AI information literacy in healthcare context: profiles and associated variables among healthcare professionals

Abstract

This study examines AI information literacy among healthcare professionals and identifies differences across professional roles. An online survey was administered to 390 healthcare professionals, including 22.3% medical doctors, 42.1% nurses, and 35.6% health professionals. AI information literacy was measured using the AI Information Literacy Scale (AILIS), which captures perceived confidence across four dimensions: process/create, assess, retrieve, and ethics. Findings indicated that across AILIS dimensions, participants reported the greatest confidence in retrieving information with AI and the least confidence in critically assessing AI-generated outputs. Medical doctors scored significantly higher than nurses and health professionals on process/create and retrieve. Across the main regression models, five predictors were included: age, perceived AI knowledge, frequency of AI use, number of AI tools used, and self-perceived understanding of how AI tools work. Self-perceived understanding showed the strongest association with all four AI information literacy dimensions. The models explained between 27.3% and 48.3% of the variance. Age showed small negative associations in most models. Self-reported AI information literacy among healthcare professionals appeared uneven across dimensions, with relatively lower confidence in critical assessment. GenAI integration in healthcare should therefore be approached as a literacy and training challenge, with particular attention to evaluation skills and reflective understanding of AI outputs.

Introduction

Healthcare systems are entering a period in which generative artificial intelligence (GenAI) tools are becoming increasingly visible in professional practice [1]. Large language model (LLM) based systems are now used to summarize information, draft text, explain complex concepts, support documentation, and provide rapid access to synthesized answers [2–4]. In health-related contexts, these tools are being explored for clinical decision support [5], communication, education, administrative work, and information seeking [6]. As a result, healthcare professionals are beginning to work in environments where AI-generated content may shape how information is accessed, interpreted, and used [7, 8].

This shift creates new demands for healthcare professionals. In healthcare, the use of AI-generated information is a high-stakes matter that requires careful judgment, responsible use, and context-sensitive interpretation [9]. Professionals must often make judgments under time pressure [10], communicate clearly with patients and colleagues [11], protect sensitive information, and distinguish between plausible outputs and trustworthy ones [12]. In such contexts, what matters is healthcare professionals’ ability to use AI critically, responsibly, and effectively. This includes retrieving relevant information, evaluating the quality of AI-generated responses, adapting outputs for professional purposes, and recognizing ethical boundaries in the use of AI tools [13].

For example, a healthcare professional may use GenAI to summarize symptoms, prepare a patient-facing explanation, draft educational material, or retrieve information about a treatment option. In such cases, limited confidence in critically assessing the output may increase the risk of relying on incomplete, misleading, or insufficiently contextualized information. Similarly, limited ethical awareness may affect decisions about entering patient information into AI tools, explaining AI-supported information to patients, or maintaining professional responsibility for the final clinical judgment.

These demands can be understood through the concept of AI information literacy. Information literacy has traditionally been defined as a set of abilities related to locating, processing, interpreting, and critically evaluating information [14]. In AI-mediated environments, these abilities remain central, but they are enacted under new conditions: users must formulate prompts, interpret outputs that may be fluent but unverifiable, assess transparency and trustworthiness, and distinguish between human-authored and machine-generated information [15–17]. Accordingly, AI information literacy refers to the ability to locate, create, process, and critically evaluate information with and through AI tools while making informed and ethical decisions about their use [18]. This perspective is particularly relevant to healthcare, where AI-generated information may support professional work but also requires careful judgment, verification, and ethical awareness.

The AI Information Literacy Scale (AILIS) was developed to operationalize AI information literacy as information literacy self-efficacy in GenAI-mediated information behavior [18–20]. Self-efficacy theory suggests that perceived capability can influence how individuals approach complex, uncertain, or unfamiliar tasks [21]. In the context of GenAI-mediated healthcare information practices, perceived capability may shape whether professionals feel able to formulate prompts, interpret responses, question outputs, and regulate their reliance on AI-generated information. At the same time, self-efficacy is not equivalent to demonstrated performance. This distinction is important because self-assessment may diverge from actual performance, especially in complex professional domains. In high-stakes clinical contexts, professionals may feel confident without accurately detecting errors, bias, missing evidence, or inappropriate recommendations in AI-generated outputs [12, 13]. Conversely, lower confidence may sometimes reflect appropriate caution rather than lack of literacy.

Unlike broader AI literacy measures, which often emphasize conceptual understanding, attitudes, or general awareness of AI [22, 23], AILIS focuses on concrete information practices performed with GenAI tools. It captures four dimensions that are relevant to healthcare use: creation and processing, assessment and critique, seeking and retrieval, and ethics. This structure makes it possible to identify uneven literacy profiles, for example when professionals feel more confident retrieving or generating information than critically assessing AI-generated outputs or making ethical decisions about their use.

AILIS is also compatible with existing health informatics and digital health competency frameworks. These frameworks commonly emphasize professionals’ ability to access, manage, evaluate, and use digital information, as well as to act responsibly in technology-mediated care environments. For example, the TIGER framework identifies core health informatics competency areas for nursing and other healthcare roles, including information literacy and information management [24]. More recently, the Digital Health Competencies in Medical Education framework organized digital health competencies into four domains: professionalism in digital health, patient and population digital health, health information systems, and health data science [25]. AILIS extends this competency-oriented perspective to GenAI-mediated information practices by focusing on how professionals retrieve, process, assess, and ethically use AI-generated information. This makes it suitable for examining AI-related learning needs in health professions education.

Although AILIS was validated in a general adult sample, little is known about AI information literacy among healthcare professionals. In particular, it remains unclear how healthcare professionals’ AI-related literacy background is distributed, how they score across the dimensions of AI information literacy, whether these patterns differ across professional groups, and which demographic, professional, and AI-related background variables are associated with AI information literacy in healthcare contexts.

The present study addresses this gap by applying AILIS to healthcare professionals. Given that AI information literacy in healthcare is still an emerging area of research, an initial descriptive profile is needed before more explanatory or intervention-based models can be developed. Healthcare work is also interprofessional: doctors, nurses, and allied health professionals often contribute to shared care processes, while differing in training pathways, scope of practice, information-seeking practices, and clinical responsibilities [26, 27]. These shared and role-specific features may shape how AI-generated information is used, evaluated, and regulated. The study therefore examines participants’ AI-related literacy background, their confidence across AI-related information practices, differences across professional groups, and the demographic, professional, and AI-related variables associated with AI information literacy. We examine the following research questions:

• RQ1. What are healthcare professionals’ self-reported levels of AI-related literacy background, including perceived AI knowledge, frequency of AI use, and number of AI tools used, and do these differ across professional groups?

• RQ2. How do healthcare professionals score on the identified dimensions of AI information literacy, and do these differ across professional groups?

• RQ3. Which demographic, professional, and AI-related background variables are associated with healthcare professionals’ AI information literacy across these dimensions?

Methods

Participants

The sample in this study included 390 participants from a range of healthcare professions and training stages. A sensitivity analysis conducted in G*Power 3.1 indicated that with N = 390, α = 0.05, and power = 0.80, the study was sufficiently powered to detect small-to-medium effects in the one-way ANOVAs (f = 0.16) and small effects in the multiple regression models with five predictors (f² = 0.033).

Participants were recruited from the medical workforce and completed the questionnaire online in Hebrew. The sample included 58 male participants (14.9%) and 332 female participants (85.1%), with ages ranging from 23 to 68 years (M = 40.64, SD = 10.67). Mean seniority in the medical field was 11.13 years (SD = 11.05). In terms of professional role, the sample included 87 participants in medical roles across training and career stages (22.3%), including 28 medical students (7.2%), 13 interns (3.3%), 16 residents (4.1%), and 30 specialists or senior physicians (7.7%). Registered nurses constituted the largest group, with 164 participants (42.1%). The remaining 139 participants (35.6%) represented other health professions, including 84 allied health professionals, such as speech therapists, physiotherapists, and occupational therapists (21.5%), 45 pharmacists or laboratory staff (11.5%), and 10 participants in other roles (2.6%).

Tools

Data were collected using a structured questionnaire consisting of three sections.

AI Information Literacy Scale (AILIS)

AI information literacy was assessed using AILIS [18], a self-report instrument that conceptualizes AI information literacy as information literacy self-efficacy in GenAI-mediated information behavior. The scale assesses perceived ability to work with AI-generated information across four dimensions: process/create, assess, retrieve, and ethics. In its original validation, AILIS was designed to capture concrete information practices performed with GenAI tools across everyday, educational, and professional contexts. The full instrument includes 39 items distributed across four subscales: process/create (16 items, e.g., “Generate titles or summaries using an AI tool”, “Convert a text into a graph or diagram using an AI tool”), assess (13 items, e.g., “Detect hallucinations and ask the tool to correct them”, “Critically edit an output generated by an AI tool”), retrieve (8 items, e.g., “Update or refine prompt after receiving an answer to improve the result”, “Ask the AI tool for examples, sources, or citations”), and ethics (2 items, e.g., “Indicate my use of AI tools in a transparent way”). Items are rated on a 5-point confidence scale ranging from 1 (not at all) to 5 (to a very great extent). In the present sample, internal consistency was high for process/create (α = 0.972), assess (α = 0.974), and retrieve (α = 0.959), and acceptable for ethics (α = 0.794).

GenAI usage patterns

To characterize participants’ AI-related background, we included four self-report indicators: perceived AI knowledge, frequency of AI use, number and type of AI tools used, and self-perceived understanding of how AI tools work. Perceived AI knowledge and frequency of AI use were rated on a 5-point Likert scale ranging from 1 (“not at all”) to 5 (“to a very great extent”). Self-perceived understanding of how AI tools work was rated on a 7-point Likert scale ranging from 1 (“not at all”) to 7 (“to a very great extent”). The number and type of AI tools used were included to reflect the breadth of participants’ GenAI tool repertoire. Perceived AI knowledge reflected participants’ subjective familiarity with AI tools, and frequency of AI use captured how often they used them. These variables were used both to describe the sample and to examine associations with AI information literacy scores.

Demographic and professional background variables

Participants also completed a brief background questionnaire that included year of birth, gender, seniority in the medical field, and professional role within the healthcare workforce.

Procedure and ethics

The survey was administered online via the iPanel internet panel service. Data were collected in December 2025 and January 2026. iPanel is an established Israeli online research platform for administering surveys among consenting respondents and enables the recruitment of geographically dispersed samples through its panel infrastructure. Participants were invited through iPanel’s distribution system, and the questionnaire, which included the AILIS items, GenAI knowledge and use measures, and background questions, required approximately 15 min to complete.

Responses were collected anonymously, and iPanel’s privacy and confidentiality procedures were applied throughout the data collection process. The study was conducted in accordance with institutional and national ethical guidelines for research involving human participants. The research protocol was reviewed and approved by the Tel-Hai University Institutional Review Board (Application No. 01-10-25; approval date: October 22, 2025). Participation was voluntary, and informed consent was obtained electronically, as participants indicated their agreement on the first survey screen before proceeding. Participants were informed that their responses would be used for research purposes only, that they could discontinue participation at any time, and that no identifying information would be stored with their responses. Data were analyzed and reported in aggregate form to protect participants’ privacy.

Analysis

Data were analyzed using IBM SPSS Statistics (Version 31). To address RQ1, descriptive statistics were computed for the AI-related background variables, including perceived AI knowledge, frequency of AI use, number of AI tools used, and self-perceived understanding of how AI tools work. Group differences in these variables were examined using one-way ANOVAs. To address RQ2, descriptive statistics were calculated for the four dimensions of AI information literacy: process/create, assess, retrieve, and ethics, and differences across professional groups were examined using one-way ANOVAs with Bonferroni-adjusted post hoc comparisons.

To address RQ3, a series of multiple linear regression analyses was conducted. Each AI information literacy dimension was entered as a separate dependent variable. The main regression models included age, perceived AI knowledge, frequency of AI use, number of AI tools used, and self-perceived understanding of how AI tools work. These models focused on age and AI-related background variables associated with AILIS scores, while professional group differences were examined directly through ANOVA. Pearson correlation analyses were conducted to examine bivariate associations among the study variables and to provide additional context for interpreting the regression models. Standardized beta coefficients were used to assess the relative magnitude of associations within each regression model.

Results

AI-related literacy background across groups

To characterize participants’ AI-related background, participants reported the number and type of GenAI tools they currently use, the frequency of AI use, their perceived AI knowledge, and their self-perceived understanding of how AI tools work (Fig. 1). The mean number of AI tools used was 1.82 (SD = 1.03). Frequency of AI use was rated on a 1 to 5 scale, with higher scores indicating more frequent use (M = 3.02, SD = 1.17). Perceived AI knowledge was also rated on a 1 to 5 scale, with higher scores indicating greater perceived knowledge (M = 3.35, SD = 1.05). Self-perceived understanding of how AI tools work was rated on a 1 to 7 scale, with higher scores indicating greater perceived understanding (M = 3.66, SD = 1.50).

Fig. 1.

AI-related literacy background among healthcare professionals. Note: Error bars represent standard deviations

To further examine whether AI-related literacy background differed across professional groups, a one-way ANOVA was conducted to compare medical doctors, nurses, and health professionals (Fig. 2). No statistically significant group differences were found for perceived AI knowledge [F(2, 387) = 2.75, p = .065], or for self-perceived understanding of how AI tools work [F(2, 387) = 0.36, p = .700]. However, significant differences emerged for frequency of AI use [F(2, 387) = 3.97, p < .05], and number of AI tools used [F(2, 387) = 6.38, p < .01]. Bonferroni comparisons showed that medical doctors reported more frequent AI use (M = 3.23, SD = 1.15) than health professionals (M = 2.81, SD = 1.13), whereas nurses did not differ significantly from either group. In contrast, health professionals reported using a larger number of AI tools (M = 2.05, SD = 1.07) than nurses (M = 1.63, SD = 1.02), whereas medical doctors did not differ significantly from either group.

Fig. 2.

AI-related literacy background by professional group. Note: Error bars represent standard deviations. Significant Bonferroni-adjusted differences were found for frequency of AI use, with medical doctors reporting more frequent AI use than other health professionals, and for number of AI tools used, with health professionals reporting use of a larger number of AI tools than nurses

Differences across AI information literacy dimensions

Overall, participants reported moderate levels of AI information literacy across the four dimensions. The highest mean score was found for retrieve (M = 3.04, SD = 1.18), followed by process/create (M = 2.80, SD = 1.16), ethics (M = 2.73, SD = 1.28), and assess (M = 2.53, SD = 1.14). This pattern suggests that participants felt relatively more confident in using AI tools to locate and retrieve information than in critically assessing AI-generated information.

To further examine whether AI information literacy differed across professional groups, a one-way ANOVA was conducted to compare medical doctors, nurses, and health professionals (Fig. 3). Significant group differences emerged for process/create [F(2, 387) = 4.71, p < .05, ηp² = 0.024], assess [F(2, 387) = 3.30, p < .05, ηp² = 0.017], and retrieve [F(2, 387) = 4.56, p < .05, ηp² = 0.023]. These effects were small. No significant group differences were found for ethics [F(2, 387) = 1.12, p = .329].

Fig. 3.

AI information literacy dimensions by professional group. Note: Error bars represent standard deviations. Significant Bonferroni-adjusted differences were found for process/create and retrieve, with medical doctors scoring higher than nurses and other health professionals

Bonferroni post hoc comparisons showed that medical doctors scored significantly higher than nurses and health professionals on process/create. Specifically, medical doctors scored higher than nurses (M = 3.13 vs. 2.72, p < .05) and health professionals (M = 3.13 vs. 2.68, p < .05). A similar pattern was found for retrieve, with medical doctors scoring higher than nurses (M = 3.36 vs. 2.98, p < .05) and health professionals (M = 3.36 vs. 2.90, p < .05). For assess, although the omnibus ANOVA reached statistical significance, the effect size was small and Bonferroni-adjusted pairwise comparisons did not reach significance. Therefore, this result should be interpreted cautiously and does not indicate clear pairwise differences between professional groups. No significant pairwise differences were found for ethics.

Variables associated with self-reported AI information literacy across dimensions

Bivariate correlations among the AI-related background variables and AILIS dimensions are presented in Table 1. Perceived AI knowledge, frequency of AI use, number of AI tools used, and self-perceived understanding of how AI tools work were all positively associated with the four AILIS dimensions. The AILIS dimensions were strongly intercorrelated, indicating that participants who reported higher confidence in one area of AI information literacy tended to report higher confidence in the other areas as well.

Table 1.

Pearson correlations among AI-related background variables and AILIS dimensions (n = 390)

Variable	1	2	3	4	5	6	7	8
1. Perceived AI knowledge	—
2. Frequency of AI use	0.706***	—
3. Number of AI tools used	0.312***	0.342***	—
4. Self-perceived understanding	0.593***	0.538***	0.342***	—
5. Process/create	0.577***	0.564***	0.402***	0.589***	—
6. Assess	0.486***	0.454***	0.344***	0.522***	0.860***	—
7. Retrieve	0.504***	0.501***	0.389***	0.542***	0.859***	0.855***	—
8. Ethics	0.410***	0.404***	0.323***	0.453***	0.713***	0.747***	0.763***	—

***p < .001, two-tailed

A series of multiple linear regression analyses was conducted to examine which demographic and AI-related background variables were associated with healthcare professionals’ self-reported AI information literacy across the four dimensions. The main regression models included age, perceived AI knowledge, frequency of AI use, number of AI tools used, and self-perceived understanding of how AI tools work. Professional group differences were examined separately through ANOVA. Sensitivity analyses were conducted with gender and professional group included as additional covariates to examine whether their inclusion changed the pattern of findings.

For process/create, the overall model was significant [F(5, 384) = 71.86, p < .001], and explained 48.3% of the variance. Higher scores were associated with greater perceived AI knowledge (β = 0.182, p < .01), more frequent AI use (β = 0.198, p < .001), use of a larger number of AI tools (β = 0.148, p < .001), and greater self-perceived understanding of how AI tools work (β = 0.312, p < .001). Age was negatively associated with process/create (β = − 0.092, p < .05).

For assess, the model was also significant [F(5, 384) = 42.11, p < .001], explaining 35.4% of the variance. Higher assess scores were associated with greater perceived AI knowledge (β = 0.160, p < .05), use of a larger number of AI tools (β = 0.128, p < .01), and greater self-perceived understanding of how AI tools work (β = 0.312, p < .001). Age was negatively associated with assess (β = − 0.086, p < .05). Frequency of AI use did not reach significance in this model (β = 0.112, p = .061).

For retrieve, the model was significant [F(5, 384) = 52.07, p < .001], and explained 40.4% of the variance. Higher retrieve scores were associated with greater perceived AI knowledge (β = 0.122, p < .05), more frequent AI use (β = 0.173, p < .01), use of a larger number of AI tools (β = 0.160, p < .001), and greater self-perceived understanding of how AI tools work (β = 0.309, p < .001). Age was negatively associated with retrieve (β = − 0.107, p < .05).

For ethics, the model was significant [F(5, 384) = 28.91, p < .001], explaining 27.3% of the variance. Higher ethics scores were associated with more frequent AI use (β = 0.128, p < .05), use of a larger number of AI tools (β = 0.137, p < .01), and greater self-perceived understanding of how AI tools work (β = 0.269, p < .001). In contrast, age (β = − 0.080, p = .083) and perceived AI knowledge (β = 0.097, p = .145) were not significant in this model.

Across all four models, self-perceived understanding of how AI tools work emerged as the strongest and most consistent variable associated with AI information literacy, while age showed small negative associations with three of the four dimensions.

Discussion

This study examined self-reported AI information literacy among healthcare professionals by applying AILIS in a healthcare context. Three main findings emerged. First, participants reported moderate AI-related literacy background overall, including moderate perceived AI knowledge, frequency of AI use, and number of AI tools used. Second, participants’ confidence was uneven across dimensions: retrieval received the highest scores, whereas assessment received the lowest. Third, AI information literacy was more strongly associated with AI-related experience and perceived understanding than with demographic or professional characteristics. Together, these findings suggest that GenAI integration in healthcare should be understood as depending not only on access to technology, but also on multiple dimensions of AI information literacy.

One central contribution of the study is showing that AI information literacy in healthcare is a multidimensional capability [18]. The uneven pattern across dimensions is theoretically and practically meaningful because it suggests that healthcare professionals may feel more confident using AI to retrieve or generate information than in judging the reliability, adequacy, clinical relevance, or ethical implications of what AI produces [28–30]. Importantly, these dimensions may interact in clinically meaningful ways. Confidence in retrieving or generating AI-supported information may become risky if it is not accompanied by sufficient confidence in assessing the accuracy, relevance, and evidentiary basis of the output. The ethical dimension may also function as a regulatory layer by prompting professionals to verify outputs, protect patient privacy, disclose or document AI use when appropriate, and decide when AI-generated information should not be used. For healthcare organizations, this highlights the importance of supporting not only AI use, but also critical, evaluative, and responsible information practices [31–33].

A second contribution of the study is the finding that self-reported AI information literacy in healthcare was consistently associated with AI-related engagement and perceived understanding [34]. The regression analyses showed that self-perceived understanding of how AI tools work was the strongest and most consistent variable associated with AILIS scores, while frequency of AI use and the number of AI tools used were also associated with several dimensions [35]. These associations should not be interpreted causally. It is possible that greater experience with AI tools strengthens professionals’ confidence in AI-mediated information practices, but the reverse direction is also plausible: professionals with higher AI information literacy self-efficacy may be more likely to use AI tools, try a wider range of tools, and report greater understanding of how AI works. By contrast, gender and professional group did not change the substantive pattern in the sensitivity analyses, and age showed only small negative associations with some dimensions.

The small negative associations with age should be interpreted cautiously. They should not be taken as evidence of age-based resistance to technology. They may reflect differences in exposure to GenAI tools, access to recent AI-related training, workplace expectations, or confidence with rapidly changing digital systems. This finding points to the need for professional development that is differentiated by prior exposure and experience rather than by age alone.

These findings have several implications for healthcare contexts. First, workforce preparation for AI should not be framed only around positive attitudes or willingness to adopt innovation. What appears more central is the development of specific competencies related to retrieving, processing, evaluating, and ethically using AI-generated information. Second, the relatively lower scores on assessment point to a particularly important area for professional learning, but they should not be interpreted automatically as a literacy deficit. Because AILIS measures self-efficacy, moderate confidence in assessing AI-generated outputs may also reflect appropriate epistemic caution in a high-stakes clinical context. At the same time, assessment remains central to safe AI use. Educational efforts should therefore support calibrated confidence: the ability to question AI outputs, verify information, recognize uncertainty, and decide when AI-generated information should not be used.

In healthcare, the ethical dimension also requires specific attention because GenAI use may involve sensitive patient information, unclear provenance of AI-generated outputs, and ambiguity regarding responsibility for AI-supported communication or decisions. Ethical AI information literacy therefore includes knowing when patient information should not be entered into AI tools, when AI use should be disclosed, and how professional responsibility is maintained when AI-generated information informs clinical, educational, or patient-facing work.

More concretely, for health professions educators, these findings suggest the need to integrate AI information literacy into professional development and curricula through scenario-based activities that require retrieval, evaluation, verification, and ethical reflection. Such activities could include comparing AI-generated responses with clinical guidelines, identifying unsupported or fabricated claims, practising prompt refinement for information retrieval, and discussing cases in which AI use should be disclosed or avoided. Because GenAI tools, clinical applications, institutional policies, and AI-related risks change rapidly, such training should be treated as an ongoing professional development process rather than a one-time technical orientation. Training materials and organizational guidelines should be revisited periodically as new tools, clinical uses, privacy risks, and forms of AI-generated error emerge. For practicing healthcare professionals, the findings point to the importance of practical routines for checking AI-generated outputs, recognizing uncertainty, and using AI-generated information responsibly in clinical and educational contexts. For students and trainees, early exposure to these practices may help develop calibrated confidence before GenAI tools become embedded in everyday professional work. In this sense, AI information literacy may offer a more actionable framework for healthcare training than broader and more diffuse notions of AI readiness.

The study also contributes conceptually by extending AI information literacy into a healthcare setting. To the best of our knowledge, this is one of the first applications of an explicit AI information literacy framework among healthcare professionals. In a field where discussions of AI often emphasize implementation, clinical performance, or attitudes, the present study highlights the value of focusing on information practices themselves. Understanding how healthcare professionals seek, interpret, evaluate, adapt, and ethically use AI-generated information may therefore be essential for responsible integration.

Several limitations should be noted. First, the study relied on self-report measures, which capture perceived competence rather than observed performance. This is especially important in healthcare, where confidence in using AI-generated information does not necessarily indicate the ability to detect inaccurate, biased, incomplete, or clinically inappropriate outputs. Relatedly, the association between self-perceived understanding of how AI tools work and AILIS scores may partly reflect common method variance, as both were collected through self-report measures in the same questionnaire.

Second, the cross-sectional design does not allow conclusions about directionality. The findings cannot determine whether AI use contributes to greater perceived AI information literacy or whether professionals with higher confidence are more inclined to use AI tools. Future longitudinal or intervention-based studies are needed to examine this issue.

Third, the sample was not designed to be representative of the broader Israeli healthcare workforce. Although it included a range of healthcare professions, the groups were uneven in size, with nurses forming the largest subgroup. Participants were recruited through an online panel and were therefore self-selected, which may be associated with greater willingness to participate in online research, familiarity with digital tools, or interest in AI. The gender composition should also be interpreted cautiously. Although the high proportion of women may partly reflect the large nursing subgroup, the sample was not weighted to match the gender distribution of the national healthcare workforce.

In addition, the study was conducted in one national and linguistic context. The findings should therefore be interpreted in relation to the Israeli healthcare context and may not transfer directly to other health systems. AI information literacy may be shaped by organizational culture, professional hierarchies, institutional policies regarding AI use, privacy and data-protection regulations, and the extent to which GenAI tools are formally integrated into clinical workflows. Future studies should compare AI information literacy across countries and healthcare systems to examine how regulatory and organizational contexts shape professionals’ confidence and practices.

Finally, the ethics subscale includes only two items and therefore does not capture the full range of ethical issues relevant to AI in healthcare, such as consent, algorithmic bias, equity, accountability, and liability. Future studies should examine healthcare-adapted versions of AILIS with a broader ethics dimension and more parsimonious subscales. More broadly, future research should combine self-report instruments with performance-based tasks, behavioral indicators, or observational data to examine whether perceived AI information literacy corresponds to demonstrated AI-related information practices, including verification behavior, error detection, disclosure decisions, and the quality of AI-supported professional work.

Conclusion

This study provides an initial profile of self-reported AI information literacy among healthcare professionals. The findings suggest that confidence in AI-mediated information practices is uneven, with stronger confidence in retrieval than in assessment, and is consistently associated with perceived understanding of how AI tools work. These results position GenAI integration in healthcare as a literacy and training issue, not only a technological one. AILIS may help healthcare systems identify areas in which professionals require targeted support, while future work should examine how perceived AI information literacy relates to observed AI-supported performance, including verification, error detection, and ethical decision-making.

Authors’ contributions

L.A. and I.L. jointly contributed to the conception and design of the study, data collection, data analysis and interpretation, and manuscript writing. L.A. and I.L. both contributed equally to the development of the theoretical framing, revision of the manuscript, and preparation of the final submitted version. All authors reviewed and approved the final manuscript.

Funding

No funding was received for this study.

Data availability

The data supporting the findings of this study are not publicly available due to the conditions of the IRB approval and the need to protect participant confidentiality. The data are therefore not available for public sharing. All information necessary to interpret the findings is provided within the paper.

Declarations

Ethics approval and consent to participate

The study was conducted in accordance with the ethical principles of the World Medical Association Declaration of Helsinki, as amended in 2024. The study protocol was reviewed and approved by the Tel-Hai University Institutional Review Board (Application No. 01-10-25; approval date: October 22, 2025). Participation was voluntary, and electronically documented informed consent was obtained from all participants before they proceeded to the questionnaire.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.