Graefe,2016;Sonni et al.,2024; Sonni,2025). The integration of AI in journalism is not without ethical challenges. Beyond the potential benefits attributed to it, one of the most debated dilemmas is the inherent bias in AI algorithms (Binns,2018;Deuze & Beckett,2022), compounded by the risk of large-scale dissemination of disinformation (Kertysova,2018). Additionally, there is the threat faced by news consumers in discerning the source and veracity of information, which can lead to trust issues in the media (Montal & Reich,2017;Wu,2024;Morosoli et al.,2025). AI can facilitate certain specific activities within the news production process (Canavilhas,2022), among which the fight against disinformation stands out (Manfredi Sánchez & Ufarte Ruiz,2020;Saeidnia et al.,2025), a matter of particular concern in democratic countries (Rodríguez Martelo et al.,2023). In fact, the advent of AI has increased the possibilities for combating disinformation (Moreno Espinosa et al.,2024) and can help Journal. Media 2025,6, 133 3 of 15 distinguish between truthful information and the distortion of reality (Flores Vivar,2019; Lange & Lechterman,2021;Santos,2023). Ballesteros and del Olmo (2024) state that AI has the capacity to analyze large amounts of information quickly and efficiently, identifying patterns and anomalies that can reveal the veracity of a news item. In this regard, some journalistic organizations have expressed their distrust regarding the evolution of the sector, due to the lack of capacity to combat disinformation and unverified information (Subiela Hernández et al.,2023). To address this concern, factcheckers have emerged as contextual agents to shape a new ecosystem that fights against disinformation (Pilo García et al.,2024). Indeed, some advocate for the incorporation of AI into content-verification processes (Tejedor & Sancho Ligorred,2023). Thus, the potential of AI for information verification—a key process in communication—is recognized (Cuartielles et al.,2023), and it has been demonstrated that the use of AI can be vital for counteracting the disruptive effects of disinformation (Rubin,2022), as it reduces detection time and increases the capacity to respond to it (Cuartielles et al.,2024). Unfortunately, these same AI tools can also prove effective in amplifying the reach and impact of disinformation by enabling the generation of false content. The misuse of AI tools represents a significant challenge in the information age (La Rosa & Luján,2024), as it is possible, at a relatively low cost, to use AI to distribute false information worldwide. This possibility is seen as a threat by the public and by a large part of information professionals, since messages produced using artificial intelligence (AI) techniques can be falsely attributed to non-existent information sources, incorporate false or inaccurate content, and be used for partisan or malicious purposes (Peña-Fernández et al.,2023). The problem of disinformation, understood as that which serves a specific purpose and includes an intentional component, poses a greater risk when situated within the context of a hyperconnected society, where any user of digital environments can simultaneously be both a victim and a conscious or unconscious source of disinformation (Benaissa Pedriza, 2024, pp. 106–107). The malicious use of these models to generate false and credible content automatically, massively, and at no cost constitutes one of their main risks and is the most immediate argument for those calling for urgent regulation (Sanguinetti,2023, p. 13). Moreover, AI not only has the capacity to create convincing and realistic content, but it also raises serious ethical challenges in multiple areas, especially regarding privacy, data protection, and copyright, among others (García,2024). In fact, the emergence of AI in journalism raises questions about the ethics, quality, and transparency of automated information processes. It is crucial that journalism professionals and media organizations understand these challenges and work to address them effectively (Esteban Regules & Calle Mendoza,2024, p. 251). In this context, this article aims to analyze Basque journalists’ perceptions of the impact of artificial intelligence on the phenomenon of disinformation. To address this main objective, this research poses the following questions: RQ1. How is AI going to impact information disorders (disinformation, false content, and deepfakes)? RQ2. Are there significant differences among journalists based on variables such as gender, years of professional experience, the type of media in which they work, and their position within the organization? RQ3. How do Basque journalists perceive risks in their professional use of AI? At which levels—individual and associated—are these risks perceived? Journal. Media 2025,6, 133 4 of 15 2. Materials and Methods This study adopts a quantitative approach, based on a structured survey administered to journalists working in the Basque Country. The sample consists of 504 journalists, identified through the Open Communication Guide of the Basque Government (Basque Government,n.d.), which lists the active media outlets in the region, and with the collaboration of the Basque Association of Journalists (Basque Association of Journalists—Basque College of Journalists,n.d.). The distribution of the survey also benefited from the support of this association. Of the participants, 276 are men (55.1%), 226 are women (44.5%), and 2 individuals identified with other genders (0.4%). According to available data (Basque Government,2022;Pérez et al.,2023), it is estimated that approximately 5000 individuals are employed in the media sector in Euskadi. Therefore, for a 95% confidence level, the survey’s margin of error is ±4.15% (Sarrionandia et al.,2025). Data acquisition was carried out between May and June 2024 through anonymous surveys, without recording any personally identifiable information or sensitive data. At no point were respondents asked to provide their name, address, email, telephone number, or any other information that could allow for their direct or indirect identification. Participation in the survey was entirely voluntary and based on informed consent, which was provided at the beginning of the questionnaire. This introductory section clearly explained the purpose of the research, the confidentiality of responses, and the respondent’s right to decline to complete the form if they wished. As the study involves no physical or psychological risks, does not process specially protected data, and complies with the provisions of the General Data Protection Regulation (GDPR) and current national legislation, it is considered that formal ethics committee approval is not required, in line with standard guidelines for low-risk research in academic settings. The questionnaire was administered primarily through an online platform, which maximized geographic coverage and participant accessibility, and was complemented by telephone support to facilitate participation. The survey was developed ad hoc for the present study. Prior to its administration, the questionnaire underwent a validation process, which included a pilot test with a small group of journalists to ensure the clarity, relevance, and comprehensibility of the items. The questionnaire included closed and multiple-choice questions, organized into two main sections: (a) sociodemographic and professional variables (gender, years of experience, type of media outlet, province of workplace, main activity in the media, and level of responsibility); and (b) perceptions and attitudes regarding the impact of AI on journalism. Among the specific issues analyzed are perceptions of the increase in the risks of information disorders (disinformation, false content, and deepfakes), the rise of biases and digital divides (gender, social class, etc.), and the identification of the main risks associated with the use of AI in the journalistic field (allowing for a maximum of two options to be selected). Only fully completed questionnaires were included in the analysis. Partially completed surveys and those missing informed consent were excluded on that basis. The heatmaps presented in this study were generated using ChatGPT 4.1. All statistical analyses, including correspondence analysis (CA), were performed using R (version 4.4.0) within the RStudio integrated development environment (IDE) (version 1.1.456). 3. Results 3.1. Journalists’ Perceptions of Disinformation Associated with AI The majority of professionals perceive that artificial intelligence will significantly increase the risks of disinformation. Specifically, 89.88% of the responses fall into the Journal. Media 2025,6, 133 5 of 15 categories “considerably” (n = 171) and “significantly” (n = 282), while only 10.12% consider that the impact to fall under the categories of “minimally” (n = 46) or “not at all” (n = 5). These results suggest that journalists view artificial intelligence as a significant challenge to the integrity of the information ecosystem. When analyzing the distribution of responses according to the type of media in which they work (Figure 1), a trend similar to that of the overall sample is observed. However, among journalists working in digital-native media, although the negative perception remains predominant (81.81%, n = 54), there is a higher percentage of responses that minimize the impact (“minimally”: 18.18%) compared to the average for other categories, such as print, radio, or television (91.03%). In contrast, professionals working in communication offices and press agencies are the most likely to consider that AI will have a high negative impact, with 96.95% (n = 84) in the highest-impact categories. Figure 1. Percentage distribution of perceived disinformation risk by media type (X-axis: perception attributes; Y-axis: media type). The perception of the risk of AI-driven disinformation is very similar across genders, with 90.22% (n = 249) for men and 89.82% (n = 196) for women in terms of relative frequency for the “significantly” and “considerably” options (Figure 2). The “other” category includes only two cases, so its representativeness is very limited. The results of Fisher’s exact test, with a p-value of 0.05419, are not conclusive, as they are slightly above the reference value of 0.05. After excluding the “other” category and grouping the “minimally” and “not at all” categories to ensure a sufficiently large sample size per category (typically above 30 cases) for the Chi-square test, the p-value increases to 0.5916. The effect size, measured by Cramér’s V, is 0.046, which is considered very small (Cohen,1988). Adjusting for multiple comparisons using the Holm and Benjamini– Hochberg methods did not alter this conclusion (adjusted p-values: 0.5916). Therefore, we can conclude that there is not sufficient statistical evidence of a relationship between these variables. With regard to the relationship between information disorders and years of experience in journalism (Figure 3), the sum of the percentages for the “significantly” and “considerably” options for each of the groups studied suggests a trend with some nuances: as years of experience increase, so does the perception of AI-related disinformation risk (more than 20 years, 92.76%; 10 to 20 years, 94.44%; 5 to 10 years, 87.32%; and five years or less, Journal. Media 2025,6, 133 6 of 15 80.77%). Journalists with five years or less of professional experience report the lowest percentage in risk perception; however, this group still represents a clear majority overall. Figure 2. Percentage distribution of perceived disinformation risk by gender (X-axis: gender; Y-axis: perception attributes). Figure 3. Percentage distribution of perceived disinformation risk by years of experience (X-axis: years of experience; Y-axis: perception attributes). In this case, the p-value (0.001937) from Fisher’s exact test clearly indicates the existence of statistically significant evidence; these data support the conclusion that there is an association between the variables studied. The effect size (Cramér’s V = 0.144) is small to medium, and the result remains significant after adjusting for multiple comparisons (adjusted p-value: 0.0058, Holm; 0.0058, Benjamini–Hochberg). Journal. Media 2025,6, 133 7 of 15 The correspondence analysis visualization (Figure 4) corroborates the previously noted trend. As professional experience in the field increases, journalists are more likely to perceive a greater influence of AI on disinformation. Nevertheless, all journalists, regardless of years of experience, attribute some impact to AI, as can be inferred from the distance between the “not at all” category (red triangle) and its counterparts representing years of experience (blue dots). Regarding the correspondence analysis, the first dimension explained 85.7% of the total inertia, while the second dimension accounted for 13.3%. Together, the first two dimensions explained 99.0% of the total variance, indicating that the two-dimensional solution provides a good summary of the association structure in the data. Figure 4. Correspondence analysis of the association between journalistic experience and perceived influence of artificial intelligence on disinformation. With regard to how the use of artificial intelligence in the workplace influences the perception of disinformation risks, Fisher’s exact test yielded a p-value of 0.0056, indicating a statistically significant association between the variables. The effect size, measured by Cramér’s V, was 0.120, which is considered small to medium according to Cohen’s (1988) criteria. This result remained significant after adjusting for multiple comparisons using both the Holm (adjusted p= 0.0112) and Benjamini–Hochberg (adjusted p= 0.0084) methods. As shown in the correspondence analysis (Figure 5), there is an inverse relationship between journalists who do not use artificial intelligence at all and their perception of its influence on disinformation, as they tend to choose the “significantly” category. In contrast, information professionals who make intensive use of this technology have a more moderate perception of its impact. The correspondence analysis revealed that Dimension 1 accounted for 70.3% of the total inertia, and Dimension 2 contributed an additional 27.6%. Combined, these two dimensions captured 97.9% of the overall variance, demonstrating that the relationship between the variables is effectively represented within a twodimensional space. Journal. Media 2025,6, 133 8 of 15 Figure 5. Correspondence analysis of artificial-intelligence usage in the workplace and journalists’ perceptions of disinformation risk. 3.2. Perceived Risks of AI Use Among Basque Journalists Regarding the specific risks associated with the direct use of artificial intelligence, Basque journalists mainly highlight the difficulty in identifying false content and deepfakes, as well as the risk of obtaining inaccurate or erroneous data. Difficulty detecting false content and deepfakes is the most frequently cited risk, accounting for 37.9% of responses (n = 382), followed by concerns about content or data inaccuracy, mentioned by 33.33% (n = 336). Other risks include becoming a victim of criminal uses, such as scams or skimming (11.61%, n = 117), and biases stemming from data origin (gender, social class, etc.) (10.81%, n = 109). Only 6.35% (n = 64) (Table 1) report no additional risks. It is important to note that the total number of responses (1008) exceeds the number of respondents (504), as each participant was allowed to select up to two risks. This design permitted respondents to indicate more than one perceived risk, resulting in a higher overall response count. Table 1. Main risks identified by journalists in their use of AI. Risk Category Count Percentage Difficulty identifying false content and deepfakes 382 37.9% Obtaining inaccurate or erroneous content/data 336 33.33% Being a victim of criminal uses (scams and skimming) 117 11.61% Biases due to data origin (gender, social class, etc.) 109 10.81% None other 64 6.35% Note: Respondents could select up to two risks. To examine how the different risks are associated in respondents’ minds, we built a 4 × 4 co-occurrence matrix that combines the two options each journalist could select. Table 2reports the ten distinct risk pairs and their relative frequencies; the three most frequent pairs account for more than three quarters of all co-occurrences. Journal. Media 2025,6, 133 9 of 15 Table 2. Co-occurrence of perceived risk pairs in journalists’ use of AI. Risk Pair Count Percentage Difficulty identifying false content and deepfakes paired with obtaining inaccurate or erroneous content/data 253 50.5% Difficulty identifying false content and deepfakes paired with being a victim of criminal uses (scams, skimming) 69 13.77% Obtaining inaccurate or erroneous content/data paired with biases due to data origin (gender, social class, etc.) 58 11.58% Difficulty identifying false content and deepfakes paired with biases due to data origin (gender, social class, etc.) 40 7.99% Obtaining inaccurate or erroneous content/data paired with being a victim of criminal uses (scams, skimming) 17 3.99% Difficulty identifying false content and deepfakes paired with none other 20 3.99% Being a victim of criminal uses (scams, skimming) paired with none other 25 4.99% Obtaining inaccurate or erroneous content/data paired with none other 8 1.6% Biases due to data origin (gender, social class, etc.) paired with0 none other 5 1% Biases due to data origin (gender, social class, etc.) paired with being a victim of criminal uses (scams, skimming) 6 1.2% Note: Because each respondent could select up to two risks, the co-occurrence counts are based on 504 respondents. Among the journalists who selected two risks, 50.5% (n = 253) combined concerns about difficulty identifying false content and deepfakes with the risk of obtaining inaccurate or erroneous content or data. This finding suggests that, for a significant portion of respondents, the challenges of detecting false content, such as deepfakes, manipulated images, or fabricated news, are closely linked to the risk of receiving inaccurate or erroneous information generated by AI systems. In other words, journalists do not view these issues in isolation; rather, they perceive them as part of a broader, interconnected threat to the reliability of information. The second most common pairing, selected by 13.8% (n = 69) of respondents, links the difficulty of identifying false content and deepfakes with the risk of being a victim of criminal uses, such as scams or skimming. The third most frequent pairing, identified by 11.6% (n = 58) of respondents, combines concerns about obtaining inaccurate or erroneous content or data, with worries about biases due to data origin, such as gender, social class, or other demographic factors. This co-occurrence indicates that, for a segment of journalists, the issue of factual correctness is closely intertwined with the question of fairness in AI-generated information. A related finding is that 7.99% (n = 40) of respondents also paired difficulty identifying false content and deepfakes with concerns about data biases. The remaining pairs have markedly lower frequencies. Two involve “none other”, suggesting a subset of respondents who perceive only a single additional risk beyond their primary choice. Finally, co-occurrences between biases and criminal uses (1.2%) or between inaccuracy and criminal uses (3.99%) appear only sporadically. 4. Discussion In line with the findings of previous studies, the results of this research show a broad consensus among journalists regarding the perceived risk that the implementation of artificial intelligence may pose in amplifying disinformation (RQ1). This view gives rise to predominantly cautious attitudes toward the application of this technology in journalism