A latent profile analysis of artificial intelligence literacy among undergraduate nursing students: a cross-sectional study

Abstract

Aims

To understand the current status of artificial intelligence (AI) literacy among undergraduate nursing students through latent profile analysis, identify potential subgroups and their population characteristics, and analyze the influencing factors of different profile categories.

Design

A cross-sectional study.

Methods

The study utilized 686 undergraduate nursing students from an undergraduate college in Guangdong Province as research subjects. The study employed a number of research tools, including demographic characteristics, the Artificial Intelligence Literacy Scale (AILS), and the Artificial Intelligence Self-Efficacy Questionnaire. A latent profile model of undergraduate nursing students’ AI literacy was analyzed using Mplus 8.3. The influencing factors of each profile model were analyzed by multiple logistic regression analysis.

Results

686 nursing students were finally included. Undergraduate nursing students’ AI literacy score was (64.69 ± 11.00). Undergraduate nursing students’ AI literacy could be categorized into three latent profile analysis: low AI literacy group (18.80%), moderate AI literacy group (58.50%) and high AI literacy group (22.70%). Logistic regression analysis showed that grade, only child status, mother’s education level, whether interested in AI technology, frequency of AI technology use in the past 3 months, whether used AI tools and AI self-efficacy were the influencing factors of potential categories among undergraduate nursing students’ AI literacy (P < 0.05).

Conclusion

The findings revealed the heterogeneity of AI literacy among undergraduate nursing students and could guide the identification and early intervention of undergraduate nursing students with low AI literacy.

Clinical trial number

Not applicable.

Introduction

With the deep integration and broad application of artificial intelligence in healthcare—encompassing voice recognition systems, computer vision applications, and generative AI models such as ChatGPT—the medical field is undergoing a profound intelligent transformation [1]. Applications range from AI-assisted image diagnosis and personalized treatment recommendations to robotic surgery and chronic disease management. In response to this trend, the World Health Organization (WHO) has issued new competency requirements for future healthcare professionals [2]. As a core component of medical practice, nursing is increasingly intersecting with AI, making the integration of AI into nursing an irreversible trend [3]. For instance, machine learning models are being used to optimize nursing resource allocation, natural language processing aids in automated nursing documentation, and smart wearable devices enable remote monitoring of patients with chronic conditions. Given the complexity of AI technologies, along with ethical considerations and the necessity to harmonize technology with humanistic care, future nurses are expected not only to serve as care providers but also to act as collaborators with intelligent medical devices and data-driven clinical decision-makers [4]. Therefore, cultivating AI literacy among undergraduate nursing students—who represent the future backbone of the nursing workforce—has become a critical objective for nursing education reform worldwide [5].

AI literacy is a multidimensional construct that extends beyond mere technical proficiency. It encompasses understanding basic AI concepts and principles, competency in applying AI technologies, critical awareness of ethical and security issues, and the ability to collaborate effectively with AI systems to solve practical problems [6]. A study conducted in China reported moderate levels of AI literacy among nursing students [7], while research in Turkey found that although nurses’ AI literacy was at a moderate level, actual AI adoption remained low [8]. Evidence suggests that higher AI literacy not only enhances human-AI collaboration capabilities, enabling nurses to better adapt to intelligent healthcare environments [9], but also contributes to improved patient safety and care quality [10]. At the same time, AI self-efficacy, rooted in Bandura’s self-efficacy theory, refers to an individual’s confidence in their ability to perform AI-related tasks, such as understanding AI concepts, operating AI tools, and solving AI-related problems [11]. This belief plays a key role in motivating behavioral choices, effort investment, and persistence [12]. Students with high AI self-efficacy are more likely to actively engage with AI technologies and demonstrate greater adaptability in human—AI collaborative scenarios [13]. Previous studies have identified a significant positive correlation between AI literacy and self-efficacy [14], indicating that self-efficacy may serve as both a core component and a psychological driver of AI literacy. This is further supported by recent nursing-specific research, which highlights how AI self-efficacy can mitigate related anxiety, underscoring its critical role in shaping adaptive psychological responses to AI technology [15].

Although AI literacy has garnered increasing research attention, most existing studies rely solely on scale scores for assessment, neglecting the heterogeneous nature of individual differences among students. This oversight limits the development of targeted educational strategies. Latent profile analysis (LPA), a person-centered approach that classifies individuals into subgroups based on similar response patterns [16], can help identify distinct profiles of AI literacy and facilitate group-specific interventions. Therefore, this study employs LPA to identify latent profiles of AI literacy among undergraduate nursing students and to examine their influencing factors. The findings aim to provide an evidence-based foundation and design tailored educational interventions for enhancing AI literacy, thereby supporting the advancement of nursing education.

Methods

Participants and sample size

From March to April 2025, a convenience sample of undergraduate nursing students was recruited from a medical college in Guangdong Province, China. Sample size estimation followed Kendall’s principle, which recommends that the sample size be 10–20 times the number of independent variables, with an additional 20% allowance for potentially invalid responses [17]. This study included 17 items from a general information questionnaire, 4 dimensions from the Artificial Intelligence Literacy Scale (AILS), and 6 items from the Artificial Intelligence Self-Efficacy Questionnaire, resulting in a total of 22 variables. Thus, the estimated required sample size ranged from 275 to 550. Furthermore, based on recommendations by Nylund-Gibson [18], a minimum sample size of 300 is advised for latent profile analysis to ensure model stability. The final sample included 686 participants, meeting both criteria. Inclusion criteria were: (1) age ≥ 18 years; (2) enrollment as a full-time undergraduate nursing student; and (3) provision of informed consent. Exclusion criteria included: (1) absence, leave of absence, or other circumstances preventing questionnaire completion; (2) completion time < 120 s; (3) incomplete responses or obvious response patterns (e.g., straight-line answering).

Measurements

Sociodemographic characteristics

A self-administered general information questionnaire was developed based on a literature review [8, 19, 20] and research group discussions. It collected data on the following 17 variables: age, gender, grade, residence, whether to be an only child, whether to serve as student leader, family average monthly income, father’s education level, mother’s education level, whether had research experience, whether interested in AI technology, whether participation of relevant courses on AI, frequency of AI technology use in the past 3 months, used AI time per day, whether hope adding AI content to the curriculum, whether used generative AI and perception of AI’s impact on nursing.

Artificial Intelligence Literacy Scale (AILS)

The AILS, was developed by Wang [21] in 2024 based on Wang’s [22] AI literacy framework. Participants were instructed to respond to the items with their experiences and understanding of contemporary, widely accessible AI—primarily generative AI tools (e.g., ChatGPT)—in mind, making it an appropriate instrument for assessing literacy in contexts where generative AI tools are prevalent. It consists of 13 items grouped into four dimensions: awareness of AI, usage of AI, evaluation of AI, and ethics of AI. Items are rated on a 7-point Likert scale from 1 (“strongly disagree”) to 7 (“strongly agree”). Total scores range from 13 to 91, with higher scores indicating greater AI literacy. In this study, the scale demonstrated high internal consistency, with a Cronbach’s α of 0.911.

Artificial Intelligence Self-Efficacy Questionnaire

This instrument, originally developed by Carolus [23] and translated into Chinese by Tseng et al. [24], was used to evaluate AI-related self-efficacy among nursing undergraduates. It comprises 6 items rated on a 5-point Likert scale (1 = “strongly disagree” to 5 = “strongly agree”). Total scores range from 6 to 30, with higher scores indicating greater self-efficacy. The Cronbach’s α for this scale was 0.900 in the present study.

Data collection and quality control methods

Data were collected online via the Wenjuanxing platform (https://www.wjx.cn) between March and April 2025. After obtaining approval from college administrators, a questionnaire link with a QR code was distributed to all 768 nursing undergraduate students of the college. The landing page outlined the study purpose, significance, instructions, and data confidentiality agreement. Participation was voluntary and required informed consent. To ensure data quality, the platform was configured to allow only one response per IP address, and all items were mandatory. Out of 750 distributed questionnaires, 686 valid responses were retained after excluding those with completion times under 2 min or exhibiting patterned responses. The effective response rate was 91.47%.

Statistical analysis

Data were analyzed using SPSS 27.0 and Mplus 8.3. Categorical variables were summarized as frequencies and percentages and compared using chi-square tests. Continuous variables were expressed as mean ± standard deviation or median (interquartile range) and compared using ANOVA or the Kruskal–Wallis H test, as appropriate. LPA was conducted in Mplus to identify distinct subgroups based on AI literacy responses. The fitting indicators of model included: Akaike Information Criteria (AIC), Bayesian Information Criteria (BIC), Sample-Size Adjusted BIC (aBIC), information Entropy, Lo-Mendell-Rubin Likelihood Ratio Test (LMR-LRT) and Bootstrapped Likelihood Ratio Test (BLRT). The smaller the values of AIC, BIC and aBIC, the better the model fit. The closer the information entropy is to 1, the more accurate the classification is. The P values for LMRT and BLRT were < 0.05, indicating that the model with k categories was significantly better than the model with k − 1 categories. Multivariate Logistic regression was used to analyze the influencing factors of the potential profile category of AI literacy of undergraduate nursing students, and P < 0.05 was used to indicate statistically significant differences.

Research hypotheses

This study aimed to investigate the latent profiles of AI literacy among nursing undergraduates and their associated factors, to inform the development of targeted educational strategies. Specifically, it tested the following hypotheses:

1. There are different potential profile categories of AI literacy among nursing undergraduates.

2. Socio-demographic characteristics and AI self-efficacy exhibit associations with subgroup of AI literacy potential profiles.

Ethical approval

This study was approved by the Ethics Committee of the First Affiliated Hospital of Guangdong Pharmaceutical University (Approval number: [2025]ⅡNo.39). Permission was also obtained from the School of Nursing at Guangdong Pharmaceutical University. Adhering to the ethical guidelines, protocols, and regulations outlined in the Declaration of Helsinki and the Measures of the Ethical Review of Life Science and Medical Research Involving Humans, the study ensured that informed consent was obtained from all participants. Participation was entirely voluntary, with individuals having the freedom to withdraw from the study at any point without facing any consequences.

Results

Sociodemographic characteristics

A total of 686 undergraduate nursing students were included in the analysis, with a mean age of 20.16 ± 1.22 years (range: 18–23). Among the participants, 80.5% were women, 37.8% were sophomores, 52.3% were student leaders, and 84.5% were not only child. 67.5% were from urban areas; 35.9% of the family monthly income was within 3001 ~ 6000. A large proportion (84.3%) had no prior involvement in scientific research projects, while 85.9% expressed interest in AI technology. Although 92.1% had used generative AI tools, 85.9% had not received any formal AI-related training. Detailed characteristics are presented in Table 1.

Table 1.

Sociodemographic characteristics (n = 686)

Characteristics	Classification	Number	Constituent ratio(%)
Gender
	Men	134	19.5
	Women	552	80.5
Grade
	Freshman	162	23.6
	Sophomore	259	37.8
	Junior	173	25.2
	Senior	92	13.4
Residence
	Urban	463	67.5
	Rural	223	32.5
Only child
	Yes	106	15.5
	No	580	84.5
Student leader
	Yes	359	52.3
	No	327	47.7
Family average monthly income (RMB)
	≤ 3000	106	15.5
	3001–6000	246	35.9
	6001–10,000	210	30.6
	≥ 10,001	124	18.1
Father’s education level
	Primary school and below	92	13.4
	Junior high school	253	36.9
	Senior school	166	24.2
	College or above	175	25.5
Mother’s education level
	Primary school and below	136	19.8
	Junior high school	248	36.2
	Senior school	166	24.2
	College or above	136	19.8
Research experience
	Yes	108	15.7
	No	578	84.3
Interested in AI technology
	Yes	589	85.9
	No	97	14.1
Participation in relevant AI courses
	Yes	97	14.1
	No	589	85.9
Frequency of AI technology use in the past 3 months
	Everyday	157	22.9
	At least once a week	344	50.1
	At least once a month	65	9.5
	Once in a while, but not more than once a month	102	14.9
	Almost never	18	2.6
Used AI time per day
	< 1 h	530	77.3
	≥ 1 h	156	22.7
The hope of adding AI content to the curriculum
	Yes	575	83.8
	No	111	16.2
Used AI tools
	Yes	632	92.1
	No	54	7.9
Perception of AI’s impact on nursing
	Largest	171	24.9
	Larger	304	44.3
	Ordinary	211	30.8

The scores of the main variables

The overall AI literacy score among participants was 64.69 ± 11.00. Dimension-specific scores were as follows: knowledge (18.00 ± 4.72), usage (15.43 ± 2.91), evaluation (14.95 ± 3.00), and ethics (16.31 ± 2.98). The mean score for AI self-efficacy was 21.37 ± 3.93.

Latent profile analysis and naming of AI literacy

Latent profile analysis was conducted based on the four dimensions of the AI Literacy Scale. Fit indices for models with latent classes are presented in Table 2. The Akaike Information Criterion (AIC), Bayesian Information Criterion (BIC) and adjusted Bayesian Information Criterion (aBIC) decreased consistently as the number of classes increased. The three-class model demonstrated higher entropy (0.832), and both the LMRT and BLRT yielded significant P values, indicating that the three-class model fit significantly better than the one- and two-class models. The P values of LMRT for 4 category models were not statistically significant and the minimum profile proportion is less than 5% [25], suggesting that it is not recommended as a separate profile. After comprehensive comparison of the fitting indexes of each model, the potential profile of AI literacy among undergraduate nursing students was finally determined to be divided into three categories. The results of the latent profile analysis of AI literacy among undergraduate nursing students are shown in Table 3; Fig. 1. Each profile was named based on its score characteristics. Profile 1, with 129 participants (18.80%), had the lowest scores in all dimensions and was labeled the low AI literacy group. Profile 2, with 401 participants (58.50%), had moderate scores across all dimensions and was termed the moderate AI literacy group. Profile 3, with 156 participants (22.70%), had the highest average scores in all dimensions and was designated the high AI literacy group.

Table 2.

Latent profile model fit indices comparison (n = 686)

Model	AIC	BIC	aBIC	Entropy	LMR	BLRT	Category probability
1	14399.233	14435.480	14410.079	-	-	-	-
2	13737.896	13796.798	13755.521	0.778	<0.001	<0.001	0.666/0.334
3	13406.762	13488.318	13431.165	0.832	<0.001	<0.001	0.188/0.585/0.227
4	13281.390	13385.601	13312.572	0.859	0.114	<0.001	0.208/0.017/0.548/0.226

Table 3.

Mean scores and standard deviation for each dimension of the three-profile model of AI literacy (n = 686)

	Low AI literacy	Moderate AI literacy	High AI literacy
AI literacy	50.20 ± 5.96	63.58 ± 5.15	79.50 ± 6.18
Awareness of Al	14.41 ± 3.35	17.20 ± 3.79	22.99 ± 3.81
Usage of Al	11.61 ± 1.82	15.26 ± 1.67	19.01 ± 1.52
Evaluation of Al	10.97 ± 2.00	14.75 ± 1.52	18.76 ± 1.59
Ethics of Al	13.21 ± 2.84	16.37 ± 2.36	18.73 ± 2.10

Fig. 1.

Distribution characteristics of latent profiles in undergraduate nursing students’ AI literacy

Univariate analysis of latent profiles of AI literacy among undergraduate nursing students

The results of the univariate analysis showed that the differences in the distribution of the latent profiles of the undergraduate nursing students of age, grade, whether to be an only child, mother’s education level, family average monthly income, whether interested in AI technology, whether participation in relevant AI courses, frequency of AI technology use in the past 3 months, whether used AI tools, perception of AI’s impact on nursing and AI self-efficacy were statistically significant (P < 0.05). No significant differences were observed for other demographic variables (P > 0.05), as shown in Table 4.

Table 4.

Comparison of sociodemographic characteristics among different profiles of AI literacy among undergraduate nursing students (n = 686)

Variable		Low AI literacy (n = 129)	Moderate AI literacy (n = 401)	High AI literacy (n = 156)	X2/F/H	P
Age (year, )		20.041.26	20.261.20	20.011.22	3.252b)	0.039
Gender	Men	31(24.0%)	67(16.7%)	36(23.1%)	4.943a)	0.084
	Women	98(76.0%)	334(83.3%)	120(76.9%)
Grade	Freshman	39(30.2%)	81(20.2%)	42(26.9%)	16.185a)	0.013
	Sophomore	57(44.2%)	149(37.2%)	53(34.0%)
	Junior	21(16.3%)	108(26.9%)	44(28.2%)
	Senior	12(9.3%)	63(15.7%)	17(10.9%)
Residence	Urban	79(61.2%)	271(67.6%)	113(72.4%)	4.037a)	0.133
	Rural	50(38.8%)	130(32.4%)	43(27.6%)
Only child	Yes	11(8.5%)	63(15.7%)	32(20.5%)	7.814a)	0.020
	No	118(91.5%)	338(84.3%)	124(79.5%)
Student leader	Yes	59(45.7%)	207(51.6%)	93(59.6%)	5.648a)	0.059
	No	70(54.3%)	194(48.4%)	63(40.4%)
Father’s education level	Primary school and below	23(17.8%)	51(12.7%)	18(11.5%)	7.340a)	0.291
	Junior high school	52(40.3%)	151(37.7%)	50(32.1%)
	Senior school	27(20.9%)	94(23.4%)	45(28.8%)
	College or above	27(20.9%)	105(26.2%)	43(27.6%)
Mother’s education level	Primary school and below	28(21.7%)	74(18.5%)	34(21.8%)	18.153a)	0.006
	Junior high school	58(45.0%)	149(37.2%)	41(26.3%)
	Senior school	29(22.5%)	99(24.7%)	38(24.4%)
	College or above	14(10.9%)	79(19.7%)	43(27.6%)
Family average monthly income (RMB)	≤ 3000	24(18.6%)	64(16.0%)	18(11.5%)	15.333a)	0.018
	3001–6000	43(33.3%)	140(34.9%)	63(40.4%)
	6001–10,000	45(34.9%)	130(32.4%)	35(22.4%)
	≥ 10,001	17(13.2%)	67(16.7%)	40(25.6%)
Research experience	Yes	12(9.3%)	69(17.2%)	27(17.3%)	4.970a)	0.083
	No	117(90.7%)	332(82.8%)	129(82.7%)
Interested in AI technology	Yes	90(69.8%)	352(87.8%)	147(94.2%)	37.739a)	<0.001
	No	39(30.2%)	49(12.2%)	9(5.8%)
Participation in relevant AI courses	Yes	13(10.1%)	52(13.0%)	32(20.5%)	7.426a)	0.024
	No	116(89.9%)	349(87.0%)	124(79.5%)
Frequency of AI technology use in the past 3 months	Everyday	19(14.7%)	84(20.9%)	54(34.6%)	70.843a)	<0.001
	At least once a week	49(38.0%)	224(55.9%)	71(45.5%)
	At least once a month	14(10.9%)	39(9.7%)	12(7.7%)
	Once in a while, but not more than once a month	34(26.4%)	52(13.0%)	16(10.3%)
	Almost never	13(10.1%)	2(0.5%)	3(1.9%)
Use AI time per day	< 1 h	105(81.4%)	314(78.3%)	111(71.2%)	4.815a)	0.090
	≥ 1 h	24(18.6%)	87(21.7%)	45(28.8%)
The hope of adding AI content to the curriculum	Yes	103(79.8%)	334(83.3%)	138(88.5%)	4.063a)	0.131
	No	26(20.2%)	67(16.7%)	18(11.5%)
AI tools used	Yes	107(82.9%)	379(94.5%)	146(93.6%)	18.605a)	<0.001
	No	22(17.1%)	22(5.5%)	10(6.4%)
Perception of AI’s impact on nursing	Largest	23(17.8%)	79(19.7%)	69(44.2%)	40.552a)	<0.001
	Larger	63(48.8%)	188(46.9%)	53(34.0%)
	Ordinary	43(33.3%)	134(33.4%)	34(21.8%)
AI self-efficacy score [M(P25, P75)]		18(16, 21)	21(19, 24)	24(23, 28)	184.072c)	<0.001

Note: a) X² = chi-square tests; b): F = one-way ANOVA; c) H = Kruskal-Wallis H; Bold values indicate that p values are < 0.05

Multicollinearity assessment

To evaluate the suitability of selected explanatory and control variables for regression analysis, variance inflation factors (VIF) and tolerance statistics were computed to quantify multicollinearity. Consistent with established methodological standards, VIF values between 0 and 5 indicate acceptable collinearity levels. In this study, VIF statistics ranged from 1.018 to 2.660 (Table 5), confirming the absence of substantial multicollinearity among independent variables.

Table 5.

VIF analysis

Variable	VIF	Tolerance
Age	2.660	0.376
Grade	2.640	0.379
Only child	1.100	0.909
Mother’s education level	1.204	0.831
Family average monthly income	1.148	0.871
Interested in AI technology	1.141	0.877
Participation in relevant AI courses	1.018	0.982
Frequency of AI technology use in the past 3 months	1.198	0.835
AI tools used	1.152	0.868
Perception of AI’s impact on nursing	1.059	0.944
AI self-efficacy	1.127	0.887

Multinomial logistic regression analysis of latent categories of AI literacy among undergraduate nursing students

With the latent profiles of AI literacy as the dependent variable, profile 1 (comprising 18.8% of students) had the lowest proportion and was therefore used as the reference group. Eleven variables found to be statistically significant in Table 4 were used as independent variables. Variable coding is summarized in Table 6. The logistic regression analysis revealed that grade, whether to be an only child, mother’s education level, interested in AI technology, frequency of AI technology use in the past 3 months, used AI tools and AI self-efficacy were the influential factors of AI literacy of nursing undergraduates (Table 7).

Table 6.

Coding method for independent variables

Independent variable	Assignment method
Age	Original value used
Grade	Freshman = 1; Sophomore = 2; Junior = 3; Senior = 4
Only child	Yes = 1; No = 0
Mother’s education level	Primary school and below = 1; Junior high school = 2; Senior school = 3; College or above = 4
Family average monthly income	≤ 3000 = 1; 3001–6000 = 2; 6001–10,000 = 3; ≥10,001 = 4
Interested in AI technology	Yes = 1; No = 0
Participation in relevant AI courses	Yes = 1; No = 0
Frequency of AI technology use in the past 3 months	Everyday = 1; At least once a week = 2; At least once a month = 3; Once in a while, but not more than once a month = 4; Almost never = 5
AI tools used	Yes = 1; No = 0
Perception of AI’s impact on nursing	Largest = 1; Larger = 2; Ordinary = 3
AI self-efficacy	Original value used

Table 7.

The results of the multinomial logistic regression analysis for the latent profiles of AI literacy among undergraduate nursing students (n = 686)

Variables	Profile 2: Moderate AI literacy group (vs. Profile 1: Low AI literacy group)					Profile 3: High AI literacy group (vs. Profile 1: Low AI literacy group)
	β	SE	Wald χ 2	P	OR (95%CI)	β	SE	Wald χ 2	P	OR (95%CI)
Age	-0.039	0.157	0.063	0.802	0.961(0.706 ~ 1.308)	-0.104	0.205	0.257	0.612	0.901(0.603 ~ 1.347)
Grade (ref: Senior)
Freshman	-1.28	0.627	4.174	0.041	0.278(0.081 ~ 0.949)	-0.852	0.816	1.091	0.296	0.427(0.086 ~ 2.110)
Sophomore	-0.942	0.484	3.786	0.052	0.39(0.151 ~ 1.007)	-1.009	0.648	2.421	0.120	0.365(0.102 ~ 1.299)
Junior	-0.206	0.464	0.198	0.657	0.814(0.327 ~ 2.022)	0.052	0.596	0.008	0.931	1.053(0.327 ~ 3.387)
Only child (ref: Yes)
No	-0.836	0.429	3.796	0.051	0.434(0.187 ~ 1.005)	-1.032	0.498	4.297	0.038	0.356(0.134 ~ 0.945)
Mother’s education level (ref: College or above)
Primary school and below	-0.752	0.450	2.787	0.095	0.472(0.195 ~ 1.140)	-0.651	0.543	1.438	0.230	0.521(0.180 ~ 1.512)
Junior high school	-0.694	0.405	2.941	0.086	0.500(0.226 ~ 1.104)	-1.160	0.487	5.684	0.017	0.313(0.121 ~ 0.814)
Senior school	-0.587	0.427	1.887	0.170	0.556(0.241 ~ 1.285)	-0.799	0.506	2.494	0.114	0.450(0.167 ~ 1.212)
Family average monthly income(RMB)(ref: ≥10001)
≤ 3000	-0.019	0.442	0.002	0.966	0.981(0.412 ~ 2.335)	-0.255	0.565	0.204	0.652	0.775(0.256 ~ 2.346)
3001–6000	0.028	0.388	0.005	0.942	1.029(0.481 ~ 2.200)	0.177	0.467	0.144	0.704	1.194(0.478 ~ 2.979)
6001–10,000	-0.17	0.387	0.194	0.660	0.843(0.395 ~ 1.800)	-0.671	0.477	1.979	0.160	0.511(0.201 ~ 1.302)
Interested in AI technology (ref: Yes)
No	-0.465	0.307	2.292	0.130	0.628(0.344 ~ 1.147)	-1.137	0.518	4.808	0.028	0.321(0.116 ~ 0.886)
Participation in relevant AI courses (ref: Yes)
No	-0.078	0.363	0.046	0.830	0.925(0.454 ~ 1.885)	-0.567	0.442	1.642	0.200	0.567(0.238 ~ 1.35)
Frequency of AI technology use in the past 3 months (ref: Almost never)
Everyday	3.206	0.908	12.473	<0.001	24.684(4.166 ~ 146.268)	2.494	1.045	5.696	0.017	12.113(1.562 ~ 93.927)
At least once a week	3.454	0.886	15.209	<0.001	31.631(5.574 ~ 179.483)	2.492	1.026	5.894	0.015	12.080(1.616 ~ 90.29)
At least once a month	3.119	0.935	11.117	0.001	22.617(3.616 ~ 141.461)	2.611	1.109	5.543	0.019	13.608(1.549 ~ 119.59)
Once in a while, but not more than once a month	2.568	0.895	8.235	0.004	13.039(2.257 ~ 75.334)	1.646	1.063	2.399	0.121	5.186(0.646 ~ 41.634)
AI tools used (ref: Yes)
No	-0.897	0.427	4.412	0.036	0.408(0.177 ~ 0.942)	-0.943	0.606	2.424	0.119	0.389(0.119 ~ 1.277)
Perception of AI’s impact on nursing (ref: Ordinary)
Largest	-0.253	0.337	0.567	0.452	0.776(0.401 ~ 1.501)	0.757	0.422	3.224	0.073	2.132(0.933 ~ 4.871)
Larger	-0.159	0.269	0.348	0.555	0.853(0.503 ~ 1.447)	-0.003	0.370	0.000	0.993	0.997(0.483 ~ 2.056)
AI self-efficacy	0.231	0.038	37.785	<0.001	1.260(1.170 ~ 1.356)	0.588	0.054	117.183	<0.001	1.801(1.619 ~ 2.003)

Note: SE: Standard Error; OR: Odds Ratio; 95% CI: 95% Confidence Interval; ref, reference group

Discussion

The present study revealed that the overall AI literacy score among nursing undergraduates was 64.69 ± 11.00, indicating a moderate level of proficiency. This finding is consistent with results reported by Sharma, S [26], but higher than those documented by Taotao He [27]. First, it is important to acknowledge the sampling context of the present study. Variations in educational environments, curricular content, and emphasis on AI training likely have divergent outcomes [28]. Our data were collected from a single medical university using convenience sampling, whereas He et al. included students from 29 higher education institutions across China, encompassing various academic levels. The participants in our study might have had comparatively better access to or exposure to AI concepts within their specific educational environment. Second, with the widespread availability of the Internet and growing accessibility of generative AI tools in recent years, many students express optimism about integrating AI into nursing curricula, considering that there is an increasing awareness of the need to develop competencies in emerging technologies [29].

Using latent profile analysis, this study identified three distinct profiles of AI literacy: “low”, “moderate” and “high” AI literacy groups. The low AI literacy group, particularly deficient in evaluation and usage, They might uncritically accept or inappropriately dismiss AI-generated recommendations (e.g., diagnostic alerts, medication suggestions) in clinical practice due to an inability to assess their validity. For these students, progressive practical training can be implemented, starting from low-risk scenarios (such as health consultation) and gradually transitioning to high-risk decision-making (such as emergency diagnosis). Students are required to make clinical judgments independently, and then analyze the rationality of AI recommendations. By comparing differences, critical thinking can be developed, information processing skills can be improved, and practical ability to apply AI technology can be improved. Moderate AI literacy group has a certain foundation, but it may not be enough for in-depth and systematic learning of AI. In clinical settings, they may become competent users of specific, trained AI tools (e.g., operating a smart pump) but struggle when faced with novel AI outputs or when required to integrate AI insights into complex, holistic patient care plans. Their performance may be adequate in routine tasks but fragile in unfamiliar or high-stakes situations. They may need more practical opportunities (transfer learning of AI tools, AI simulation of composite clinical scenarios, etc.) to learn how to apply AI technology to specific study and life. The high AI literacy group can better access, communicate, and apply AI tools, and can better leverage evaluative thinking and ethical awareness to use AI-related information critically. This may be because students with more in-depth learning and use of generative AI may be more concerned about its potential limitations or ethical implications, which is often accompanied by a more cautious attitude and ethical risk perception in their application of AI technology [30]. These future nurses are more likely to effectively interrogate AI suggestions, identify potential biases, explain AI-assisted decisions to patients and families, and responsibly integrate algorithmic insights with clinical judgment and patient preferences. Educators can encourage and support students with high AI literacy to guide those with insufficient AI skills through workshops, group learning, or “AI partners” in courses or clinical internships. This not only enhances the overall AI capabilities of the student body but also deepens the understanding and critical thinking of the instructors through the process of “teaching and learning together”. These findings highlight significant heterogeneity in AI literacy among nursing students, underscoring the need for differentiated educational strategies rather than treating the student population as homogeneous.

Nursing students who was lower grade had lower scores for AI literacy. Junior students are often occupied with foundational medical courses and lack clinical exposure, limiting their opportunities to relate AI technologies to real-world nursing contexts—such as smart infusion pumps, AI-assisted diagnostics, or electronic health records. With the practical pressure such as internship demand, academic education, job hunting and so on, senior students will take the initiative to understand and learn AI-related skills to enhance their competitiveness, so as to realize the natural improvement of AI literacy under the “demand driven“ [31]. Nursing educators should introduce the enlightenment module of AI early in the curriculum and gradually incorporate elective courses and practical projects to raise awareness and competence [32, 33]. The only child demonstrated higher levels of AI literacy than non-only child. Based on the resource dilution model [34], family resources such as financial support and emotional investment are dispersed as the number of children increases [35, 36]. One-child families may provide more concentrated and better quality financial and educational support, which may facilitate them exposed to and familiar with various digital technologies and information resources earlier and more frequently, thus cultivating stronger information acquisition ability, technology adaptability and self-learning confidence. This comprehensive “digital native” advantage is reflected in higher literacy performance when interacting with emerging technologies like artificial intelligence. This study showed that students in households with a higher level of maternal education reported higher AI literacy. According to the social role theory, the Chinese family division of labor tends to the traditional model of “men take the lead outside and women take the lead inside“ [37], and mothers usually take more responsibility for family education and provide communication and emotional support to children [38, 39]. Attention should be paid to the cases of woman science and technology leaders (especially in the field of medical science and technology) should be introduced into the curriculum to compensate for the lack of role models in the families of these students. Notably, when interpreting the association of AI literacy with only child status and maternal education, these demographic variables should not be interpreted as an intrinsic advantage for only children or as a disadvantage from families with less educated parents. Educational institutions should not select students based on these demographic indicators. Instead, there should be a stronger awareness that all students should have equal access to educational resources and support, regardless of their family structure. Potential gaps in exposure, encouragement, or confidence that may arise from different family backgrounds should be actively identified and reduced to provide all students with the basic exposure and practical opportunities necessary to develop AI skills.

Interested in AI technology were associated with higher AI literacy. Interest is the driver of learning motivation [40]. Learning behavior changes from “let me learn” to “I will learn” [20]. If nursing students are truly interested in AI technology, they actively search for information, videos, courses, or cases related to AI, which can expand the breadth and depth of their learning. In the process of nursing education, teachers should also pay attention to cultivating student’ interest in AI, hold lectures and competitions on AI related topics, create a scientific research atmosphere to explore AI, and promote the improvement of AI literacy.

The frequency of AI application use was positively associated with classification into the moderate or high AI literacy groups, which is consistent with previous studies [8, 22]. According to experiential learning theory, learning is a cyclic process through concrete experience, reflective observation, abstract conceptualization, and active experimentation [41]. Frequent use of AI applications provides a wealth of specific experience. Users are directly exposed to the functions and interfaces of AI through practical operation, and will naturally compare and think about the behaviors, output results and potential errors of AI in multiple uses— a process associated with a deeper intuitive understanding of AI performance [26, 42]. Universities can build nursing AI virtual simulation laboratories, with built-in various common medical AI software (such as intelligent medication systems, virtual medical record analysis tools, nursing operation simulation human AI, etc.) to provide students with an “immersive” experience environment, and students are required to carry out a large number of simulation operations in and out of class [43, 44]. The proposed strategies are designed not only to build literacy but also to foster a mindset and skill set that can better bridge the gap to future clinical practice, preparing students to adapt and contribute thoughtfully as AI integrates into healthcare environments. For smaller colleges or environments with limited resources, educators can utilize widely available and usually free generative artificial intelligence platforms (such as ChatGPT, Deepseek, and kimi) as well as common office suites with artificial intelligence capabilities to conduct teaching activities. Course content can be designed using these tools to achieve core learning objectives. For example: Students can use chatbots to simulate interactions with patients and evaluate the communication and suggestions of artificial intelligence; they can use AI assistants to draft and revise care plans or literature reviews, and then have classmates and teachers analyze the accuracy, bias, and applicability of the output content [24, 45]. If there is a lack of practical experience in using advanced tools, educators can carefully select high-quality, open-access online courses and case studies on artificial intelligence in healthcare. For example: Conduct group discussions on the ethical dilemmas presented in the cases, simulate scenarios of explaining AI-assisted diagnosis to patients. Students can strengthen the basic concepts, limitations, and ethical impacts of artificial intelligence in the field of nursing through self-study, writing assignments, and group discussions, and develop an evaluation mindset towards artificial intelligence, enabling them to respond wisely when new technologies emerge in the future. Furthermore, each institution can actively seek to establish virtual partnerships with universities or medical institutions that have more resources. This can be done through hosting lectures, sharing online seminars, or using cloud simulation platforms, etc. By collaborating rather than making capital investments, resources can be expanded [46].

Studies have shown that nursing students with high AI self-efficacy are more likely to take the initiative to accept learning challenges, show a stronger sense of purpose, and believe in their ability to learn and harness AI technology to solve problems creatively [47]. It is suggested that nursing educators should not only focus on the teaching of practical knowledge when conducting AI literacy training, but also pay attention to individualized teaching and learning situation analysis, fully refer to students’ existing knowledge level, strengthen positive feedback (such as recognition of learning achievements), and incorporate emotional regulation strategies (such as resilience training and mindfulness) into the curriculum to help build learning confidence [48].

Limitations

Several limitations should be considered when interpreting the findings. First, the use of convenience sampling from a single medical university significantly restricts the generalizability (external validity) of the results. While this approach provided feasible access to a pertinent sample, it limits our ability to confidently extend the conclusions—such as the precise mean AI literacy score, the prevalence of the three latent profiles, and the strength of the identified influencing factors (e.g., the role of being an only child)—to all nursing undergraduates in China. Students from other universities, especially those in regions with different levels of AI technological infrastructure, from vocational colleges, or with different curriculum designs, may exhibit different patterns of AI literacy. Similarly, the emphasis on maternal education and only child status as influential factors may reflect specific socio-cultural dynamics in China, and their relevance might diminish in cultures with different family structures and gender roles. Therefore, we recommend that educators and policy makers do not view these recommendations as a universal prescription, but rather conduct a comprehensive assessment of the local educational environment, student demographics, available resources, and cultural factors before implementing them.

Second, self-reported data are susceptible to recall and social expectation biases, and the latent profile of this study comes entirely from the four subdimensions of the AI literacy scale (awareness, use, evaluation, ethics). While this approach effectively captures the core competency structure as defined by the scale, it may not incorporate a comprehensive profile structure. Future studies are encouraged to incorporate objective evaluation or peer review, multivariate methods to construct profiles to improve the richness and practical relevance of the identified subgroups.

Third, as AI technology evolves rapidly, literacy levels and influencing factors may change over time. Our findings cannot directly speak to how frequently or in what specific ways nurses employ AI technologies in their daily clinical work. Longitudinal studies tracking students into their professional careers are also needed to understand how pre-graduation AI literacy influences post-graduation clinical engagement with AI.

Conclusion

This study employed a person-centered latent profile analysis to reveal the intrinsic heterogeneity of AI literacy among nursing undergraduates. Three distinct profiles were identified: a “low AI Literacy” group characterized by pronounced deficits in evaluation and usage of AI information; a “moderate AI Literacy” group with basic capabilities but lacks in-depth and systematic learning of AI tools; and a “high AI Literacy” group demonstrating advanced competency across all dimensions—awareness, usage, evaluation, and ethics. These three profiles are in grade, only child status, mother’s education level, whether interested in AI technology, the frequency of AI technology use in the past 3 months, whether AI tools used and AI self-efficacy. Nursing educators should incorporate the assessment of AI literacy into the assessment of nursing courses. Based on the LPA classification results, priority should be given to providing targeted training (e.g., add introductory practical courses, progressive practical training, emotional regulation strategies) to nursing students in the low AI group. For the “moderate AI Literacy” profile, case-based simulations and clinical scenarios that require students to integrate AI-derived data into patient care planning can help bridge the gap between basic knowledge and competent application in practice-like contexts. At the same time, teaching or supervisory roles should be assigned to the students in the high AI literacy group to promote peer learning and enhance students’ AI literacy.

Author contributions

Shuyi Zhu: Writing - Original Draft, Conceptualization, Software, Formal analysis, Investigation, Data Curation, Visualization, Project administration. Rui Li: Writing - Original Draft, Methodology, Software, Formal analysis, Data Curation. Xuan Ren: Formal analysis, Writing - Review & Editing. Jiawen Huo: Writing - Review & Editing. Qiqing Tan: Writing - Review & Editing. Xiangdi Hu: Writing - Review & Editing. Lin Wang: Validation, Writing - Review & Editing. Lishan Huang: Writing - Review & Editing, Supervision. Aoxiang Luo: Conceptualization, Investigation, Resources, Data Curation, Writing - Review & Editing, Supervision, Project administration, Funding acquisition.

Funding

This study was funded by Guangdong Joint Training Postgraduate Demonstration Base [Guangdong Teaching Research Letter (2024) No. 1–57], Education and Teaching research and reform project of the Open Online Course Alliance for Universities in the Guangdong-Hong Kong-Macao Greater Bay Area in 2025 [WGKM2025Ⅱ019] and Guangdong Pharmaceutical University 2025 University-level Smart Course Project Construction Project.

Data availability

The data that support the findings of this study are available from the corresponding author upon reasonable request.

Declarations

Ethics approval and consent to participate

This study was approved by the Ethics Committee of the First Affiliated Hospital of Guangdong Pharmaceutical University (Approval number: [2025]ⅡNo.39), and the permission was also obtained from the School of Nursing at Guangdong Pharmaceutical University. The study strictly adhered to the principles of the Declaration of Helsinki. All participants voluntarily participated and gave informed consent. Data collection and use followed the principles of confidentiality and anonymity.

Consent for publication

Not applicable.

Declaration of generative AI

During the preparation of this work, the authors used used Grammarly and Deepseek to in order to improve language. After using this tool, the authors reviewed and edited the content as needed and take full responsibility for the content of the publication.

Competing interests

The authors declare no competing interests.