This observational, cross-sectional survey was conducted among pharmacy students and pharmacy graduates in Saudi Arabia using an online, self-administered questionnaire. The instrument was structured according to a Knowledge-Attitudes-Practice (KAP) framework and informed by constructs from technology-acceptance models (perceived usefulness, perceived ease of use, and behavioral intention) ensuring comprehensive assessment of both cognitive and behavioral dimensions of AI adoption in pharmacy education and practice. Item generation was based on a review of published surveys on AI and digital health in pharmacy and other health-professional education, as well as broader reports on digital transformation in Saudi healthcare. Existing KAP instruments on AI and digital health in pharmacy and other health professions were reviewed and adapted to guide the domain structure (KAP and practice) and the choice of response scales. In this study, items referring to AI “accuracy,” “reliability,” or “safety” were intentionally classified under the perception domain, as they reflect respondents' subjective beliefs and trust judgments rather than objectively verifiable factual knowledge.

The final questionnaire comprised four sections: (i) sociodemographic and educational characteristics (e.g., age, gender, region, current role, and highest pharmacy qualification); (ii) knowledge of AI concepts and applications in healthcare and pharmacy; (iii) attitudes and perceptions regarding the benefits, risks, and future role of AI in pharmacy education and practice; and (iv) self-reported practices and intended use of AI tools in educational and professional contexts. Knowledge items were single-best-answer questions focusing on fundamental definitions, examples of AI applications across the medication-use process, and awareness of specific AI platforms, including both widely discussed general-purpose tools and specialized healthcare AI systems (e.g., IBM Watson, PathAI), with response options such as “Yes/Correct,” “No/Incorrect,” or “Not sure.” These items were designed to capture recognition and conceptual understanding rather than hands-on experience with these tools. Many of the specialized healthcare AI systems included in the knowledge items require institutional licensing or integration into clinical environments and are therefore not routinely accessible to students or early-career practitioners, which informed the distinction between awareness and actual use. For analysis, correct answers were coded as 1 and incorrect/“Not sure” as 0, and individual items were summarized as proportions of correct responses; in addition, total knowledge scores were derived by summing item scores, with higher scores indicating better knowledge.

Attitudes and perceptions were assessed using five-point Likert-type items (strongly disagree to strongly agree) that captured agreement with statements about AI impact on efficiency, safety, professional roles, and ethical/data-governance issues; higher scores reflected more positive attitudes and greater perceived usefulness of AI. Practice items asked respondents to indicate how often they used AI tools (e.g., LLM-based chatbots, AI-enabled drug-information systems, or electronic prescribing decision support) for learning, research, or clinical tasks, using ordinal frequency options (never, rarely, sometimes, often, very often), with higher values indicating more frequent use. These items did not specify a fixed recall window (e.g., past month or past year) and were intended to capture respondents' typical or usual frequency of AI use, rather than usage within a predefined time period. The choice of Likert-type and ordinal frequency scales was based on their established use in KAP and technology-acceptance research and their alignment with previous AI-in-education surveys.

Attitudes and perceptions were assessed using five-point Likert-type items (strongly disagree to strongly agree) that captured agreement with statements regarding the impact of artificial intelligence on efficiency, safety, professional roles, and ethical or data-governance issues; higher scores reflected more positive attitudes and greater perceived usefulness of AI. Practice was assessed using categorical items that asked respondents to identify the artificial intelligence tools or software they had used for academic, research, or professional purposes (e.g., large language model-based chatbots, AI-enabled drug-information resources, or electronic prescribing and clinical decision-support systems). These items were designed to capture self-reported exposure and hands-on use of specific AI applications, rather than the frequency or intensity of use. As no fixed recall window (e.g., past month or past year) was specified, responses reflect participants' general experience with AI tools rather than usage within a predefined time frame. The choice of item formats was guided by their established use in KAP-based surveys and previous studies on AI adoption in health-professional education.

To further clarify construct boundaries, the perception domain was conceptualized to capture respondents' beliefs, trust, and subjective appraisals of AI tools, distinguishing it from both the knowledge and attitude domains. Knowledge items assessed factual understanding of AI concepts and applications, whereas attitude items reflected evaluative or affective judgments about AI integration in pharmacy education and practice. In contrast, perception items addressed respondents' interpretations of reliability, accuracy, safety, and potential risks, such as concerns about false information, data accuracy, or possible misuse. These items reflected belief-driven judgments rather than factual correctness, thereby justifying retention of perceptions as a separate analytical domain despite anticipated heterogeneity. These “accuracy”-related items were framed as belief-based evaluations of AI performance rather than assessments of respondents' factual correctness, thereby distinguishing perception from knowledge. Accordingly, accuracy-related items were retained within the perception domain rather than reclassified as knowledge items.

Data collection was conducted between October 2024 and November 2024 using an online survey platform. The survey link was disseminated via institutional email lists, student and alumni groups, and professional social media channels, using an open invitation approach. Participants were also encouraged to share the survey link within their academic and professional networks to enhance participation from regions with initially low response rates. Because the survey was distributed through open electronic channels rather than a closed sampling frame, it was not possible to determine how many individuals actually received, viewed, or were eligible to participate in the survey. Consequently, a conventional response rate (i.e., the proportion of invited individuals who responded) could not be calculated. This approach may introduce selection bias, as participation was voluntary and may have preferentially attracted respondents with greater interest or familiarity with artificial intelligence. As a result, the findings should be interpreted with caution, particularly with respect to representativeness and generalizability beyond the surveyed population.

Participants were eligible for inclusion if they were (i) currently enrolled in a pharmacy programme (Diploma, B Pharm, M pharm or PharmD) or (ii) pharmacy graduates holding a Diploma, Bachelor's, Master's, or PhD degree, and (iii) residing in Saudi Arabia at the time of the survey. Participants were excluded if they did not provide informed consent or submitted incomplete questionnaires.

Because these channels did not allow tracking of how many individuals actually viewed or received the link, a conventional response rate could not be calculated. Further, to encourage multi-region participation, we collaborated with faculty contacts in multiple regions. Despite this, participation was disproportionately higher from the Asir region, and regional comparisons are therefore interpreted as exploratory. Accordingly, sampling was purely convenience-based rather than stratified by region.

The target sample size for the primary objective (estimation of overall KAP toward AI among pharmacy learners and graduates) was calculated using a single-proportion formula, assuming a conservative expected proportion, 95% confidence level, and 5% margin of error, yielding a minimum required sample of approximately 333 participants. No formal power calculation was performed for regional subgroup comparisons because the primary aim was overall KAP estimation in a multi-region convenience sample. Accordingly, any analyses by region are presented as exploratory and interpreted with caution, recognizing that some regions contributed relatively small numbers of respondents. As a result, the study is not powered to detect modest regional differences, and we have avoided making strong generalizations about regional effects. Participants were eligible if they were currently enrolled in a pharmacy programme or held a pharmacy degree (Diploma, BPharm, PharmD, Master's, or PhD) and were residing in Saudi Arabia at the time of the survey. Individuals who did not provide informed consent or submitted incomplete questionnaires were excluded.

Statistical analyses were performed using SPSS (version 24.0; IBM Corp., Armonk, NY, USA). Descriptive statistics (frequencies and percentages) were used to summarize sociodemographic characteristics and item-level responses for the KAP domains. For bivariate analyses, associations between KAP variables and key sociodemographic/educational factors were examined using chi-square tests or logistic regression, as appropriate. A p-value < 0.05 was considered statistically significant. Cramer's V values are reported alongside corresponding p-values in relevant tables.

Bivariate analyses were performed using logistic regression to examine associations between demographic variables and knowledge, attitude, and perception domains related to artificial intelligence. Results are presented as odds ratios (ORs) with 95% confidence intervals (CIs). For categorical variables with small cell counts, results were interpreted with caution due to the potential instability of estimates. Regional subgroup analyses were conducted as exploratory analyses because of the uneven geographic distribution of participants and the relatively small sample sizes within certain regions. No post-hoc category merging was performed unless conceptually justified, in order to preserve interpretability. Accordingly, findings from subgroup and regional analyses are intended to be hypothesis-generating rather than definitive.

The internal consistency of the questionnaire was assessed using Cronbach's alpha (α). Each domain of the KAP, Perception, and Need for AI was evaluated separately to determine the reliability of its items. Cronbach's alpha values were computed after data collection to examine how well the items within each scale measured their respective constructs. Alpha values ≥0.70 were considered acceptable, while values ≥0.80 indicated good reliability. This analysis ensured that the questionnaire items were coherent, consistent, and suitable for measuring the targeted domains within the study population.

This study involved human participants who voluntarily completed an online survey. Ethical approval was obtained from the King Khalid University Research Ethics Committee (HAPO-06-B-001; Approval No. ECM#2024-2805). The research adhered to the principles of the World Medical Association Declaration of Helsinki. All participants were informed about the study objectives, assured of confidentiality, and notified that no personally identifiable information would be collected. Participation was entirely voluntary, and individuals could withdraw from the survey at any time.

A total of 333 participants responded to the distributed questionnaire over a 2-month period. Table 1 shows the sociodemographic characteristics of participants. Of these, more than half participants were female 194 (58.3%). Among the study population, the majority of respondents were from the Asir region (21.9%, n = 73), followed by Najran (12.3%, n = 41), Khamis Mushayt (16.8%, n = 56), and Al Madinah Al Munawwarah (3.6%, n = 12). Al Madinah Al Munawwarah (2.7%, n = 9). This correction was made to ensure the description accurately reflects the data presented in Table 1. Most participants (98.2%) were classified as youth (18–39 years). Almost half of the respondents were students pursuing a pharmacy degree (48.0%, n = 160), followed by B. Pharmacy graduates (46.3%, n = 157). A smaller number were enrolled in M. Pharmacy (2.7%, n = 9) and Ph.D. programs (1.5%, n = 5). Regarding occupational distribution, the majority were students (55.9%, n = 186), followed by participants employed in the private sector (21.0%, n = 70).

A notable regional skew was observed in the sample composition. Approximately 60% of the respondents were from the Asir region when all cities within Asir including Abha, Khamis Mushayt, Barq, Tharban, Bisha, Mahayil, Sarat Ubaida and The Majardah. The remaining 40% of participants represented other regions across Saudi Arabia (non-Asir region). This distribution reflects the higher accessibility and engagement of participants from the Asir region, which may influence the representativeness of regional perceptions toward AI. However, clarifying this distribution provides transparency and ensures accurate interpretation of the findings.

The results showed that a strong majority of participants demonstrated awareness of pharmacy-related AI tools, including clinical decision-support systems, AI-based medication management platforms, and commonly used large language models.

Table 2 illustrates participants' knowledge of AI across different pharmacy education levels. A majority of pharmacy students (47.4%) and bachelor's pharmacists (44.4%) reported having heard about AI programs, indicating a strong general awareness (p = 0.6, Cramer's V = 0.091). Most respondents acknowledged AI as a useful tool for medication management (33 and 35.4%; p = 0.1, Cramer's V = 0.133), patient safety (34.2 and 36.0%; p = 0.2, Cramer's V = 0.128), and cancer research (28.5 and 33.9%; p = 0.2, Cramer's V = 0.120). Participants also recognized AI's utility in clinical trials (p = 0.8, Cramer's V = 0.080), drug discovery (p = 0.2, Cramer's V = 0.121), personalized treatment (p = 0.4, Cramer's V = 0.107), and pharmacy supply chain management including clinical decision-support systems (χ² = 0.6, p = 0.097, Cramer's V = 0.097). Only a few respondents expressed uncertainty or lack of knowledge. Overall, the results demonstrate that a strong majority of participants were aware of pharmacy-related AI tools, including clinical decision-support systems, AI-based medication management platforms, and commonly used large language models.

Table 3 presents the attitudes of pharmacy students and professionals toward the application of AI in pharmacy practice. A majority of respondents, particularly pharmacy students (36.0%) and bachelor's degree pharmacists (37.5%), showed a positive attitude toward the usefulness of AI in identifying medication errors, high-risk drug dosing, and drug–drug interactions (p = 0.08, Cramer's V = 0.145). Around 27.3% of students and 23.4% of bachelor pharmacists agreed that AI- and ML-based systems can support personalized dosage recommendations (p = 0.2, Cramer's V = 0.124). Regarding emerging technologies, attitudes toward the usefulness of NFC tags in pharmacy practice were mixed, with 29.7% of students and 18.0% of bachelor pharmacists expressing agreement, while a larger proportion remained skeptical or disagreed (p = 0.9, Cramer's V = 0.034).

A higher proportion of participants (34.5% students and 36.0% bachelor pharmacists) strongly agreed that AI supports public health monitoring and detection of disease trends (p = 0.6, Cramer's V = 0.089). Additionally, 33.3% of students and 38.4% of bachelor pharmacists recognized AI's role in optimizing supply chain management, and this association was statistically significant (p = 0.02, Cramer's V = 0.187). Furthermore, 28.2% of students and 29.4% of bachelor pharmacists acknowledged the usefulness of interactive chatbots like Buoy Health for promoting treatment adherence (p = 0.7, Cramer's V = 0.075). Overall, the findings indicate a generally positive attitude toward AI adoption in pharmacy operations, with bachelor's degree pharmacists and students showing the most supportive responses across all parameters.

It should be noted that Table 3 assesses respondents' conceptual awareness and perceived usefulness of various AI platforms used in healthcare and pharmacy practice, many of which (e.g., IBM Watson for Oncology, PathAI, and DeepMind applications) are institutionally deployed tools that typically require organizational licensing and are not intended for routine personal use. In contrast, Table 5 captures self-reported personal use of readily accessible AI tools, such as ChatGPT or Siri. This distinction explains the apparent discrepancy between high awareness or perceived usefulness of specialized AI systems and their limited reported personal use.

Collectively, these findings indicate broad acceptance of AI applications in pharmacy practice, with limited variation across educational levels and strongest support observed for supply chain optimization.

The study evaluated self-reported experience with AI programs among pharmacy students and pharmacists across different qualifications (Table 4). Practice items captured whether respondents had ever used specific AI applications for academic, research, or professional purposes, rather than use within a predefined time frame (e.g., past month or year). The study evaluated the use of AI programs among pharmacy students and pharmacists across different qualifications (Table 4). Overall, a majority of participants reported using AI programs, with pharmacy students leading at 44.7%, followed by pharmacists with bachelor's degrees (36.3%), while usage was lower among diploma, master's, and Ph.D. holders (p < 0.001, Cramer's V = 0.238). When specifically asked about using AI for academic purposes, 43.5% of students and 35.4% of bachelor's pharmacists reported engagement, showing a statistically significant difference (p = 0.002, Cramer's V = 0.229). Diploma, master's, and Ph.D. pharmacists consistently reported lower usage across all AI applications, including retrieving drug information, exam preparation, performing medication-related duties, and writing research articles (p-values ranged from 0.01–0.1, Cramer's V = 0.142–0.193). These findings highlight that younger or academically enrolled participants are more likely to integrate AI into their studies and professional tasks. Overall, the trend suggests increasing AI adoption among pharmacy professionals, particularly for educational and research purposes.

Among the participants, ChatGPT was the most frequently used AI program, with 42.3% of pharmacy students and 36.0% of pharmacists with a bachelor's degree reporting usage (p < 0.001, Cramer's V = 0.244). Siri was used by 9.0% of students and 7.2% of bachelor's pharmacists (p = 0.01, Cramer's V = 0.195). Other AI tools, including Clalit AI, Drawn AI, Google Assistant, Pharmee, Robotic App, and miscellaneous programs, were used by a small fraction of participants (0.3%−4.8%) with no statistically significant differences between groups (p > 0.05, Cramer's V = 0.067–0.142). A large proportion of respondents reported not using AI programs, highlighting limited adoption across pharmacy professionals. Overall, ChatGPT appears to be the most preferred AI tool among participants, particularly among students and bachelor's degree holders (Table 5).

Table 6 summarizes participants' perceptions and concerns regarding the use of AI programs in pharmacy education and practice. In this context, “perception” reflects respondents' subjective judgments, confidence, and concerns regarding AI-generated outputs rather than objective verification of AI performance. Items addressing data accuracy were interpreted as perceived reliability rather than factual knowledge. Approximately 12.3% of students and 12.9% of bachelor pharmacists perceived AI-generated drug data to be accurate, whereas 19.5% of students and 14.7% of bachelor pharmacists expressed skepticism regarding data accuracy (p = 0.07, Cramer's V = 0.147). A substantial proportion of respondents (36.0% of students and 33.6% of bachelor pharmacists) perceived that AI-based data collection may contain false information (p = 0.9, Cramer's V = 0.068).

Despite these concerns, attitudes toward the educational value of AI were predominantly positive, with 42.3% of students and 38.7% of bachelor pharmacists agreeing that AI programs are helpful for learning (p = 0.3, Cramer's V = 0.118). However, approximately one-quarter of respondents perceived that AI tools could facilitate academic dishonesty (p = 0.3, Cramer's V = 0.117). Overall, participants perceived AI programs as beneficial for enhancing understanding and motivation (p = 0.8, Cramer's V = 0.083), while simultaneously expressing concerns related to data reliability and potential misuse.

Table 7 presents respondents' views on the perceived need for integrating AI into pharmacy education, healthcare practice, and research. This variable assesses participants' interest in incorporating AI-related topics into pharmacy curricula, adopting AI-enabled tools in clinical settings, and expanding their own knowledge of AI applications in healthcare and research. A substantial proportion of pharmacy students (37.2%) and bachelor pharmacists (39.0%) expressed interest in including AI-related content in the pharmacy curriculum (p = 0.4, Cramer's V = 0.104). Similarly, 38.4% of students and 39.6% of bachelor pharmacists supported implementing AI-enabled tools in community pharmacies and hospitals to improve patient care (p = 0.6, Cramer's V = 0.090). Interest in learning more about AI applications in both hospital and community settings was also high, with 41.1% of students and 41.7% of bachelor pharmacists responding positively (p = 0.7, Cramer's V = 0.071). Overall, these findings reflect a strong perceived need for integrating AI into pharmacy education, healthcare practice, and research domains, underscoring growing recognition of its potential to enhance efficiency, accuracy, and patient outcomes.

The internal consistency of the scales used in this study was assessed using Cronbach's alpha (α). The Knowledge scale, comprising eight items, demonstrated good reliability with an α value of 0.798. The Attitude scale, which included 14 items, showed excellent internal consistency (α = 0.884). The Practice scale (six items) also exhibited good reliability (α = 0.804). The Perception scale, consisting of five items, yielded a moderate reliability coefficient (α = 0.592), indicating that some items may require refinement to enhance internal consistency. Lastly, the three-item Need for AI scale demonstrated acceptable reliability with an α value of 0.783. Overall, the majority of the scales showed good to excellent reliability, supporting their suitability for use in the study (Table 8).

Before interpreting the regional comparisons presented in Table 9, it is important to note the uneven geographic distribution of the study sample. Of the 333 respondents, 200 participants (60.1%) were from the Asir region, while 133 participants (39.9%) represented all other regions combined. Consequently, all regional comparisons were conducted as exploratory analyses. These findings are not nationally representative and should be interpreted with caution, as they are intended to be hypothesis-generating rather than definitive assessments of regional differences in knowledge, attitudes, and perceptions toward artificial intelligence.

The bivariate analysis assessed the association between demographic characteristics and participants' KAP toward the use of AI in pharmacy practice (Table 9). Gender did not show any significant association with knowledge (OR = 0.812, p = 0.3), attitude (OR = 1.192, p = 0.5), or perception (OR = 0.871, p = 0.5). Similarly, age and occupational categories were not significantly related to these domains (p > 0.05). Although higher educational qualifications, particularly among pharmacists with master's degrees, exhibited relatively higher odds of good knowledge (OR = 12.00, p = 0.07), these differences did not reach statistical significance.

Exploratory regional analyses indicated statistically significant associations in selected locations. Within the Asir region, participants from Abha (p = 0.03), Barq (p = 0.04), and Khamis Mushayt (p = 0.03) showed higher odds of positive perception toward AI. In addition, respondents from Tharban demonstrated higher odds of good knowledge (OR = 7.80, p = 0.03), while those from The Majardah showed the strongest association with positive perception (OR = 18.00, p = 0.009).

Among non-Asir regions, exploratory findings showed that participants from Riyadh exhibited notable associations with attitude (OR = 16.250, p = 0.003) and perception (OR = 7.031, p = 0.02), while associations with knowledge did not reach statistical significance (OR = 3.273, p = 0.09). Jazan and Mecca demonstrated significant associations with perception (p < 0.05), and participants from Tabuk showed a significant association with attitude (OR = 10.00, p = 0.05).

Overall, these exploratory regional findings suggest localized variation in AI-related knowledge, attitudes, and perceptions; however, due to the predominance of respondents from the Asir region (60.1% Asir, 39.9% other regions), these results should be interpreted as hypothesis-generating and not as evidence of national regional differences.

Several limitations should be considered when interpreting the findings of this study. First, the sample exhibited a notable regional imbalance due to non-probability, convenience-based online recruitment. Because invitations were distributed through open electronic channels without the ability to track how many individuals viewed the survey link, a conventional response rate could not be calculated. Approximately 60% of respondents were from the Asir region and 40% from other regions of Saudi Arabia. This uneven representation limits the generalisability of the findings to the national level and underscores that the regional comparisons reported in this study should be interpreted as exploratory rather than definitive.

Second, the reliance on self-reported questionnaire data introduces the possibility of response bias and social desirability bias. The study population was also limited to pharmacy students and graduates, excluding other healthcare professionals whose perspectives may differ, thereby constraining external validity. In addition, the cross-sectional design precludes causal inference and does not permit assessment of changes in knowledge, attitudes, or practice over time.

Third, the temporal gap between data collection (November 2024) and manuscript revision represents an important limitation. Given the rapid evolution of AI technologies and their expanding integration into pharmacy education and practice, participants' knowledge, attitudes, and usage patterns may have evolved since the time of data collection. Accordingly, the present findings should be interpreted as a baseline snapshot during an early phase of AI adoption, and future studies using longitudinal or repeated cross-sectional designs are recommended.

Fourth, a conceptual limitation was identified within the perception domain, which contained heterogeneous items assessing perceived data accuracy, educational value, and ethical concerns, including belief-based evaluations of AI accuracy rather than objectively verifiable knowledge items. This conceptual breadth likely contributed to the moderate internal consistency (Cronbach's α = 0.592). Future surveys should differentiate more clearly between belief-based perceptions, objective knowledge, and attitudinal constructs to enhance psychometric precision and interpretability.

Additionally, the practice domain did not impose a fixed recall window (e.g., past month or past year), which may have resulted in variability in how participants interpreted “typical” AI use. Similarly, some contingency table cells contained small counts, which may affect the appropriateness of chi-square testing for those specific comparisons and should be interpreted with caution. In addition, some subgroup comparisons involved relatively small participant numbers, which may reduce statistical power and increase the risk of unstable estimates. Findings from these smaller subgroups should therefore be interpreted with caution Finally, given the number of statistical tests performed, the possibility of Type I error inflation cannot be ruled out. Future research with larger samples and, where appropriate, statistical correction procedures (such as Bonferroni adjustments) is warranted.