Prior research indicates that individuals with higher information literacy self-efficacy are more confident in their ability to search diverse sources, synthesise information, and apply it to their work (Pinto et al., 2013; Wu, 2018). Individuals with higher information literacy self-efficacy are more likely to effectively engage with AI technologies and utilise advanced intelligent information systems. Although information literacy self-efficacy is an individual-level construct, it has significant implications for knowledge management processes (Al-Emran et al., 2018). Workers with strong information literacy self-efficacy are more likely to actively engage in knowledge acquisition by consulting diverse sources, including scientific data systems, AI-enabled databases and machine learning–driven information retrieval systems, field surveys, and stakeholder feedback (Ali & Warraich, 2020; Alzahrani & Seth, 2021). They are also more inclined to participate in knowledge sharing because confidence in information handling reduces fear of judgment and enhances willingness to exchange insights with colleagues (Budiardjo et al., 2017; Choi & Song, 2020; Dong et al., 2016). Similarly, high information literacy self-efficacy facilitates knowledge utilisation, as individuals who believe in their competence are more likely to apply information to practical problem-solving (Jasin et al., 2024; Karlinsky-Shichor & Zviran, 2016).

Empirical studies provide support for these links. For example, Tsai and Tsai (2003) found that internet self-efficacy positively influenced students’ ability to acquire relevant information online. Information literacy students were more effective in collaborative knowledge sharing (Hwang et al., 2018; Rohma & Khoirunnisa, 2024; Zamir & Kim, 2022). Safdar, Batool, and Mahmood (2020) demonstrated that confidence in information skills predicted knowledge-sharing behaviour. Drawing on these findings, we propose the first hypothesis:

User satisfaction with information systems is widely recognised as a key measure of system success (Bhattacherjee, 2001; DeLone & McLean, 2003). User satisfaction with information systems reflects the extent to which users perceive that an information system meets their needs, supports their tasks, and provides value (Al-Emran et al., 2018). AI-driven information systems enhance user experiences by enabling intelligent data processing and personalised information retrieval. Machine learning algorithms improve the accuracy and relevance of information provided to users. These AI capabilities increase system responsiveness and support more effective decision-making in research environments (Jarrahi, 2018; Wilson & Daugherty, 2018). In agricultural research contexts, user satisfaction with the information systems is crucial because researchers depend on these systems for accessing global agricultural databases, analysing field data, and disseminating research outputs (Lopez et al., 2020; Shahzad et al., 2023).

The effectiveness of knowledge management processes is closely tied to user satisfaction with information systems. When knowledge acquisition mechanisms are AI-augmented and intelligent, workers perceive information systems as providing relevant and timely resources. When sharing practices are encouraged, systems are seen as supportive of collaboration and collective learning (Alhendawi, 2019; Laily et al., 2024). When utilisation is emphasised, workers find that information systems contribute directly to solving problems and enhancing productivity (Shahzad et al., 2023). Empirical studies suggest that well-functioning knowledge management processes enhance trust in information systems, reduce redundancy, and increase perceived system usefulness (Gold et al., 2001; Lee & Choi, 2003). Accordingly, we hypothesise:

Several lines of evidence suggest that information literacy self-efficacy directly shapes user satisfaction with information systems (Ibrahim et al., 2019; Ibrahim et al., 2023; Naveed & Mahmood, 2021; Tsai & Tsai, 2003). Users leverage AI-enabled information systems more effectively when they possess strong information literacy self-efficacy. They critically evaluate machine learning–generated outputs and integrate them into their research tasks with greater confidence (Alon & Malinoff, 2026; Ng, 2012; Siddiq et al., 2016). Confident information users are more capable of managing complex search interfaces, AI-driven databases, interpreting data outputs, and managing bibliographic references (Al-Abyadh & Azeem, 2022). These competencies reduce frustration and enhance the perceived ease of use and usefulness of information systems—two key determinants of user satisfaction with information systems (Zamir & Kim, 2022). Bacarrisas (2023) found that individuals with higher information literacy skills were more satisfied with digital library systems. Similarly, Atikuzzaman and Ahmed (2023) reported that self-assessed competence in information tasks was strongly associated with satisfaction in academic settings.

In the context of agricultural knowledge workers, higher information literacy self-efficacy is likely to enhance satisfaction with systems used for accessing scientific journals, agricultural statistics, and policy documents (Assefa et al., 2011). Workers with high information literacy self-efficacy are better able to independently navigate and utilise AI-enabled information systems, including machine learning–driven functionalities, leading to higher user satisfaction with information systems. Therefore, we hypothesise.

H₃: Information literacy self-efficacy positively impacts user satisfaction with information systems.

Information literacy self-efficacy is consistently linked to better personal knowledge and information management practices, which are foundational to effective knowledge management processes in organisations (Burchard & Myers, 2019; Calaguas & Consunji, 2022; Shahzad et al., 2023). Studies show that knowledge management processes significantly affect user satisfaction with information systems and organisational performance, and that user information literacy self-efficacy is a key factor in successful knowledge management system adoption and satisfaction (Sahibzada et al., 2019). AI-driven tools support these processes by automating data organisation and retrieval. Machine learning algorithms analyse large agricultural datasets to guide knowledge sharing and utilisation. Researchers leverage AI-enabled platforms to integrate and synthesise information across disciplines efficiently (von Krogh, 2018; Wolfert et al., 2017). While direct studies on the mediation model are scarce, research supports that self-efficacy enhances knowledge management practices and user satisfaction (Batool et al., 2023; Sahibzada et al., 2019; Shahzad et al., 2023). In agricultural research organisations, where multidisciplinary collaboration is essential, information literacy self-efficacy is likely to enhance user satisfaction with information systems, primarily when it contributes to effective knowledge management processes that improve decision-making and innovation. Therefore, we hypothesise.

H₄: The impact of information literacy self-efficacy on user satisfaction with information systems is mediated by knowledge management processes.

Figure 1 presents the proposed mediation model of the study

The study employed a stratified random sampling design with proportionate allocation to ensure that all categories of agricultural research institutes under the research wing of the Punjab Agriculture Department were adequately represented. The sampling frame was obtained from the office of the Chief Information Officer, which provided a complete list of knowledge workers employed in these institutes. From this list, 600 respondents were randomly selected proportionate to the size of each stratum. Data were collected from 600 respondents using a mixed approach that combined online and in-person methods to maximise participation. Specifically, a structured Google Form questionnaire was circulated to the study respondents, and follow-up in-person visits to each institute were conducted to improve response rates and address any queries related to the questionnaire. This approach ensured comprehensive coverage of the target population and minimised the risk of non-response bias. Moreover, 415 surveys were returned giving a response rate of 69%. Out of 415 questionnaires, five had many missing values, four were incomplete due to respondent abandonment, and seven showed respondent misconduct. Finally, 399 surveys were usable for data analysis.

The study was conducted in accordance with ethical standards. All participants were informed about the study objectives and procedures. The study participants gave written consent for review and signature before starting the study questionnaire. Confidentiality and anonymity of the study participants’ information were strictly maintained throughout the research.

All constructs in this study were measured using validated instruments adapted from prior literature. The items were contextualised to reflect the work environment of agricultural knowledge workers in Punjab, Pakistan. Responses for all scales were recorded on a 7-point Likert scale (1 = strongly disagree, 7 = strongly agree). Full scales are provided in Appendix A; only illustrative sample items are presented below.

Data analysis was conducted using SmartPLS 4, a variance-based structural equation modelling (SEM) technique suitable for handling complex hierarchical constructs. The evaluation proceeded in multiple stages. At the first stage, the measurement models for the first-order reflective constructs were assessed to establish reliability and validity. The latent variable scores obtained from this step were then saved and incorporated into the dataset. At the second stage, these scores were used to model the second-order formative constructs, following the disjoint two-stage approach, which is considered robust for estimating higher-order constructs. Once both levels of the measurement model were validated, the structural model was tested to examine the hypothesised relationships among information literacy self-efficacy, knowledge management processes, and user satisfaction with information systems. This approach ensured rigorous validation of the hierarchical measurement structure while enabling robust hypothesis testing.

Figure 2 presents the sociodemographic characteristics of the study sample. In terms of gender, 290 were male (72.5%), while 110 were female (27.5%). Regarding qualifications, 281 held an M.Phil./M.Sc. (Hons.) degree (70.3%), 112 held a PhD (28%), 6 held a Master’s degree (1.4%), and 1 held a B.Sc. (Hons.) degree (0.3%). In terms of designation, 210 were Scientific Officers (52.4%), 158 were Senior Scientists (39.4%), 19 were Principal Scientists (4.8%), 10 were Chief Scientists (2.5%), and 3 held other positions (0.8%).

With respect to professional experience, 196 were mid-career professionals with 11–20 years of experience (49%), 130 were early-career professionals with 1–10 years of experience (32.6%), and 74 were late-career professionals with 21 years or more (18.4%). Participants represented a wide range of agricultural disciplines, including Plant Breeding and Genetics with 136 respondents (34%), Soil Science with 100 respondents (24.9%), Agronomy with 45 respondents (11.3%), Entomology with 38 respondents (9.6%), Horticultural Sciences with 34 respondents (8.5%), Plant Pathology with 25 respondents (6.2%), Food Technology with 11 respondents (2.8%), Agricultural Engineering with 4 respondents (1.1%), Biotechnology with 1 respondent (0.3%), and 4 respondents from other fields (1.1%). Overall, the sample was largely composed of mid-career researchers, with Plant Breeding and Soil Science emerging as the most represented areas of specialisation.

Data analysis began with an assessment of the measurement model for the lower-order constructs, followed by validation of the higher-order constructs, and finally testing of the structural model to evaluate the proposed hypotheses. Reliability and validity indicators were computed for each construct to ensure the adequacy of the measurement model.

Internal consistency reliability was assessed using Cronbach’s alpha (α) and composite reliability (CR). Following Hair and Alamer (2022), α and CR values above 0.70 are considered acceptable for social science research. All constructs in the study met or exceeded this threshold, indicating strong internal consistency (Table 1). Convergent validity was examined using average variance extracted (AVE) and item loadings. An AVE of 0.50 or higher is regarded as evidence of convergent validity (Fornell & Larcker, 1981), while standardised loadings above 0.70 indicate adequate indicator reliability (Hair & Alamer, 2022). All constructs in the model exhibited AVE values well above the recommended 0.50 threshold, and most items loaded above 0.70, confirming convergent validity. Multicollinearity among items was assessed using the variance inflation factor values. Variance inflation factor values below 5.0 are generally accepted as indicating no problematic collinearity (Hair & Alamer, 2022). All items in this study met this criterion, showing that the indicators did not exhibit collinearity issues.

Additionally, discriminant validity was assessed using three complementary approaches: the Heterotrait–Monotrait (HTMT) ratios of correlations (Table 2), Fornell–Larcker criterion (Table 3), and cross-loadings (Appendix B). According to the Fornell and Larcker (1981) criterion, the square root of the AVE for each construct should exceed its correlations with other constructs. The results indicated that all constructs met this criterion, as the diagonal values (square roots of AVE) were greater than the off-diagonal correlations. Next, HTMT ratios were examined to ensure that constructs were distinct from one another. Consistent with the guidelines of Henseler et al. (2015) and Hair and Alamer (2022), HTMT values below 0.85 (or more conservatively 0.90) indicate adequate discriminant validity. All HTMT ratios in this study fell below the 0.85 threshold, supporting discriminant validity. Finally, cross-loadings were examined to ensure that each indicator loaded highest on its intended construct compared to all other constructs. These cross-loadings are presented in the Appendix B.

Together, these findings demonstrate that the lower-order constructs achieved satisfactory reliability and construct and discriminant validity according to standard guidelines. Using the disjoint two-stage approach, latent variable scores from the first-stage model were used to estimate the higher-order constructs of Information Literacy Self-Efficacy and Knowledge Management Processes. This two-stage procedure allowed for robust testing of the higher-order model and the subsequent structural model (Sarstedt et al., 2019).

Higher-order constructs offer several benefits to researchers. They streamline complex models by reducing the number of direct paths between variables, thus improving parsimony (Edwards, 2001; Johnson et al., 2011; Polites et al., 2012). By combining multiple related dimensions into a single overarching construct, researchers can simplify relationships in their models and avoid redundancy between lower-order components and outcomes. Higher-order constructs also address the bandwidth–fidelity dilemma (Cronbach & Gleser, 1957), balancing the breadth of information captured with the depth of measurement accuracy. In addition, they help mitigate collinearity among formative indicators by redistributing items or dimensions across more clearly defined subdimensions of the overall construct (Hair et al., 2023).

Information literacy self-efficacy and knowledge management processes were higher-order constructs in the study. Information literacy self-efficacy was comprised of basic, intermediate, and advanced information literacy skills as reflective lower-order constructs. Knowledge management processes had knowledge acquisition, knowledge sharing, and knowledge utilisation as reflective lower-order constructs.

Several steps were taken to validate both higher-order constructs. First, multicollinearity was assessed to check the collinearity issues with formative models. Variance inflation factor values were checked for both information literacy self-efficacy and knowledge management processes higher-order constructs. When the Variance inflation factor is higher, multicollinearity is higher. Variance inflation factor values of 5 or higher indicate multicollinearity issues. In the present study, multicollinearity did not pose any serious threat because Variance inflation factor values were less than the recommended cut-off value of 5. Next, the statistical significance and relevance of outer weights were assessed. Outer weights were found to be significant. This further validates the higher-order constructs of information literacy self-efficacy and knowledge management processes. Finally, outer loadings were found to be greater than 5 and significant for each indicator of the formative higher-order constructs, information literacy self-efficacy and knowledge management processes. Since all criteria are met (Table 4), the validity of the higher-order constructs is established.

Following the assessment of measurement models, the next step is the evaluation of structural paths for the evaluation of path coefficients (relationships among the study constructs) and their significance.

H₁ evaluate whether information literacy self-efficacy positively and significantly affects knowledge management processes. The results reveal that information literacy self-efficacy had a significant and positive impact on knowledge management processes (β =.464, t = 11.305, p < 0.001). Hence, H₁ is supported. H₂ assess whether knowledge management processes positively and significantly affect user satisfaction with information systems. The findings demonstrate that the knowledge management processes have a positive and significant impact on user satisfaction with information systems (β = 0.457, t = 11.332, p < 0.001). Consequently, H₂ is supported. H₃ ascertains whether information literacy self-efficacy positively and significantly impacts user satisfaction with information systems. The results show that information literacy self-efficacy has a positive and significant impact on user satisfaction with information systems (β = 0.253, t = 5.903, p < 0.001). Thus, H₃ is supported. The results are presented in Table 5.

H₄ evaluates whether knowledge management processes mediate the relationship between information literacy self-efficacy and user satisfaction with information systems. The results demonstrate that knowledge management processes mediate the relationship between information literacy self-efficacy and user satisfaction with information systems (b = 0.212, t = 7.834, p < 0.001). Therefore, H₄ is supported. Notably, the direct impact of information literacy self-efficacy on user satisfaction with information systems in the presence of the mediator is significant and positive (β = 0.253, t = 5.903, p < 0.001). Indirect impact was also found to be significant and positive. Hence, there is partial and complementary mediation in the structural model. The summary of mediation results is presented in Table 6.

These findings contribute to the literature in several ways. First, they extend the study of information literacy self-efficacy beyond academic libraries and student populations to public-sector research organisations in a developing country. The findings show that information literacy self-efficacy remains critical for knowledge management processes and user satisfaction with information systems. Second, they validate the disjoint two-stage approach for higher-order constructs in the agricultural research context. The findings also demonstrate a robust methodological path for future studies. Third, they highlight the importance of knowledge management processes as a mediating mechanism between information literacy self-efficacy and user satisfaction with information systems—an area often overlooked in information systems and knowledge management research. Finally, they add geographic and sectoral diversity to the literature, which has traditionally focused on business or educational settings in developed economies.

The results underscore the importance of investing in both individual and organisational capacities. Training programmes to improve information literacy—particularly in digital searching, bibliographic management, and data analysis—may directly enhance user satisfaction and indirectly improve it by strengthening knowledge management processes. At the organisational level, policies encouraging knowledge sharing and utilisation can amplify the benefits of individual skills, leading to more efficient use of information systems. By creating an integrated environment where skilled individuals and supportive systems reinforce each other, research institutes can improve not only satisfaction but also innovation and evidence-based decision-making.