Knowledge, attitudes, and practices toward artificial intelligence in medicine among iranian physicians and medical students: a cross-sectional survey

Abstract

Background

Artificial intelligence (AI) is rapidly transforming healthcare worldwide, yet the readiness of physicians and medical students to adopt AI—particularly in low- and middle-income settings—remains insufficiently understood. Examining knowledge, attitudes, and practices (KAP) toward AI is essential for informing educational priorities and implementation strategies.

Objectives

This study aimed assessing AI-related knowledge, attitudes, and practices among Iranian physicians and medical students, examine demographic differences, and identify barriers to AI adoption.

Methods

A cross-sectional survey was conducted using an online convenience sampling approach across medical universities and clinical settings in Iran between June and August 2024. A validated questionnaire was developed based on a comprehensive literature review of existing KAP surveys related to artificial intelligence and digital health which assessed AI-related knowledge (3 items), attitudes (27 items), and practices (5 items). The reliability of the questionnaire was confirmed using Cronbach's alpha and ICC. Descriptive statistics and multivariable logistic regression were used to examine associations between demographic variables and KAP outcomes. Statistical significance was set at p-value < 0.05.

Results

A total of 238 participants completed the survey, including 165 physicians (69.7%) and 72 medical students (30.3%). Overall, 61.3% demonstrated good knowledge, 83.2% had positive attitudes, and 58.4% showed good AI-related practice. Medical students had higher mean knowledge scores than physicians (0.62 (SD = 0.32) vs. 0.58 (SD = 0.35); p = 0.025). Younger participants (18–25 years) showed higher mean knowledge than those aged > 65 years (0.70 (SD = 0.29) vs. 0.36 (SD = 0.36); p = 0.036). Although attitudes toward AI were uniformly positive (97.4% considered AI essential in medicine), practice levels were lower, particularly among participants aged 50–65 years (Mean practice score (SD) = 0.52 (0.21); p < 0.01). In adjusted analyses, age 50–65 years was independently associated with lower odds of good practice (AOR = 0.12; 95% CI: 0.04–0.41).

Conclusions

Despite strong support for AI integration among Iranian physicians and medical students, gaps remain between positive attitudes and real-world AI use, especially among older clinicians. Structured, practice-oriented AI training and improved institutional infrastructure are needed to facilitate effective and equitable adoption of AI in Iran’s healthcare system.

Supplementary Information

The online version contains supplementary material available at 10.1186/s13104-026-07851-1.

Background

AI has rapidly become an important part of modern healthcare. It is now used in many areas, including diagnostic imaging, clinical decision support, disease prediction, workflow management, and medical education [1, 2]. Recent advances in machine learning, natural language processing, and large language models have further accelerated the use of AI across different medical specialties. These developments have generated both excitement and uncertainty about how AI will shape future clinical practice [3, 4]. Researches have shown that AI systems can perform as well as, and in some cases even better than physicians.

As AI becomes more widely integrated into healthcare systems around the world, increasing attention has been given to how prepared healthcare professionals are to work with these technologies. Several barriers may limit AI adoption among physicians, including fear of being replaced by AI, limited knowledge and training, financial constraints, and the absence of formal AI content in medical curricula [5]. Surveys from countries such as Syria, Jordan, Pakistan, several European nations, and regions of sub-Saharan Africa consistently show while students and clinicians are generally enthusiastic about AI, their practical knowledge and hands-on experience have remained low [5–7]. For example, a multi-country study in Africa found that most participants believed AI could improve healthcare quality, yet fewer than 10% had ever used AI tools in clinical practice [7]. A comparable gap has also been reported among medical students in Canada, where awareness of AI is high but structured training and practical exposure are still limited [8].

This growing mismatch between strong enthusiasm for AI and limited knowledge or experience—often described as the “optimism–competence gap”—has become increasingly apparent in medical education. Although many students recognize AI as essential for the future of medicine, most training programs still do not offer structured AI education. In addition, misunderstandings about AI safety, reliability, and medico-legal responsibility remain common, as shown in recent scoping reviews of AI curricula [4, 9]. Closing this gap may require both basic understanding of AI concepts and exposure to real-world clinical AI applications.

While high-income countries have invested substantial resources in AI research, clinical implementation, and dedicated educational programs, many low- and middle-income countries (LMICs) face significant challenges due to financial and infrastructural limitations [10]. In these settings, digital transformation is advancing rapidly, but training opportunities and system readiness often have left behind. Studies from LMICs suggest that AI adoption is frequently hindered by limited access to training, weak digital infrastructure, unclear regulatory frameworks, and concerns about algorithm transparency and trust [5–7]. For AI to be used safely and effectively, both healthcare providers and data scientists must understand the basic principles behind AI systems and how AI-generated outputs should be evaluated in clinical contexts [8]. Without targeted capacity-building efforts, the rapid expansion of AI would widen existing global health inequalities.

Although interest in AI is increasing across the Middle East, comprehensive assessments of AI-related knowledge, attitudes, and practices among Iranian physicians and medical students remain limited. Iran has made notable investments in digital health infrastructure, health data systems, and biomedical research [11, 12]. However, it is still unclear how prepared clinicians and trainees are to meaningfully engage with AI tools in everyday practice. Understanding current levels of AI literacy, perceptions, and use is essential for guiding educational reforms, professional development programs, and national health policies.

Therefore, this study aimed to assess knowledge, attitudes, and practices related to AI in medicine among physicians and medical students in Iran. By examining demographic and professional factors associated with AI readiness and identifying perceived barriers to adoption, this study provides evidence to support curriculum development, digital health planning, and the responsible integration of AI into the Iranian healthcare system.

Material and methods

Study design and participants

This cross-sectional study was conducted in Iran between June and August 2024 and used a non-probability, convenience sampling method. The online Porsline questionnaire link (https://porsline.ir/) was disseminated through predefined digital channels, including: Professional WhatsApp groups of physicians affiliated with Masih Daneshvari Hospital (NRITLD), Loghman Hakim Hospital, Taleghani Hospital, and private pulmonology and internal medicine networks, Telegram channels of medical students, interns, and residents at Shahid Beheshti University of Medical Sciences, Tehran University of Medical Sciences, and Shahrekord University of Medical Sciences, University mailing lists and student research committee networks and Clinical and educational networks related to the Ministry of Health.

Participants were eligible and included the study if they were: medical students enrolled in Iranian universities (pre-clinical, clinical, or internship level) or licensed physicians (general practitioners, residents, or specialists) currently active in clinical, teaching, or research settings in Iran, aged ≥ 18 years at the time of participation, able to read Persian fluently and access the questionnaire via digital platforms (WhatsApp, Telegram, or email) and confirmed to sign the provided electronic informed consent before initiation of the survey.

Participants were excluded if they: Submitted incomplete responses (> 20% missing data) or abandoned the survey before the final page (excluded via listwise deletion), displayed nonsensical, repetitive, or patterned answers (e.g., identical responses across Likert items, inconsistent demographic information), reported non-medical occupations or resided outside Iran or participated in the pilot reliability testing (n = 30) to avoid duplication.

Sample size calculation and sampling process

The minimum required sample size was estimated using the prevalence-based method recommended by Bujang (2021) for cross-sectional studies [13]. Assuming an expected prevalence of 50% (a conservative estimate yielding the maximum sample size), a 95% confidence level, and a precision of 0.07, the required sample size was approximately 196 participants. A total of 424 individuals accessed the survey link, of whom 238 completed the questionnaire (response rate = 56.1%). No financial incentives were offered Fig. 1.

Fig. 1.

Participant flow diagram

Questionnaire development and validation

The questionnaire was developed based on a comprehensive literature review of existing KAP surveys related to AI and digital health. Databases searched included PubMed, Scopus, Web of Science, and Google Scholar. An initial pool of items was generated and refined through expert review.

A panel of 10 experts (eight clinical specialists and two epidemiologists) evaluated the items for content relevance and clarity. Content validity was acceptable, with Content Validity Index (CVI) ranging from 0.82 to 1.00 and Content Validity Ratio (CVR) ranging from 0.90 to 1.00. A pilot test was conducted with 30 participants to assess clarity and feasibility.

The final questionnaire consisted of four sections: (1) Demographics; (2) Knowledge domain consisted of three binary items measuring participants’ general and applied understanding of AI (e.g., familiarity with AI, recognition of its medical applications, and awareness of AI-based tools such as Grammarly). Each correct response was scored 1 and incorrect 0. The total knowledge score ranged from 0 to 3. (3) Attitude domain consisted of twenty-seven 5-point Likert and categorical items, encompassing trust, perceived usefulness, ethical concerns, and educational opinions about AI usage. (4) Practice (five items, binary or nominal). Item Q31 was reverse-coded. Reverse-coded items included questions expressing negative or over-confident views about AI. (Supp. 1).

Internal consistency of the attitude domain was satisfactory (Cronbach’s alpha = 0.87). Test–retest reliability for the knowledge domain was assessed in a subsample of 30 participants over a 14-day interval using a two-way mixed-effects, absolute-agreement intra-class correlation coefficient (ICC), (ICC = 0.81; 95% CI: 0.74–0.88). Agreement for practice items was moderate (Cohen’s kappa = 0.64).

Reliability of the questionnaire was evaluated using complementary approaches. Internal consistency of the attitude domain was examined using Cronbach’s alpha, which showed good reliability (α = 0.87). To assess test–retest stability, a subsample of 30 participants completed the questionnaire twice, with a 14‑day interval. ICC was calculated using a two‑way mixed‑effects, absolute‑agreement model, demonstrating excellent reliability for the knowledge domain (ICC = 0.81; 95% CI: 0.74–0.88). For practice items, which were binary, Cohen’s kappa = 0.64 indicated moderate agreement. Collectively, these results confirmed satisfactory stability and internal consistency of the instrument.

Statistical analysis

Statistical analyses were conducted using R version 4.3.2 and SPSS version 27.0. Continuous variables were summarized using mean ± standard deviation or median with interquartile range, depending on distribution. Categorical variables were presented as frequencies and percentages.

For inferential analysis, the total scores for knowledge, attitude, and practice were categorized into “good” versus “poor” using a threshold of 50% of the maximum possible score, consistent with previous exploratory KAP research. This method has been commonly applied to classify respondents into ‘good’ versus ‘poor’ levels for descriptive interpretation and logistic regression analyses [14, 15]. The cut‑off was therefore used for analytical convenience rather than as a validated standard.

Associations between demographic variables and each KAP domain were assessed in univariate analyses using chi-square tests, Fisher’s exact tests, or non-parametric tests as appropriate. Variables with p < 0.10 in univariate screening were considered for multivariable modeling.

Separate multivariable logistic regression models were developed for good knowledge, positive attitude, and good practice. Age group and gender were included in all models regardless of their univariate significance as priori confounders. Adjusted odds ratios (AOR) with 95% confidence intervals were reported. Statistical significance was defined as p < 0.05. Given the exploratory nature of the study, emphasis was placed on effect sizes and confidence intervals rather than multiple-comparison adjustments.

Ethics approval

The study protocol was approved by the Institutional Review Board of Mashhad University of Medical Sciences (IR.MUMS.REC.1403.123). Electronic informed consent was obtained from all participants prior to data collection. Participation was voluntary and anonymous, and no incentives were provided. The study adhered to the principles of the Declaration of Helsinki. All procedures involving human participants were conducted in accordance with institutional ethical standards. To protect participant privacy, all survey data were fully de-identified at the point of collection; IP addresses were automatically removed, and no personal identifiers were stored. Data were securely stored on a password-protected institutional server with access restricted to the study investigators only. In accordance with institutional data governance policies, the study data will be retained for a period of five years following publication and then securely deleted. De-identified data may be shared upon reasonable request to the corresponding author, subject to institutional approval and ethical considerations Table 1.

Table 1.

Structure and psychometric characteristics of the questionnaire

Domain	No. of items	Response type	Score range (min.–max.)	Reliability indicator
Demographics	7	Nominal	–	–
Knowledge*	3 (Q1–Q3)	Binary (0/1)	0–3	ICC = 0.81 (95% CI 0.74–0.88)
Attitude^	27 (Q4–Q31)	Likert (5-point)	27–135	Cronbach’s α = 0.87
Practice#	5 (Q32–Q37)	Binary / Nominal	0–5	Cohen’s κ = 0.64

ICC: intraclass correlation coefficient; CI: confidence interval.

*The knowledge domain assessed participants’ factual understanding and awareness of AI, including basic concepts, recognition of AI applications in medicine, and familiarity with commonly used AI-based tools.

*Knowledge items scored 0 = incorrect/1 = correct.

^Attitude items scored from 1 (strongly disagree) to 5 (strongly agree).

^The attitude domain evaluated participants’ beliefs, perceptions, and feelings toward AI in medicine. This included perceived usefulness, trust, ethical concerns, perceived risks, expectations about AI’s role in healthcare, and opinions regarding AI education and integration into clinical practice.

#The practice domain measured participants’ real-world use and engagement with AI tools in clinical, educational, or research settings. This domain reflects actual behaviors and experiences rather than intentions or theoretical acceptance of AI.

#Practice items scored as 0 = no/1 = yes. And nominal questions according to the answer.

Results

Participant characteristics

A total of 238 individuals completed the questionnaire, of whom 120 participants (50.6%) were male. Most respondents were between 25 and 50 years old (59.2%). 17.4% were aged 18–25, 17.9% were aged 50–65, and 4.2% were older than 65. Overall, 165 (69.7%) participants were graduate physicians and 72 (30.3%) were undergraduate medical students. Table 2 presents the baseline characteristics of participants.

Table 2.

Baseline Characteristics of Participants

Characteristic	Category	Frequency	Percentage (%)
Gender	Male	120	50.6
	Female	117	49.4
Age	18–25	41	17.4
	25–50	142	59.2
	50–65	43	17.9
	> 65	10	4.2
Qualification Level	Graduate physician	165	69.7
	Undergraduate medical student	72	30.3
Undergraduates’ year of study	Year 1	17	23.9
	Year 2	15	21.1
	Year 3	6	8.5
	Year 4	1	1.4
	Year 5	4	5.6
	Year 6	11	15.5
	Year 7	17	23.9
Graduate status	Not faculty member	112	67.8
	Assistant professor	28	16.9
	Associate professor	15	9.0
	Professor	9	5.4
Years of experience (graduates)	< 5 years	47	28.7
	5–10 years	19	11.6
	10–20 years	38	23.2
	> 20 years	60	36.6
Workplace*	University (academic setting)	32	12.5
	Ministry of Health	3	1.1
	Research center	17	6.6
	Private clinic	54	21.1
	Government clinic	35	13.7
	Educational/clinical hospital	79	30.9
	Private hospital	35	13.7

Workplace was a multiple‑response item; therefore, totals exceed the number of participants. (Supp 5.)

The total scores for knowledge, attitude, and practice were categorized into “good” versus “poor” using a threshold of 50% of the maximum possible score determining the distribution of participants across the knowledge, attitude, and practice domains. Overall, 61.3% demonstrated good knowledge, 83.2% reported positive attitudes toward AI in medicine, and 58.4% showed good AI-related practice. (Supp. 6).

Descriptive results of knowledge, attitude, and practice

Knowledge

Most respondents (77.6%) reported being familiar with AI, and 67.5% were aware of its applications in medicine. According to the applied scoring threshold, 61.3% of respondents were classified as having good knowledge. (Supp. 2).

Attitude

Attitudes were strongly positive. A total of 97.4% believed that AI is essential in medicine, 94.5% agreed that AI provides rapid access to information, and 88.1% supported integrating AI training into medical education. Overall, 83.2% of respondents were classified as having good practice. (Supp. 3).

Responders chose “Education” (n = 118) and “Clinical finding interpretation” (n = 100) as the most preferable areas they think AI can be useful in medical field. “Treatment of the disease” (n = 37) and the “Other choices” (n = 5) were the least frequent answers Fig. 2.

Fig. 2.

Distribution of answers to the question “In which of the following do you think AI could be the most useful?”

Responders mostly believed AI manufacturing companies, legislators, media professionals and the medical community are all responsible for encouraging the use of AI in medical field. (Supp. 8).

Most of the respondents believed AI application is the most necessary to use for general practitioners and urology was the least selected field to apply AI. (Supp.9).

“Lack of familiarity, proper education, and poor information” selected as the most frequent obstacle in using AI in medical field, while, “conflict of interest” (n = 2) was valued as the least influential in this area. (Supp.10).

“Misinformation and errors caused by artificial intelligence” emerged as the most concern of responders about using AI in medicine announced by 74 individuals (Supp.11).

Practice

Despite positive perceptions, actual use of AI tools was low. Most participants (63.0%) rarely or never used AI in their clinical or educational workflow. Only 8.0% reported using AI several times a day. Overall, 58.4% of respondents were classified as having good practice. Among participants who responded the question, 85.2% reported that they found AI easy to use. (Supp. 4).

Distribution of response to the question, “How often do you use artificial intelligence programs?” has been shown in Fig. 3.

Fig. 3.

Distribution of response to the question, “How often do you use artificial intelligence programs?

Table 3. presents mean KAP scores across demographic groups with corresponding p-values. Knowledge differed significantly by age and qualification status. Practice differed significantly by age. Attitude showed no significant differences between groups.

Table 3.

Association of mean Knowledge, attitude, and practice scores with gender, graduate state and age

Groups	Knowledge mean (SD)	Attitude mean (SD)	Practice mean (SD)
Male	0.63 (0.34)	0.68 (0.15)	0.64 (0.17)
Female	0.60 (0.36)	0.65 (0.14)	0.60 (0.20)
Undergraduate	0.62 (0.32)*	0.65 (0.15)	0.65 (0.14)
Graduate	0.58 (0.35)*	0.66 (0.14)	0.60 (0.20)
Age 18–25	0.70 (0.29)*	0.66 (0.16)	0.65 (0.14) *
Age 25–50	0.69 (0.35)*	0.65 (0.15)	0.64 (0.18) *
Age 50–65	0.57 (0.35)*	0.69 (0.11)	0.52 (0.21) *
Age > 65	0.36 (0.36)*	0.63 (0.18)	0.70 (0.11) *

^*P value < 0.05 was considered statistically significant

Item-level analysis

Item-level analysis was performed to identify specific strengths and weaknesses across the knowledge, attitude, and practice domains. The highest-scoring items were those reflecting strong agreement with the benefits of AI, including improved healthcare quality (97.4%) and rapid information processing (94.5%). In contrast, several items showed notably low performance. The lowest-scoring knowledge item was “Grammarly is AI-based,” with only 39.2% correct responses. Among attitude items, only 11.3% agreed that AI could replace medical staff, and 15.6% believed AI threatens patient safety. Within the practice domain, low levels of engagement were evident in items related to safe AI use (30.9%) and the ability of AI to support individualized care (Supp.7).

Univariate analysis

In univariate analyses, several demographic variables showed significant associations with the KAP domains. Knowledge scores differed significantly by both age group and qualification level. Participants aged 18 to 25 years achieved higher knowledge scores than older respondents (p = 0.036), and undergraduate medical students had higher knowledge score than graduate physicians (p = 0.025). Practice scores also varied significantly across age groups, with older participants reporting lower engagement with AI-related activities (p = 0.007).

In contrast, attitude scores did not differ significantly across gender (p = 0.39), age group (p = 0.42), or qualification level (p = 0.80). Gender was not associated with any of the three KAP domains, yielding non-significant results for knowledge (p = 0.47), attitude (p = 0.39), and practice (p = 0.26).

Variables meeting the predefined screening threshold (p < 0.10), including age group for practice and qualification level for knowledge, were subsequently entered into the multivariable logistic regression model alongside age and gender as a priori confounders.

Multivariable logistic regression

In multivariable logistic regression analyses adjusted for age group, gender, and qualification level, no demographic variables were significantly associated with good knowledge or positive attitude toward AI (all p > 0.05). Participants older than 65 years showed marginally lower odds of good knowledge compared with those aged 25–50 years (AOR = 0.26, 95% CI: 0.07–1.03, p = 0.055), and participants aged 50–65 years demonstrated marginally higher odds of positive attitude (AOR = 7.46, 95% CI: 0.92–60.60, p = 0.060) Table 4.

Table 4.

Multivariable logistic regression adjusted for age group, gender, and qualification level

Predictor	Category (vs reference)	Good Knowledge AOR (95% CI)	p-value	Positive Attitude AOR (95% CI)	p-value	Good Practice AOR (95% CI)	p-value
Age group	18–25 vs 25–50	0.88 (0.28–2.74)	0.826	0.67 (0.14–3.13)	0.609	0.67 (0.14–3.28)	0.618
	50–65 vs 25–50	0.93 (0.41–2.10)	0.857	7.46 (0.92–60.60)	0.060	0.12 (0.04–0.41)	0.001
	> 65 vs 25–50	0.26 (0.07–1.03)	0.055	0.42 (0.09–1.91)	0.262	NE*	–
Gender	Female vs Male	0.69 (0.36–1.31)	0.255	1.28 (0.51–3.18)	0.596	0.20 (0.07–0.59)	0.003
Qualification	Graduate vs Undergraduate	0.74 (0.30–1.82)	0.507	0.67 (0.18–2.48)	0.548	0.88 (0.27–2.84)	0.831

Reference categories: age 25–50 years; male; undergraduate medical student

AOR: adjusted odds ratio; CI: confidence interval; NE: not estimable

^*NE: Not estimable due to quasi-complete separation caused by very small sample size

In the practice domain, participants aged 50–65 years had significantly lower odds, compared with those aged 25–50 years (AOR = 0.12, 95% CI: 0.04–0.41, p = 0.001). In addition, female participants had significantly lower odds of AI practice after adjustment for age and qualification (AOR = 0.20, 95% CI: 0.07–0.59, p = 0.003). Also, Qualification level was not associated with good practice (AOR = 0.88, 95% CI: 0.27–2.84, p = 0.831) Table 4.

Discussion

This study provides a comprehensive assessment of knowledge, attitudes, and practices related to AI among Iranian physicians and medical students, offering one of the first national-level insights into AI readiness within Iran’s healthcare and medical education system. Overall, participants demonstrated high awareness of AI and strongly positive attitudes toward its role in medicine. However, actual engagement with AI tools in clinical and educational practice remained limited, indicating a clear gap between favorable perceptions and real-world implementation.

In descriptive and univariate analyses, undergraduate medical students and younger participants appeared to have higher AI knowledge scores than graduate physicians and older participants. However, these differences did not persist after adjustment for age and gender in multivariable logistic regression models. No demographic variable independently predicted good knowledge or positive attitude, suggesting that AI knowledge is influenced more by contextual exposure and training opportunities than by age or professional status.

A considerable proportion of participants reported familiarity with AI and recognized its medical applications, consistent with similar findings from Syria, Jordan, and Pakistan, where healthcare students reported increasing awareness of the expanding role of AI in diagnosis and clinical workflows [1, 5, 6]. The strong agreement that AI will be essential in future clinical practice underscores growing readiness for AI integration within Iran’s healthcare system.

The unadjusted trend toward higher knowledge among students may nonetheless reflect broader changes in medical education. Medical students are increasingly exposed to digital tools, online learning environments, and informal AI-related content, whereas many practicing physicians completed their training before AI became a visible component of medical discourse. Similar patterns have been reported in studies from Pakistan, Jordan, Syria, Nigeria, and Vietnam, where students consistently report high interest in AI despite limited structured training [1, 3, 5, 6, 8]. Because the sample was convenience-based and included many younger participants, this finding should be interpreted with caution and not generalized to all Iranian trainees or clinicians.

Consistent with regional and international studies, attitudes toward AI were strongly positive. Participants in this study expressed broad optimism regarding the benefits of AI in enhancing efficiency, reducing medical errors, and supporting clinical decision-making. This aligns with international evidence showing that clinicians in countries such as France and Vietnam expect AI to streamline diagnostic processes and improve patient safety [1, 7, 8, 16, 17]. Many respondents also recognized AI’s capacity to strengthen healthcare systems in underserved regions, consistent with earlier work demonstrating the utility of AI in low-resource settings [18, 19]. The similarity between Iranian findings and international evidence suggests that enthusiasm for AI exists across different healthcare systems despite limited access to AI tools.

Concerns about job displacement were relatively low in our survey. Only 11.3% of respondents believed AI might replace medical staff, a level markedly lower than reports from Korea (35%) and moderately lower than the Syrian survey (21.5%) [6]. Similar global studies among pediatricians, dermatologists, and psychiatrists have also shown limited fears of workforce replacement [7, 20, 21]. These consistent findings suggest that healthcare workers generally view AI as a supportive tool rather than a replacement, and consider human expertise and clinical judgment essential to medical care. The shared belief that human judgment and ethical oversight are essential may reduce resistance to AI and support its future use in clinical practice.

Despite positive attitudes, actual AI use remained low, a finding repeatedly observed in low- and middle-income countries. In multivariable analyses, lower AI-related practice was independently associated with older age (50–65 years) and female gender. These associations likely reflect differences in access to training, digital confidence, institutional support, and workload rather than lack of interest in AI itself. Similar age-related and structural barriers have been documented in multinational studies and surveys from Pakistan, Vietnam, and Europe, where AI engagement was shaped more by training opportunities and infrastructure than by attitudes alone [1, 7, 8, 21]

The Iranian context adds important nuance to the existing literature. Iran has a centralized, publicly funded medical education system and has invested in digital health infrastructure and national health information systems; however, formal AI curricula and widespread clinical AI implementation remain limited [11, 12]. Unlike some neighboring countries, AI education is not yet systematically embedded in undergraduate or postgraduate medical training. The strong support observed for integrating AI into medical curricula highlights a clear opportunity for national-level intervention. Structured curriculum modules, hands-on workshops, and continuing professional development programs could help bridge the gap between awareness and practice. In parallel, investment in digital infrastructure, regulatory guidance, and institutional support is essential to ensure safe and equitable AI adoption.

Limitations

This study has several limitations that should be considered when interpreting the findings. The use of a non-probability, convenience sampling strategy based on online recruitment through social media and professional networks introduces potential selection bias and limits external validity. Participants with greater digital literacy or interest in AI may have been more likely to respond, potentially leading to overestimation of knowledge and engagement levels. Non-response bias may also have influenced results.

The reliance on self-reported data introduces the possibility of social desirability and recall bias, particularly in the practice domain, where reported behaviors may not accurately reflect real-world clinical use. The cross-sectional design precludes causal inference, and residual confounding may remain despite multivariable adjustment. Additionally, the knowledge domain consisted of a limited number of binary items, restricting its ability to capture the multidimensional nature of AI literacy. Future studies should use more comprehensive assessment tools and longitudinal or interventional designs to evaluate the impact of structured AI training.

Conclusion

In conclusion, Iranian physicians and medical students demonstrated high levels of awareness and strongly positive attitudes toward AI; however, actual engagement with AI tools in clinical practice remained limited, highlighting a clear gap between favorable perceptions and real-world implementation. older age and female gender were independently associated with lower AI-related practice, suggesting that barriers to implementation may be linked to access, training opportunities, institutional support, and contextual factors rather than resistance to AI itself. These findings emphasize the importance of structured, practice-oriented AI training programs that specifically address generational and contextual barriers to adoption. Interventions should focus on improving access to hands-on AI experience, enhancing institutional support, and integrating applied AI competencies into medical education and continuing professional development. Addressing these gaps may facilitate more equitable and effective integration of AI technologies within Iran’s healthcare system.

Author contributions

Concept/design: Elham Nazari, Shadi Shafaghi; Data collection: Zahra Hajimoradi, Hanieh Sadeghi Koupaei, Mahshid Hajiali, Hannaneh Yousefi-Koma; Data analysis, interpretation: Babak Sharif-kashani, Farah Naghashzadeh, Raha Parnia, Farnaz Ahmadi; Revision of the article: Elham Nazari, Shadi Shafaghi, Maryam Hajimoradi; Writing of the manuscript: Ali Bozorg savoji, Maryam Hajimoradi, Masoume Avateffazeli, Rahem Rahmati.

Funding

The study was self-funded.

Data availability

The study data is available from the corresponding author upon reasonable request. In accordance with institutional data governance policies, the study data will be retained for a period of five years following publication and then securely deleted. Due to ethical and confidentiality considerations, the data have not been deposited in a public repository.

Declarations

Ethics approval and consent to participate

Not applicable.

Competing interests

The authors declare no competing interests.