An Investigation Into the Comprehension, Perceptions, Application, and Determinants Affecting the Utilization of Deepseek Artificial Intelligence by Healthcare Professionals: A Cross‐Sectional Study Authors: Mei Cha, Qin Zhang, Linlin Yu, Hairong Wang, Erdan Luo, Nan Huang, Wenjie Qing, Wenxu Yang, Mengjun Luo, Yiting Du, Yuanhu Lu, Haibo Yao, Jinghua Ye, Yinghong Fan, Qionghua Huang, Xiaobing Ren, Jie Lu, Liangyin Deng, Jianghui Cai, Yonghong Lin Published: 2025 ABSTRACT ABSTRACTBackground and ObjectiveDeepSeek LLM, launched on January 5, 2024, integrates natural language processing, data analytics, and medical picture analysis. The use of it among frontline healthcare personnel in Western China has not been previously examined. This study evaluates Chengdu clinicians' knowledge, attitudes, and usage habits on DeepSeek while identifying significant impacting factors.MethodsA WeChat‐based survey was administered to physicians, nurses, medical technicians, and administrative personnel at two tertiary maternity and pediatric hospitals in Chengdu from February 24 to 26, 2025. The gathered data encompassed demographics, educational attainment, job designation, familiarity with DeepSeek, frequency of use, openness to training, and preferred functionalities. Of the 865 distributed questionnaires, 788 valid replies (91.10%) were analyzed using SPSS 25. Group disparities were analyzed using chi‐square tests, and forward likelihood‐ratio logistic regression determined predictors of awareness, utilization, and willingness to undergo training (p < 0.05).ResultsA total of 94.2% of participants indicated awareness of DeepSeek, predominantly via media sources (83.65%) and peer endorsements (49.53%). Usage was indicated by 76.14% of respondents, comprising 98.6% of physicians and 89.61% of individuals with a master's degree or higher. The desire to train attained 87.44%, with predominant preferences for hands‐on practice (99.4%), a platform overview (93.1%), and case study talks (90.4%). Univariate analyses revealed substantial disparities in awareness, utilization, and training interest based on education level, job title, and job type (p < 0.05). Logistic analysis indicated that possessing a master's degree or higher significantly enhanced the likelihood of awareness (OR: 8.36; 95% CI: 1.13–61.87; p < 0.05). Clinicians aged 51 and older exhibited a reduced likelihood of utilizing DeepSeek (OR: 0.29; 95% CI: 0.13–0.64; p < 0.05). Individuals with mid‐level professional titles demonstrated increased utilization (OR: 4.08; p < 0.001) and a greater readiness to undergo training (OR: 13.26; p < 0.001), while those holding a bachelor's degree showed a reduced inclination to engage in training (OR: 0.47; p < 0.05).ConclusionThere were substantial deficiencies in the awareness and utilization of DeepSeek among healthcare professionals in Chengdu. Educational attainment, professional status, and occupation significantly influence outcomes. Customized training and assistance are essential to facilitate wider and more successful implementation. Introduction As Artificial Intelligence (AI) technology improves faster, there is more evidence that AI applications are being used more and more in the medical field. These applications include predicting and diagnosing diseases, creating personalized treatment plans, managing patients' health, and efficiently allocating healthcare resources, among other things, showing that AI has a lot of benefits [1, 2, 3, 4]. DeepSeek, a versatile AI model, unveiled its inaugural large‐scale model, DeepSeek LLM, on January 5, 2024, and owing to its robust data processing and analytical capabilities, DeepSeek has revealed substantial potential for utilization within the medical sector [5]. Many healthcare organizations have chosen to use the DeepSeek system to improve the efficiency of diagnosis and treatment as well as the quality of care. This is because it has amazing technological benefits and real‐world results in AI‐driven medical image analysis, accurate disease diagnosis, personalized treatment design, and lowering healthcare costs. Even with the leaps made in AI technology in the realm of healthcare, current studies reveal that many healthcare workers show a lack of profound understanding about AI technology, especially regarding trust in the system, privacy measures, and ethical implications [6, 7]. Additional research has demonstrated that a considerable proportion of healthcare practitioners have not undergone structured training in AI throughout their professional journeys, which consequently results in feelings of being overwhelmed when faced with AI technologies [8]. This difference in thinking might be because healthcare professionals have trouble accepting and understanding AI technologies. They worry about how reliable the technology is, how it will affect ethics, and how it will affect privacy. So, it is very important to look into how healthcare workers learn about the DeepSeek platform, how they feel about it, and how they actually use it in their work. Most prior research has examined AI in medical imaging or focused on diagnostic tools, primarily sampling clinicians in eastern China and major cities like Beijing and Shanghai. As a result, how general‐purpose AI platforms such as DeepSeek are adopted in Western hospitals remains largely unknown [9, 10]. Relying on Kelman's “Knowledge‐Attitude‐Practice” framework [11], our study is the first to use a systematic, quantitative approach at two tertiary maternity and pediatric hospitals in Chengdu. We surveyed 788 healthcare workers—covering clinical, technical, and administrative roles—to clarify DeepSeek's awareness, usage patterns, and key influencing factors in frontline practice. These findings offer a theoretical basis and practical recommendations for medical institutions planning AI training programs. Objects and Methods Objects of Study The medical staff at one of Chengdu City's two tertiary maternity and pediatric hospitals were specifically chosen to take part in an online fact‐finding survey using the Questionnaire Star software. Between February 24 and February 26, 2025, 865 completed questionnaires were collected. We present the survey's findings as follows. Research Tools We used a self‐made questionnaire after conducting a thorough literature review. It was structured according to the Knowing, Believing, and Doing framework and included important demographic information about the research participants as well as the medical staff's knowledge, attitudes, and use behaviors regarding DeepSeek and the artificial intelligence platform. Six experts—three clinical doctors, one hospital administrator, one healthcare AI researcher, and one psychology researcher—used the Likert scale to rate the validity of the questionnaire on ten different dimensions, including how useful it was, how complete it was, how representative it was, how clear it was, how culturally adaptable it was, how appropriate the response options were, how biased they were, how redundant they were, how accurate the measurements were, and how ethically they were done. We evaluated the questionnaire twice following a critical discussion. The final round of evaluation showed that the Item‐Level Content Validity Index (I‐CVI) for all parts was at least 0.80, and the Scale‐Level Content Validity Index (S‐CVI) for the whole survey was 0.92. A total of 139 healthcare professionals were asked to fill out the questionnaire for the preliminary analysis that came before the formal execution of the questionnaire. The Kaiser‐Meyer‐Olkin (KMO) test and Bartlett's Test of Sphericity both showed that the scale had a KMO score of 0.919 and a chi‐square statistic of 584.433 (p < 0.001). This indicated that the scale had significant construct validity. We found the Cronbach's alpha coefficient for the overall scale, which included the standardized items, to be 0.889. This value proved that the questionnaire was reliable and valid according to the survey's requirements. The study employed a web‐based survey platform (Questionnaire Star) with predetermined start and end times. Electronic invitations were distributed through the hospital's official WeChat public account to recruit participants. We utilized a convenience sampling approach, enrolling medical staff who were both available and willing to complete the online questionnaire during the survey period. This study obtained informed consent from participants and was approved by our hospital's Medical Ethics Committee (No. Scientific Research Commission 2025 (45)) The questionnaire had 32 carefully thought‐out questions covering four main areas: awareness of technology, acceptance of technology, fear of substitution, perceived ethical risk, organizational support, and expectations for future development. It also had three separate groups of control variables: organizational characteristics, individual characteristics, and technology experience. The questionnaire employed a 5‐point Likert scale (1 = entirely disagree to 5 = completely agree) coupled with a multiple‐choice/fill‐in‐the‐blank format, incorporating logical progressions to facilitate differentiated measurement across various demographic segments. The content was delineated into several sections: (1) Basic information section: encompassing details such as hospital name, gender, type of position, practice type, years of experience, hospital grade, education level, title, and more, with the objective of gathering fundamental demographic data and work background of the study subjects. (2) Awareness and use of the DeepSeek platform: this section asked the participants if they knew about the DeepSeek platform, how they learned about it, what they thought were its main functions, how useful they thought it was for the healthcare sector, and how it could make work more efficient. This was done to find out how aware and positive the medical staff were about the platform. (3) Usage Behavior Section: This part looked at how often participants used DeepSeek, how they interacted with it, and how well they thought it was designed for ease of use. This helped us understand how the platform could be used in clinical settings and how the overall user experience was. (4) Training and support: This section looked at how willing participants were to learn how to use the platform, what kind of training they would like to see, and how much help their institutions gave with training. The goal was to find out what kind of education healthcare professionals need and how much support the AI platform has from institutions. Quality Control Before the survey's execution, we put together a focused team made up of five public health authorities, one expert in information systems, and four clinical researchers, which comprised two medical doctors and two registered nurses. During the execution phase of the survey, the team provided prompt elucidations to medical personnel who exhibited a lack of comprehension, thereby ensuring the fidelity of the survey responses. In the data validation phase, questionnaires from people who weren't participating were carefully looked over. This revealed a number of apparent outliers, logical discrepancies within the data, answer patterns that were too consistent (e.g., choosing the same option for all questions), and an excessive lack of critical data, which led to the elimination of some questionnaires. Statistical Analyses The study employed IBM SPSS Statistics 25 software to scrutinize the information and assemble the database. The reliability and validity of the instrument underwent a detailed assessment through the methods involving Cronbach's alpha coefficient, the KMO test, and Bartlett's sphericity test. Measurement data were articulated as mean ± standard deviation (M ± SD). We used frequency and percentage (n, %) to show categorical data, and the Pearson's chi‐square (χ²) test (two‐sided, α = 0.05) to look at differences between groups. Subgroup stratification was analyzed by post hoc test. We also used binary logistic regression with forward stepwise selection (likelihood ratio method, entry p < 0.05, removal p > 0.10) to look at the factors that affect how much people know about and use DeepSeek, as well as how ready they are to go through training. Each regression model was tested by multicollinearity diagnosis, whose variance inflation factor (VIF) with threshold VIF < 5. Results General Information We acquired a total of 788 valid questionnaires throughout this survey, culminating in an effective response rate of 91.10%. The survey revealed that the average age of participants stood at 37.81 ± 8.27 years, with a considerable number being female (88.20%), and primarily, they possessed Bachelor's degree qualifications (66.62%). The average duration of professional experience among respondents was 14.08 ± 9.10 years, with the share of individuals holding intermediate titles exceeding 50% (50.76%). This study's demographic group came from multiple divisions, including pediatrics, outpatient and health services, medical technology, obstetrics and gynecology, and also administration and logistics. The predominant occupational categories were comprised of nurses (46.45%) and physicians (26.78%), with the majority identified as practitioners of Western medicine. Detailed information is presented in Table 1. Table 1General information on medical personnel.Demographic characteristicsNumber of persons(n(%))GenderFemale695 (88.20)Male93 (11.80)Age~ 21176 (22.34)~ 31342 (43.40)~ 41200 (25.38)~ 5170 (8.88)Education LevelCollege and below109 (13.83)Bachelor's degree525 (66.62)Master's degree136 (17.26)Doctoral degree18 (2.28)TitleHigh level147 (18.65)Middle level400 (50.76)Junior209 (26.52)Not have32 (4.06)Type of positionNurse366 (46.45)Surgeon211 (26.78)Medical technician103 (13.07)Administrative staff101 (12.82)Trained/interned/fellowship doctors7 (0.89)DepartmentPaediatrician182 (23.10)Outpatient and Health Care Units126 (16.00)Medical department114 (14.47)Department of gynecology and obstetrics111 (14.09)Administrative and Logistics Section110 (13.96)Emergency department48 (6.09)Department of critical care medicine (medicine)30 (3.81)Adult Surgery24 (3.05)Paediatric surgery20 (2.54)Chinese medicine13 (1.65)Adult Internal Medicine10 (1.27)John Wiley & Sons, Ltd. General information on medical personnel. Healthcare Professionals Are Aware of Deepseek and Its Channels of Dissemination A total of 742 individuals (94.2%) demonstrated awareness of DeepSeek, primarily due to media coverage (83.65%) and referrals from colleagues or acquaintances (49.53%). When it came to functionality, the creative writing feature had the best recognition rate (89.99%). It was followed by accurate translation (80.80%) and data processing (70.14%), refer to Figure 1. Medical Staff Awareness of DeepSeek and Dissemination Channels. (a) The Number of medical staff aware of DeepSeek through various means. (b) The number of medical staff who are aware of DeepSeek's various functions. Attitudes and Willingness of Healthcare Professionals Towards Deepseek 655 of the 742 people who took part in the study and knew about DeepSeek thought it was “beneficial” to the healthcare sector, while only 19 people (2.56%) thought it was “not beneficial.” Also, 516 people (69.6%) thought it “enhanced” productivity, while only 73 people (9.8%) thought it was “not beneficial.” Finally, 532 people (71.8%) thought it “enhanced” productivity, while only 73 people (9.8%) thought it was “not beneficial.” A total of 69.6% of the respondents anticipated that DeepSeek would “enhance” productivity, while merely 73 participants (9.8%) thought it would “not assist”; 532 subjects (71.8%) believed that DeepSeek would “enhance” the quality of healthcare services, with only 43 individuals (5.8%) asserting that it was “not beneficial.” When asked what made work more efficient, 553 participants (75.68%) said that fewer repetitive tasks were the main reason, followed by 54.79% who said that diagnostic accuracy had gone up, 50.41% who said that medical record management had been improved, and 48.91% who said that better communication with patients had become easier. Ultimately, 771 participants (97.84%) expressed a willingness to utilize DeepSeek, whereas only 2.15% indicated that they were “not eager” or “not willing at all” to engage with it. Refer to Figure 2. Medical Staff Attitudes and Willingness to DeepSeek. (a) Attitudes of medical staff towards DeepSeek's help in improving the healthcare sector, efficiency and quality of care. (b) Reasons why medical staff think DeepSeek improves efficiency. (c) Willingness of medical staff to use DeepSeek. Current Status and Functional Requirements of Medical Staff Using Deepseek 600 people (76.14%) have used DeepSeek for work, with 30.33% saying they used it “frequently (on a daily basis or several times within a week)” and 48% saying they used it “occasionally (a few times per month).” 93% of those people accessed the service through mobile apps, 28.67% through official web platforms, 15.67% through WeChat, 3% through localized deployments, and 2.67% through API integrations. 67.5% of those people said it was “user‐friendly,” while only 2.5% said it was “not user‐friendly.” Among the 788 individuals involved in the research, 69.22% indicated they wanted DeepSeek to improve education and training in the medical field, while 68.87% believed it would enhance communication with patients and elevate service quality, 68.18% expressed a desire for it to optimize hospital administration and resource distribution, 65.82% searched for help in making clinical decisions and diagnostics, 65.57% wanted backing in medical research and innovative practices, and 58.67% were hopeful for advancements in the efficiency of processing medical records. Refer to Figure 3. Current status of DeepSeek usage and functionality requirements for medical professionals. (a) Frequency of DeepSeek use by medical personnel. (b) Use of DeepSeek by medical personnel. (c) Evaluation of DeepSeek's ease of use by users in the medical profession. (d) Medical Staff Functionality Requirements for DeepSeek. Willingness and Need for DeepSeek Training Among Medical Personnel Among the 788 research participants, 87.44% expressed a willingness to engage in the training for the DeepSeek AI platform facilitated by their respective units. Out of the 707 research participants, 99.4% indicated a preference for the training to encompass practical exercises, 93.1% desired an overview of the platform's fundamental functions, 90.4% sought inclusion of case studies and applications, and 85.5% requested a focus on data security and privacy protection. Refer to Figure 4. Medical staff demand for DeepSeek training content. Differences in Knowledge, Utilisation, and Willingness to Train for DeepSeek Among Different Groups According to the univariate analyses, there were differences in how aware people were of DeepSeek across groups with different levels of literacy, job titles, and job types (p < 0.05). For example, people with a master's degree and senior positions were the most aware of DeepSeek (ALL p < 0.005). Also, doctors were more aware than nurses (p < 0.008), as shown in Table 2. When looking at how DeepSeek was used in professional settings, there were big differences between groups based on literacy levels, job titles, and job types (p < 0.05). For example, groups with master's degrees and senior job titles had the highest rates of use (ALL p < 0.005), followed by doctors and administrators with high rates of use (p < 0.008), as shown in Table 2. Different age groups, literacy levels, job titles, and job types also showed different levels of willingness to participate in DeepSeek training (p < 0.05). For example, Table 2 shows that people with bachelor's degrees or higher, senior job roles (ALL p < 0.005), and doctors and administrators (p < 0.008) were more likely to participate in training. Table 2Differences in medical staff knowledge, usage and willingness of training of DeepSeek.VariableGroupTotalKnowledgeUsageWillingness to DeepSeek training n (%)χ2 p n (%)χ2 p n (%)χ2 p Age~ 21176162 (92.04)4.4890.213131 (74.43)3.6100.307143 (81.25)8.2820.041~ 31342323 (94.44)256 (74.85)305 (89.18)~ 41200193 (96.50)162 (81.00)180 (90.00)~ 517064 (91.43)51 (72.86)61 (87.14)GenderFemale695653 (93.96)0.4530.501523 (75.25)2.5700.109608 (87.48)0.0110.916Male9389 (95.70)77 (82.79%)81 (87.10)Education levelCollege and below10997 (88.99)13.0370.00173 (66.97)21.634< 0.00182 (75.23)20.223< 0.001Bachelor's degreea 525492 (93.71)389 (74.09)463 (88.19)Master's degree or abovea 154153 (99.35)138 (89.61)144 (93.51)TitleJunior and untitledb 241221 (91.70)12.0880.002172 (71.37)15.845< 0.001196 (81.33)25.051< 0.001Middle levelb 400374 (93.50)298 (74.50)348 (87.00)High level147147 (100.00)130 (88.43)145 (98.64)Type of positionAdministrative staff10197 (96.04)14.3350.00289 (87.61)40.880< 0.00190 (89.11)20.461< 0.001Doctorc 218215 (98.62)191 (98.6)208 (95.41)Medical technician10396 (93.20)74 (71.84)88 (85.44)Nurse366334 (91.26)246 (67.21)303 (82.79)Years of work experience~ 0321301 (93.77)1.2730.736243 (75.70)0.8890.828274 (85.36)2.1300.546~ 11306291 (95.10)232 (75.82)272 (88.89)~ 21116109 (93.96)92 (79.31)103 (88.79)~ 314541 (91.11)33 (73.33)40 (88.89) Note: Post Hoc test results were showed as following: a Compared with college and below, p < 0.005. b Compared with high level title, p < 0.005. c Compared with nurse, p < 0.008; dCompared with medical technician, p < 0.008.John Wiley & Sons, Ltd. Differences in medical staff knowledge, usage and willingness of training of DeepSeek. Note: Post Hoc test results were showed as following: Compared with college and below, p < 0.005. Compared with high level title, p < 0.005. Compared with nurse, p < 0.008; dCompared with medical technician, p < 0.008. Differences in the Needs of Different Groups for Deepseek Training Content The results show that there are differences in the desire for practical training activities based on gender and job position (p < 0.05), with women and people in higher positions showing a greater interest in these activities (p < 0.05; p < 0.017). There is a statistically significant difference in the demand for an introduction to the basic functions of the training platform based on gender and education level (p < 0.05). Men and Master's degree or above have lower demand for this introductory training (p < 0.05; p < 0.017). Moreover, different educational levels, titles, and type of position (p < 0.05) shows different interest in Case Analysis and Application. And persons holding advanced degrees and higher job positions, along with administrative staffs and doctor, demonstrating a heightened interest in evaluating and utilizing training case studies (p < 0.017; p < 0.013); on the other hand, there is no noteworthy statistical difference in the interest for the security and confidentiality of training information among various demographic sectors (p > 0.05), as illustrated in Table 3. Table 3Differences in demand for DeepSeek training content across populations.VariableGroupTotal number of peoplePractical operation exercise (n [%])χ2 p Introduction of basic functions of platform (n [%])χ2 p Case analysis and Application (n [%])χ2 P Data security and privacy (n [%])χ2 p Age~21176153 (86.93)2.0470.563142 (80.68)4.7590.190150 (85.23)4.3070.230138 (78.41)6.7750.079~31342305 (89.18)280 (81.87)271 (79.24)250 (73.10)~41200180 (90.00)174 (87.00)165 (82.50)165 (82.50)~517065 (92.86)62 (88.57)53 (75.71)52 (74.28)GenderFemale695627 (90.2)6.1520.013587 (84.5)3.9220.048567 (81.58)0.9270.336539 (77.55)1.9960.158Male9376 (81.7)71 (76.3)72 (77.42)66 (70.97)Education levelCollege and below10992 (84.40)3.2170.20092 (84.4)6.6320.03676 (69.72)13.5850.00177 (70.64)4.2290.121Bachelor's degree525471 (89.71)448 (85.3)428 (81.52)414 (78.86)Master's degree or above154140 (90.91)118 (76.6)b 135 (87.66)b,c 114 (74.03)TitleJunior and untitled241204 (84.6)a 8.9910.011196 (81.33)2.7760.250189 (78.42)a 6.6000.037186 (77.18)2.8870.236Middle level400361 (90.3)a 333 (83.25)320 (80.00)a 299 (74.75)High level147138 (93.9)129 (87.76)130 (88.43)120 (81.63)Type of positionAdministrative staff10186 (85.15)5.2760.15381 (80.20)1.0270.79583 (82.18)13.6360.00379 (78.22)1.0540.788Doctor218202 (92.66)182 (83.49)194 (89.00)171 (78.44)Medical technician10389 (86.41)86 (83.49)80 (77.67)d 80 (77.67)Nurse366326 (89.07)309 (84.43)282 (77.05)d 275 (75.14)Years of work experience~0321286 (89.10)2.1170.548260 (81.00)2.5520.466269 (83.80)3.4010.334241 (75.08)1.2200.748~11306272 (88.89)260 (84.97)239 (78.10)236 (77.12)~21116102 (87.93)99 (85.34)95 (81.90)92 (79.31)~314543 (95.56)39 (86.67)36 (80.00)36 (80.00) a Compared with High level title, p < 0.017. b Compared with Bachelor's degree, p < 0.017. c Compared with College and below, p < 0.017. d Compared with Doctor, p < 0.013.John Wiley & Sons, Ltd. Differences in demand for DeepSeek training content across populations. Compared with High level title, p < 0.017. Compared with Bachelor's degree, p < 0.017. Compared with College and below, p < 0.017. Compared with Doctor, p < 0.013. Factors Influencing Medical Staff's Knowledge, Use and Willingness to Train on DeepSeek Employing demographic and occupational data as independent variables, alongside factors such as familiarity with DeepSeek, utilization of DeepSeek within professional settings, and the willingness to engage in DeepSeek training as dependent variables, logistic regression analyses were executed utilizing the forward‐stepping (likelihood ratio) approach. Individuals who held a Master's degree or above showing 18.928 times higher awareness level than college and below (p < 0.01), as outlined in Table 4. Additionally, factors such as age, titles, and job classification emerged as key influences in the use of DeepSeek. And individuals aged 51 and older demonstrating lower usage compared to those under 30. While individuals holding high titles made more frequent 4.080 times use of DeepSeek than those with junior roles, and both administrators and doctors showed a higher incidence of DeepSeek utilization compared to nurses at 4.608 times and 2.811 times, respectively (Table 5); furthermore, professional title and literacy levels were identified as significant determinants affecting the willingness to engage in DeepSeek training, with individuals possessing bachelor's degree as well as master or above showing 2.111 times and 3.374 times more willingness to partake in DeepSeek training compared to a college degree or lower study participants, whereas holders of senior titles exhibited a 13.262 times more increased willingness to engage in DeepSeek training compared to their junior and untitled holding counterparts, as illustrated in Table 6. Table 4Results of logistic regression analysis of DeepSeek awareness.Variable β SE Wald p OR (95% CI) Education Level 8.9700.011College and belowReferenceBachelor's degree0.6120.3552.9750.0851.844 (0.920 ~ 3.698)Master's degree or above2.9411.0497.8600.00518.928 (2.423 ~ 147.884)John Wiley & Sons, Ltd. Results of logistic regression analysis of DeepSeek awareness. Table 5Results of logistic regression analysis of DeepSeek usage.Variable β SE Wald p OR (95% CI) Age 9.4450.024~ 21Reference~ 31−0.3390.2721.5570.2120.712 (0.418 ~ 1.213)~ 41−0.5470.3282.7720.0960.579 (0.304 ~ 1.102)~ 51−1.2390.4089.2490.0020.290 (0.130 ~ 0.644) Type of position 32.666< 0.001NurseReferenceDoctor1.0340.24417.947< 0.0012.811 (1.743 ~ 4.535)Medical technician0.2160.2520.7340.3911.242 (0.757 ~ 2.036)administrative staff1.5280.35019.034< 0.0014.608 (2.320 ~ 9.154) Title 13.6130.001Junior and untitledReferenceMiddle level0.3870.2432.5390.1111.473 (0.915 ~ 2.372)High level1.4060.38313.468< 0.0014.080 (1.926 ~ 8.647)John Wiley & Sons, Ltd. Results of logistic regression analysis of DeepSeek usage. Table 6Results of logistic regression analysis of willingness to participate in DeepSeek training.Variable β SE Wald p OR (95% CI) Education level 11.1340.004College and belowReferenceUndergraduate0.7470.2727.5630.0062.111 (1.240 ~ 3.596)Master's degree or above1.2160.4059.0090.0033.374 (1.525 ~ 7.463) Title 12.7320.002Junior and untitledReferenceMiddle level0.2860.2321.5240.2171.331 (0.845 ~ 2.096)High level2.5850.73512.365< 0.00113.262 (3.140 ~ 56.015)John Wiley & Sons, Ltd. Results of logistic regression analysis of willingness to participate in DeepSeek training. Discussion The findings of this investigation revealed that healthcare personnel exhibited a notably elevated awareness rate of 94.2% regarding DeepSeek. The level of knowledge was particularly pronounced within the cohort possessing advanced educational qualifications and higher professional titles. Specifically, the demographic with a master's degree or above demonstrated an impressive knowledge rate of 99.35%, whereas those with a college degree or lower recorded the least knowledge rate at 88.99%. In a similar vein, the awareness rate among individuals holding senior titles reached an exceptional 100%, further indicating that professional designation and educational attainment exert a significant positive influence on awareness of artificial intelligence technology. This observation fits with the findings of Hamedani Z et al. and Perrier E et al. [12, 13], which showed that higher levels of literacy and higher job titles are usually linked with awareness of AI technology. For example, master's and doctoral candidates in medicine are more likely to embrace and understand AI technology than their Bachelor's degree counterparts. This shows how important literacy is for raising technology awareness. The study also found that 76.14% of medical professionals used DeepSeek. Of those professionals, 89.61% had master's degrees or higher, while only 66.97% had college degrees or less. This result suggests that people with more education are more likely to take the initiative and rely on AI technology. This may be because they are better at learning new things and are more open to using new technologies. Notably, the rate of use among doctors rose to 98.6%. Research by Ahmed Z et al. showed that, even though most doctors don't know much about the specific uses of AI, they are generally positive about its prospects in the medical field and want to use AI technology in their future practices [14]. There were some differences in how healthcare professionals used and understood the DeepSeek platform. However, a large majority of the study participants (87.44%) said they would be willing to take relevant training. The demand for DeepSeek training content was different across different demographic groups. The older demographic, particularly individuals aged 51 and above, demonstrated a heightened interest in practical exercises and a foundational introduction to the platform's basic functions, with the demand for practical exercises reaching an impressive 92.86%. According to the Technology Acceptance Model (TAM), both perceived ease of use and perceived usefulness are critical determinants of training needs. Specifically, older medical staff demonstrated lower perceived ease of use due to unfamiliarity with AI system operation procedures, consequently requiring more systematic hands‐on guidance. Across gender dimensions, female medical personnel—particularly nursing staff—exhibited consistently higher training demands than their male counterparts in all training categories, with particularly significant differences in practical operation, case analysis, and application modules. This pattern may stem from their earlier recognition of AI‐assisted decision‐making value through daily nursing practice and patient communication, resulting in stronger perceived usefulness. Furthermore, groups possessing higher educational qualifications exhibited relatively diminished demand for introductory content regarding platform functions and data security measures. The survey data indicates that the cohort with master's degrees and higher displayed a lower demand for foundational platform functionalities, recorded at merely 76.6%, whereas the bachelor's degree cohort exhibited a demand of 85.3%. The idea behind this trend is that people with more education may have a strong technical background and the ability to learn on their own, which means they may not need as much basic educational content. Additionally, individuals occupying higher job titles demonstrated a comparatively elevated demand for training content. The necessity for practical exercises, case analysis, and application was markedly greater among the higher‐title group than among those with junior and intermediate titles, particularly in the realm of case analysis and application, where the demand from the higher‐title group reached 88.43%. This finding suggests that healthcare professionals with higher titles often have to make more complicated clinical decisions. This means that there is a greater need for practical operational exercises and application case studies that use AI technology. The findings of the survey indicate that while 88.3% of the participants in the study perceive DeepSeek as advantageous to the healthcare sector, a notable 11.7% maintain that DeepSeek is merely somewhat beneficial or not beneficial at all. This observation suggests that, despite a predominant inclination among healthcare professionals toward the prospective capabilities and applications of DeepSeek, a fraction remains skeptical regarding its tangible efficacy and utilization. The difference in these examples shows how healthcare professionals may be afraid of new technologies and the unknown when it comes to using them. In‐depth interviews further revealed that this group harbors doubts about DeepSeek's decision‐making accuracy and reliability in complex clinical scenarios and fears potential data breaches when handling sensitive information. Other clinicians worry that the platform could undermine or even replace their professional value, and—given their already busy schedules—they view the extra effort required to learn and use an AI system as an “added burden,” which in turn reduces their willingness to adopt it. To overcome these barriers and improve DeepSeek's clinical uptake, we recommend: strengthening the platform's credibility through clinical demonstrations and comparative pilot studies; enhancing data privacy protections and clarifying legal responsibility frameworks; providing tiered, hands‐on training and deeply integrating DeepSeek with existing information systems to lower usage costs; and emphasizing AI support's added value in performance assessments and career development so that medical staff perceive it as an aid rather than a competitor. Additionally, 85.5% of the people who participated in the study said they wanted training that covered data security and privacy protection. This shows that healthcare workers are very worried about AI technology when it comes to managing data. Additionally, through comprehensive interviews with individual medical practitioners, the author discerned that numerous healthcare professionals harbor profound anxieties regarding technological displacement, privacy safeguards, ethical dilemmas, and trustworthiness. Concerns about technological displacement are mostly caused by the fast development of AI in the medical field. This could lead to changes in the roles and responsibilities of healthcare providers or even partial replacement, which could make their jobs less secure and less satisfying [15, 16]. Privacy protection emerges as another critical source of unease. With AI's role expanding in the healthcare sector, the difficulty of obtaining and processing patient information has increased, causing anxiety about data protection and privacy. Medical professionals are concerned that the retrieval or revelation of individuals' personal health data without permission could jeopardize their privacy rights [17, 18]. Concerning moral issues, using AI to make medical decisions might affect doctors' freedom and ability to make good decisions, leading to a problem of not being clear who is responsible for what. Moreover, the opacity and lack of interpretability of AI algorithms may elicit inquiries regarding fairness and impartiality [19, 20]. The matter of trustworthiness also constitutes a significant element in the anxiety experienced by healthcare professionals concerning AI. Medical staff may exhibit skepticism regarding the reliability and precision of AI systems, particularly when the determinations made by AI conflict with their professional judgment; this distrust may hinder the effective integration and application of AI within medical practice [21, 22]. As a result, enhancing the training for healthcare practitioners becomes essential, especially in fields like data protection, privacy safeguards, ethical matters, and technical knowledge. By addressing these apprehensions, the widespread adoption and application of AI technologies in the healthcare sector can be effectively advanced. DeepSeek's clinical implementation must strictly comply with data protection regulations, including China's Personal Information Protection Law, guaranteeing that sensitive patient information remains on local computers and is encrypted during transfer [23]. The open‐source design renders it vulnerable to adversarial exploitation, potentially resulting in erroneous diagnostic recommendations and jeopardizing patient safety. Consequently, hospitals ought to implement DeepSeek on segregated internal networks, conduct regular adversarial robustness assessments, and employ strategies such as differential privacy to protect sensitive data and uphold system integrity. Although this investigation clarifies healthcare workers' familiarity, attitudes, usage behaviors, and influencing factors related to DeepSeek, several limitations remain. First, data were gathered only from two tertiary maternity and pediatric hospitals in Chengdu, resulting in a small sample and limited institutional diversity that could reduce the applicability of our conclusions. Second, distributing the questionnaire via the hospitals' WeChat account—while boosting participation—may have introduced social desirability bias, encouraging respondents to rate the AI tool more positively. Third, the 72‐h window for data collection, though helpful in reducing recall errors, likely excluded staff on leave or working alternative shifts. Fourth, relying on convenience sampling—despite its efficiency—could compromise the sample's representativeness. Additionally, this study did not assess clinicians' perception of empathy when using DeepSeek, even though empathy plays a vital role in clinical communication and the physician–patient relationship. Research suggests that AI systems equipped with emotion recognition and context‐aware responses can demonstrate empathy levels comparable to or exceeding those of human physicians in certain scenarios [24]; however, DeepSeek focuses mainly on multimodal data processing and decision support, and its ability to convey empathy during clinician–patient interactions remains untested. Future work should expand the geographic and institutional scope, employ diverse recruitment methods (including both online and offline channels) or incorporate social desirability measures to detect bias, extend the data collection period to account for different shift schedules, and use stratified or mixed sampling by role and seniority to ensure balanced representation across professional categories and experience levels. Moreover, subsequent research should measure how DeepSeek expresses empathy in clinician–patient interactions, evaluate its impact on patient satisfaction, emotional support, and trust, and apply these findings to refine training materials and clinical applications so that DeepSeek can deliver greater value in practice. In conclusion, substantial discrepancies exist in medical personnel's knowledge, engagement with the DeepSeek platform, and willingness to partake in training, particularly with respect to cultural background, professional designation, and job classification. Healthcare institutions ought to align the occupational characteristics of medical staff with the promotion of AI technology and tailor the training content and utilization scenarios to facilitate the effective integration of AI technologies such as DeepSeek within the medical domain, ultimately striving to elevate the quality of diagnostic and therapeutic services rendered by medical entities. Author Contributions Mei Cha: writing – original draft, conceptualization, formal analysis, data curation, project administration. Qin Zhang: formal analysis, investigation, writing – original draft, data curation. Linlin Yu: writing – review and editing, formal analysis, data curation. Hairong Wang: writing – review and editing, investigation. Erdan Luo: formal analysis, writing – original draft. Nan Huang: data curation, formal analysis. Wenjie Qing: writing – review and editing, methodology, validation. Wenxu Yang: conceptualization, writing – review and editing. Mengjun Luo: supervision, writing – review and editing. Yiting Du: methodology, formal analysis. Yuanhu Lu: conceptualization, formal analysis. Haibo Yao: investigation, writing – review and editing. Jinghua Ye: data curation. Yinghong Fan: formal analysis, writing – review and editing. Qionghua Huang: investigation, writing – review and editing. Xiaobing Ren: formal analysis. Jie Lu: supervision, formal analysis. Liangyin Deng: conceptualization, investigation, writing – review and editing. Jianghui Cai: investigation. Yonghong Lin: supervision, resources, project administration. Ethics Statement Ethical approval was granted by Chengdu Women's and Children's Central Hospital Ethics Committee. Conflicts of Interest The authors have no conflict of interest. Transparency Statement The lead author Mei Cha, Yonghong Lin affirms that this manuscript is an honest, accurate, and transparent account of the study being reported; that no important aspects of the study have been omitted; and that any discrepancies from the study as planned (and, if relevant, registered) have been explained.