The development of generative artificial intelligence (Generative AI), especially large language models (LLMs), has profoundly advanced transformations across various fields. ‘It can creatively generate new content based on input’ (Zhang et al., 2023), significantly changing ‘the production, circulation, and utilization of information and knowledge’ (Zhao & Cao, 2023). However, the powerful generative capabilities of large language models have also made the creation of misinformation more convenient and rapid. ‘AI-generated content (AIGC) possesses characteristics such as complex corpus sources, variable modalities, and a combination of reality and fiction’ (Li et al., 2023), further exacerbating issues like deepfakes and expanding their scope of harm. In the era of AI algorithms, the total amount of misinformation is rapidly increasing and becoming more difficult to distinguish, leading to unprecedented harm to information users.

Among these issues, the harm of misinformation to the elderly population is particularly prominent. Due to various factors such as physical function (Wu, 2022), cognitive ability (Eich et al., 2014), digital literacy (Song & Peng, 2024), and psychological factors (Lichternberg et al., 2016), older adults are more susceptible to adopting misinformation and being misled by it, making them the primary vulnerable group to misinformation. Therefore, this study recruits older adults with experiences in recognizing and adopting misinformation, selecting misinformation recognition events obtained through in-depth interviews as research subjects. We aim to explore the main factors influencing misinformation recognition among older adults to gain a deeper understanding of the internal mechanisms of how they identify misinformation and further promote age-friendly improvements in the information environment of the AI era.

Unlike factual information, misinformation refers to information that is false or inaccurate (Wardle & Derakhshan, 2017), often capable of intentionally or unintentionally misleading the decision-making behavior of information recipients. Against the backdrop of exponential growth in misinformation, issues related to older adults and misinformation have gained attention in academia. Existing research has compared older adults with other groups, exploring the characteristics of the elderly when facing misinformation in terms of focusing on contradictory details (Huff & Umanath, 2018), dissemination and sharing (Zhou et al., 2023), and information management (Zhou et al., 2023). Studies have focused on the relationship between older adults and various categories of misinformation such as rumours (Wu et al., 2019; Song et al., 2020), health misinformation (Cao et al., 2022; Choudhury, 2021; Peng et al., 2024), and false memories (West & Stone, 2014), with particular emphasis on the harm of misinformation to older adults (Chen, 2020; Tessoulin et al., 2020; Zhang & Su, 2019) and corresponding governance measures (Wu, 2022). However, existing research has paid limited attention to the issue of misinformation identification by older adults. Understanding the factors influencing misinformation identification among older adults is an important foundation for comprehending its internal mechanisms and has significant implications for mitigating the harm of misinformation to the elderly population. Therefore, this paper attempts to explore the internal mechanisms of misinformation identification among older adults, specifically addressing the following research questions:

RQ1: What are the main factors older adults rely on to judge the authenticity of information?

As part of a larger research project, this study conducts quantitative analysis based on preliminary grounded theory findings. The data were collected through in-depth interviews with 19 adults aged 60 and above who had experience in identifying and adopting misinformation. The sampling process continued until theoretical saturation was achieved, with no new themes emerging from the final three interviews. Each 30-70 minutes interview was recorded with informed consent and transcribed, yielding over 152,000 words of standardized text materials. Interview duration varied based on participants’ number of relevant experiences and communication styles, while ensuring core questions were thoroughly addressed.

The misinformation recognition events used in this study were collected from the interview texts. Initially, 83 potential cases were identified from 19 interviewees’ narratives. Through fact-checking using authoritative sources, official documents, and expert consultation, we verified the authenticity of each case. After excluding cases where the information’s authenticity could not be conclusively determined, 55 confirmed misinformation recognition cases from 18 elderly individuals (A1-A18) were retained for analysis. These cases covered nine major categories including politics, society, entertainment, life, health, work, finance, and knowledge. The basic characteristics of the interviewees are shown in Table 1.

Based on the preliminary grounded theory research, this study uses quantitative content analysis to quantify the interview data, extracting various factors relied upon by older adults in misinformation identification events. Each event case was scored dichotomously (present = 1, absent = 0) for nine variables. The coding framework and examples are shown in Table 2.

Following this coding framework and standards, each of the 55 events of different categories was scored, and the statistical frequencies of each variable are shown in Table 3. Since the dependent variable (misinformation identification result) in this study is a binary variable (correct recognition = 1, incorrect recognition = 0), this paper uses IBM SPSS Statistics 26.0 to analyse the influence of independent variables through a binary logistic regression model. The data analysis is conducted in two steps: First, analysing the influence of nine original independent variables on the identification result; Second, combining the independent variables by category (see Table 2) to analyse the influence of information sources, information itself, and information users on the dependent variable.

When conducting logistic analysis on the nine variables, the model accuracy was 0.73, indicating relatively high predictive accuracy. The Hosmer-Lemeshow test result (p = 0.481 > 0.05) showed good fit. The regression results are shown in Table 4, where Exp(B) represents the odds ratio for each variable. An odds ratio greater than 1 indicates that the variable increases the likelihood of correctly identifying misinformation, while less than 1 decreases the likelihood. The larger the value, the greater the influence of that variable.

To further explore the relationships between variables, this study combined the 9 original independent variables into 3 dimensions:

x1 (information source) = Public information sources + Interpersonal information sources

x2 (information) = Information content + Information viewpoints + Evidence + Information usefulness + Corroborating information

x3 (information user factors) = Knowledge structure + Practice

After dimension combination, the logistic regression model accuracy was 0.82, and the Hosmer- Lemeshow test result (p = 0.884 > 0.05), showing significantly improved model accuracy and fit. The regression results are shown in Table 5.

The logistic model analysis indicates that older adults’ misinformation identification is influenced by multidimensional factors. Among these, public information sources, information content, and interpersonal information sources are the three most significant factors.

Based on text data obtained from in-depth interviews and building upon preliminary grounded theory research, this paper conducted content analysis and binary logistic regression analysis to reveal key factors influencing misinformation identification among older adults. The results show that public information sources, interpersonal information sources, information content, information viewpoints, and evidence significantly influence misinformation identification. Furthermore, in older adults’ misinformation identification, information sources play a crucial role, the influence of information dimensions has not been fully realized, and the direct influence of the information user factors is limited. These findings are significant for understanding and addressing the challenges older adults face with misinformation and are expected to provide insights for promoting age-friendly improvements in the information society.

Moreover, based on the findings of this study, we are currently examining the factors influencing misinformation identification by LLMs and their differences from older adults. In the next phase of our research, we will discuss how to better leverage LLMs to assist older adults in identifying misinformation.