Artificial intelligence-assisted chatbot: impact on breastfeeding outcomes and maternal anxiety

Abstract

Background

Artificial intelligence (AI) is increasingly used in healthcare interventions to provide accessible, continuous, and personalized patient support. This study investigates the impact of a mobile breastfeeding counseling application developed with artificial AI on mothers’ breastfeeding self-efficacy, success, and anxiety levels.

Methods

A quasi-experimental design was employed, involving 60 mothers. Participants were divided into two groups: 30 mothers received AI-based counseling, and 30 mothers were provided a booklet. Data collection tools included a personal information form, Breastfeeding Charting System and Assessment Tool (LATCH), Postnatal Breastfeeding Self-Efficacy Scale, and Beck Anxiety Inventory. Data were collected from mothers who delivered at a state hospital’s obstetrics and gynecology department and were followed for ten days postpartum (postpartum days 1, 3, 7, and 10).

Results

No significant differences were found in the demographic characteristics of the two groups (p > 0.05). Statistically significant improvements were observed in breastfeeding self-efficacy over time for both groups (AI group: f = 36.356, p = 0.000; booklet group: f = 43.349, p = 0.000). At day 10, the AI group scored significantly higher than the booklet group (Z=-2.216, p = 0.027). For breastfeeding success, as measured by the LATCH tool, significant differences were also noted over time for both groups (AI group: f = 68.466, p = 0.000; booklet group: f = 68.088, p = 0.000). At day seven, the AI group outperformed the booklet group (Z=-2.995, p = 0.003). Anxiety levels showed no significant differences between groups.

Conclusions

AI-based breastfeeding counseling positively impacts breastfeeding self-efficacy and success. The findings highlight the potential of AI applications in healthcare. AI-based chatbots can serve as effective tools for breastfeeding education, offering accessible, personalized, and continuous support. The significant improvements in breastfeeding outcomes indicate that innovative AI-assisted interventions can effectively support mothers during the critical early postpartum period. This research demonstrates the feasibility of integrating AI technology into maternal care and serves as a foundation for future studies.

Clinical trial number

Not applicable.

Background

Breast milk, recognized as the ideal source of nutrition for infants for the first six months, provides all the essential vitamins and minerals needed for an infant’s development [1]. Breast milk supports the immune system and protects against infections [2]. Increasing breastfeeding rates worldwide could potentially save the lives of about 800,000 children under the age of five each year [3]. While exclusive breastfeeding for the first six months and continued breastfeeding for up to two years is strongly recommended [2, 4], research shows that many mothers face breastfeeding challenges in the postpartum period [5,6,7]. During the postpartum period, mothers encounter the need to initiate and maintain breastfeeding for the first time, which is a fundamental step in the life of their newborn. Providing support and guidance to mothers during this period is an important responsibility of nurses [8].

Perceived breastfeeding self-efficacy is a key factor influencing breastfeeding duration and success in the postpartum period, encompassing a mother’s determination, emotional readiness, beliefs, and ability to overcome challenges [9]. A positive relationship between breastfeeding self-efficacy and breastfeeding success has been demonstrated. Breastfeeding counseling designed to increase self-efficacy can improve mothers’ breastfeeding outcomes [10]. In a study of breastfeeding mothers, a negative correlation was found between anxiety levels and breastfeeding self-efficacy [11]. Effective breastfeeding counseling has been reported to have a positive impact on breastfeeding initiation and continuation [12]. A study found that prenatal and postpartum care and counseling positively impact breastfeeding success. Among 637 mothers, those receiving the tell-what-you-learn method showed higher breastfeeding success compared to telephone-based counseling [13]. Various forms of breastfeeding counseling have been reported in the literature, including community-based counseling, peer-supported counseling, and formal counseling [14,15,16,17,18,19].

Technological advances, including artificial intelligence (AI)-powered mobile applications, are increasingly being used in healthcare, among other fields. Although AI has traditionally been confined to powerful mainframes and data centers, recent innovations have brought AI capabilities directly to patients [20]. AI-based mobile health (mHealth) applications have the potential to improve healthcare by empowering individuals with chronic or short-term health conditions [21]. One notable innovation is the Chat-Generative Pre-Trained Transformer (ChatGPT) developed by OpenAI. This system collects data from the whole Internet for its training and generates highly accurate responses to user prompts [22]. ChatGPT draws information from diverse online sources, such as books, articles, and websites, and refines its text-generation capabilities through human feedback [23]. It employs deep learning algorithms to mimic human learning and provide human-like responses. However, concerns remain about its reliability, including risks of misinformation, overemphasis on alternative treatments, and inadequate representation of conventional treatment risks [24]. These risks can lead to delays in treatment and recovery. However, when used in a professional setting where healthcare professionals act as decision-makers, the application may offer positive potential [25]. For example, studies suggest that this AI application and its variants can be trained on specific topics to create effective chatbots under expert supervision [26,27,28].

In the coming years, AI technologies are expected to be increasingly integrated into various aspects of nursing care, including general care and counseling [29]. AI systems possess predictive capabilities to help nurses foresee potential clinical situations and design personalized interventions accordingly. These technologies are poised to transform traditional nursing roles by introducing new assessment, diagnosis, and patient support approaches. Within the context of nursing counseling, AI-supported systems like chatbots can serve as complementary tools to nursing care by delivering standardized health information, responding to frequently asked questions, and offering emotional support when nurses are not immediately available. While these technologies do not replace the role of the nurse, they can enhance care delivery by improving accessibility, reducing workload, and promoting continuity of care [30]. Nonetheless, it is critical to safeguard the humanistic elements of nursing, such as empathy and personalized communication, when integrating such technologies into practice [31].

Today, patients are increasingly using such tools to access health information, highlighting the need to establish quality standards and improve and adapt existing content for patient use. Although, literature indicates that the application of this technology in healthcare is becoming increasingly widespread worldwide, a closer examination reveals that current research on the use of AI-based technologies in nursing predominantly focuses on the early stages of technological development, offering limited evidence on their practical impact and implementation [32]. There are studies utilizing AI-based technologies in nursing education (e.g., AI-supported nursing cases, ethical decision-making) [33, 34], patient and/or caregiver training (e.g., for dementia patients) [35], and in enhancing nursing interventions (e.g., improving clinical decision-making for delirium) [36]. However, a review of the literature shows that, to date, no studies on this specific breastfeeding topic have been conducted in Turkey or elsewhere in the world. Given the importance of breastfeeding training and counseling, there is a clear need for further research to explore the potential benefits of training and using mobile applications supported by AI in this area. Therefore, the present study aims to examine the impact of using a mobile breastfeeding counseling application developed with an AI infrastructure on mothers’ breastfeeding self-efficacy, breastfeeding success, and anxiety levels. Accordingly, the following hypotheses were formulated:

H0

AI-assisted breastfeeding counseling does not affect mothers’ breastfeeding self-efficacy, breastfeeding success, or anxiety levels.

H1

AI-assisted breastfeeding counseling affects mothers’ breastfeeding self-efficacy, breastfeeding success, or anxiety levels.

H2

Differences exist between the breastfeeding self-efficacy, breastfeeding success, and anxiety levels of mothers using the AI-assisted breastfeeding counseling application and those mothers in the booklet group.

Methods

Study aim and design

This quasi-experimental study aims to evaluate how breastfeeding counseling provided through an AI-assisted mobile application affects mothers’ perceptions of breastfeeding self-efficacy, breastfeeding success, and anxiety levels. The study was conducted in the postpartum ward and breastfeeding unit of a state hospital. The postpartum ward has a total of 23 beds and employs 14 nurses providing care. The hospital’s breastfeeding unit employs eight nurses, with two breastfeeding nurses assigned during night shifts.

Participants

The study population comprised mothers seeking breastfeeding training at the postpartum services and breastfeeding unit of a state hospital between March and November 2024. The sample included primiparous mothers meeting the inclusion criteria: voluntary participation, no congenital anomalies affecting breastfeeding, gestational age over 37 weeks, and infants aged 0–28 days. The sample size was determined through power analysis, using the study of Tokat and Okumuş [37], which examined the relationship between breastfeeding success and self-efficacy. Accordingly, based on the Breastfeeding Charting System and Assessment Tool scores, the required sample size was calculated as 25 participants per group, for a total of 50 participants, with an effect size of 1.04, an alpha error margin of 0.05, and a power of 95%. To account for potential data loss during data collection, the sample size was increased by 20% for each group, resulting in a planned total of 60 participants, with 30 participants per group.

Data collection instruments

Data were collected using a personal information form, the Breastfeeding Charting System and Assessment Tool, the Postnatal Breastfeeding Self-Efficacy Scale, and the Beck Anxiety Inventory.

Personal information form

This personal information form was developed by the researchers based on existing literature [6, 7, 10]. The 25-item form collected information on mothers’ demographic characteristics (including age, education level, employment status, monthly income, social security, and marital status), obstetric information (including the number of pregnancies, history of miscarriage, history of abortion, number of children, method of delivery, planned pregnancy status, initiation of prenatal care, health checks during pregnancy, satisfaction with prenatal care, pregnancy-related health problems), infant information (including sex, gestational age at birth, birth weight), and breastfeeding characteristics (including planned duration of breastfeeding, timing of first breastfeeding within 24 h of birth, breastfeeding education history, breast-related problems, and health problems) (See supplementary document).

Breastfeeding Charting System and Assessment Tool (LATCH)

The LATCH Scale, developed to assess breastfeeding success, follows a structure similar to the APGAR scoring system [38]. It takes approximately 5–8 min to complete. The scale assesses five criteria: L: Latch on the breast, A: Audible swallowing, T: Type of the nipple, C: Comfort breast/nipple, H: Hold/Help. Each item is scored as 0, 1, or 2. Maximum total score is 10, and higher scores indicate greater breastfeeding success. Yenal and Okumus [39] conducted the Turkish reliability and validity study for cultural adaption. The Cronbach’s alpha reliability coefficient of the original scale was 0.93, while in the Turkish adaptation study Cronbach’s alpha coefficient was 0.95. In the present study, Cronbach’s alpha values were 0.75, 0.81, 0.72, and 0.77 for the first, second, third, and final measurements, respectively.

Postnatal breastfeeding self-efficacy scale

Fourteen-item scale was developed by Dennis [9] and it assesses the perceived breastfeeding efficacy of mothers. It has a 5-point Likert format ranging from “(1) Not at all sure”, “(2) Not so sure”, “(3) Sometimes sure”, “(4) Sure”, “(5) very sure”. Scale’s minimum score is 14 and maximum 70 points. An increase in the mothers’ scores on the scale indicates an increase in their perception of breastfeeding self-efficacy. A validity and reliability study was conducted by Tokat, Okumuş and Denis [40] to determine its suitability for Turkish culture. The Cronbach’s alpha reliability coefficient of the scale was found to be 0.96 by Dennis [9], and 0.86 by Tokat et al. [40]. In the current study, Cronbach’s alpha values were 0.88, 0.96, 0.85, and 0.90 for the first, second, third, and final measurements, respectively.

Beck anxiety inventory (BAI)

The BAI, developed by Beck et al. [41], measures the frequency of anxiety symptoms. This 4-point Likert-type scale consists of 21 items. Scores range from zero (none) to three (severe). Total scores range from a minimum of 0 points to a maximum of 63 points. Score interpretations are as follows: 0–9 indicates minimal anxiety, 10–18 indicates mild anxiety, 19–29 indicates moderate anxiety, and 30–63 indicates severe anxiety. The Cronbach’s alpha reliability coefficient of the original scale was 0.92. Its Turkish language adaptation study was conducted by Ulusoy, Şahin, and Erkmen [42]. Higher scores indicate a higher level of anxiety. The Cronbach’s alpha reliability coefficient was calculated as α = 0.93. In the current study, the Cronbach’s alpha values of the scale were 0.879, 0.744, 0.800, and 0.843 for the first, second, third, and final measurements, respectively.

Breastfeeding counseling booklet

Researchers developed this booklet by reviewing relevant literature and incorporating topics from the Breastfeeding Counseling Practitioner Handbook issued by the Ministry of Health [43]. Topics covered included how to assess effective latch, recommended breastfeeding positions, and common breastfeeding-related problems along with practical solutions. However, the booklet did not include any content specifically related to psychological support, such as guidance on managing anxiety or postpartum emotional well-being. Five pediatric nursing experts evaluated the booklet for content appropriateness and clarity. Based on expert feedback, only structural changes were made before the booklet was finalized. The Breastfeeding Counselling Booklet provided to mothers in the control group included comprehensive information on various aspects of breastfeeding.

Breastfeeding counseling mobile application

This mobile application incorporates the content of the booklet, which was developed by the researchers. The chatbot application was coded by software developers with a background in Computer Engineering. React Native-based application integrates ChatGPT API to provide intelligent responses and personalized guidance to postpartum mothers. The application is designed to address potential challenges, provide customized solutions, and serve as a research tool to analyze maternal queries. The ChatGPT was trained using the custom breastfeeding guide to ensure responses were aligned with the guide’s content.

The application was implemented through Expo, allowing users to access it by scanning a QR code through the Expo Go application on smartphones. The chatbot supports both iOS and Android operating systems, ensuring broad accessibility while leveraging AI-driven conversational capabilities to enhance user interaction.

Procedure

The data were collected between March and November 2024 from mothers who delivered in the obstetrics and gynecology department of a state hospital and were subsequently followed up in breastfeeding training units. The study sample consisted of 60 mothers, with 30 participants in the pre-trained AI chatbot group (30 mothers), and booklet groups (30 mothers), who consented to participate. During the study, six mothers were excluded from the AI group (three due to technical problems preventing the use of the application, and three who became unreachable after delivery). Four mothers were excluded from the booklet group (two reported insufficient time to review the booklet and two became unavailable after delivery). These mothers’ measurements were not included in the analyses due to insufficient data. When mothers withdrew from the study, new mothers who met the inclusion criteria for the study groups were included, adhering to the randomization protocol. The study used random group assignment by handing out two envelopes to mothers who agreed to participate in the research. During the first interview, researchers explained the subject and purpose of the study and obtained informed consent from participating mothers. At this first meeting, both groups completed the personal information form, the breastfeeding charting system and assessment tool, the postnatal breastfeeding self-efficacy scale, and the Beck Anxiety Inventory.

Both groups received standard hospital postpartum breastfeeding training. Following this routine breastfeeding counseling, the study group received access to an AI-based chatbot breastfeeding counseling application compatible with iOS and Android mobile devices, developed by the researchers in collaboration with computer engineers specialized in software development. Study group mothers were instructed to actively use this application for ten days. Mothers were encouraged to use the chatbot daily whenever they had questions or needed support related to breastfeeding. Following the completion of the study, usage data indicated that mothers accessed the chatbot on average three times per day. The chatbot was designed to provide evidence-based informational counseling in response to user-generated questions related to breastfeeding. The booklet group received a researcher-developed breastfeeding counseling booklet after their routine counseling session and were instructed to use this resource for the same ten-day period.

Mothers in both groups completed follow-up assessments using the LATCH, the Postnatal Breastfeeding Self-Efficacy Scale, and the BAI at four timepoints: the first measurement on postpartum day one, the second during their routine hospital visit on day three, the third on day seven, and the fourth on day 10.

Ethical considerations

The purpose of the research and measurement procedures were explained to participating mothers. Written and verbal consent was obtained from the mothers. The researchers obtained the necessary institutional approvals from the relevant hospital. This study was conducted according to the Declaration of Helsinki. The study received ethical approval from the Kafkas University Non-Interventional Clinical Research Ethics committee on February 27, 2024 (approval number: 80576354-050-99/369).

Data analysis

The collected data were analyzed with SPSS 26.0 software. Descriptive statistics included percentages, means, and standard deviations. Normal distribution was assessed using skewness and kurtosis values, with values between − 1 and + 1 indicating normal distribution [44]. Due to the non-normal data distribution across measurement times (p < 0.05), non-parametric analyses were used. The Mann-Whitney U test was used to compare measurements between the AI and booklet groups. The r value was calculated to see the effect size of the significant difference. In the analysis, if the r value is below 0.1, it is considered a small effect, if it is around 0.3, it is considered a moderate effect, and if it is 0.5 and above, it is considered a large effect [45]. Friedman’s test was used to compare the repeated measurements of the 1st measurement (T1-day one), 2nd measurement (T2- day three), 3rd measurement (T3- day seven), and 4th measurement (T4- day 10) tests within the groups. The Wilcoxon signed-rank test with Bonferroni correction was used to determine the measurement times (post hoc) that caused the difference that emerged as a result of this analysis. As four comparisons were conducted, the alpha level was adjusted to 0.0125 (0.05/4) to control for Type I error. Kendall’s W coefficients quantified Friedman’s test effect sizes, interpreted as small (< 0.10), medium (0.10–0.30), or large (≥ 0.30) [45].

Results

The study sample consisted of 60 mothers, equally divided between the chatbot group (n = 30) and the booklet group (n = 30).

Table 1 compares mothers’ demographic characteristics in the study groups. No significant difference was found between the demographic characteristics of the participants in the groups (p > 0.05), which shows that the mothers who participated in the study were homogeneous in terms of demographic characteristics.

Table 1 Comparison of descriptive information of mothers participating in the study

Table 2 compares the information about pregnancies of mothers in the study groups. There was a significant difference between mothers in the groups about their pregnancies only in terms of the sex of the baby (p < 0.05). There was no significant difference in other variables (p > 0.05).

Table 2 Comparison of Pregnancy-Related knowledge of mothers participating in the study

Table 3 presents the results examining the temporal change in the mean scores of mothers in the AI and booklet groups on the “Breastfeeding Self-Efficacy” scale. Accordingly, over time mean scores on the scale revealed statistically significant differences between the four measurement times for both the AI and booklet groups (AI f = 36.356, p = 0.000; booklet f = 43.349, p = 0.000). In the AI group, post hoc comparisons revealed a progressice increase in breastfeeding self-efficacy scores across measurement times, with a moderate effect size (W = 0.182), indicating a meaningful within- group change over time. The mean score of the fourth measurement was significantly higher than those of the first, second, and third measurements. In addition, the mean score of the third measurement was higher than the mean scores of the first and second measurements. Finally, the second measurement mean score was found to be higher than the first measurement mean score. Similarly, in the booklet group, scores increased significantly across time points, also with a moderate effect size (W = 0.217), supporting the effectiveness of the intervention over time. Post hoc comparisons showed that the 4th measurement mean score was higher than the mean scores of the 1st, 2nd, and 4th measurements. In addition, both the third and second measurement mean scores exceeded the first measurement score.

Table 3 Change in the breastfeeding Self-Efficacy scale between measurement times

Intergroup comparisons showed no significant differences in mean scores between the AI and booklet groups for the first (Z=-1.222, p = 0.222), second (Z=-0.053, p = 0.958), and third (Z=-1.522, p = 0.128) measurements. However, at the fourth measurement, the AI group had significantly higher mean scores than the booklet group (Z=-2.216, p = 0.027), with a liarge effect size (r = 0.404), suggesting that the AI-based intervention may lead to more substantial improvements in breastfeeding self-efficacy by the end of the follow-up period.

Table 4 presents the results examining the temporal change in the mean scores of mothers in the AI and booklet groups on the “LATCH”. Accordingly, analysis over time in mean scores on the LATCH revealed statistically significant differences between the four measurement times for both groups (AI f = 68.466, p = 0.000; and, booklet f = 68.088, p = 0.000). For the AI group, post hoc comparisons revealed that the fourth measurement’s LATCH mean score was significantly higher than the first, second, and third measurements. In addition, both the third and second measurement mean scores were higher than the first measurement score. These results indicate an increase in LATCH scores for the AI group over time, with a moderate effect size (W = 0.342). In the booklet group, post hoc comparisons demonstrated that the fourth measurement mean score was higher than all previous measurements. In addition, both the third and second measurement mean scores exceeded the first measurement score. These results indicate an increase in LATCH scores for the booklet group across measurement times, with a moderate effect size (W = 0.340).

Table 4 Change in breastfeeding charting system and assessment tool (LATCH) between measurement times

Intergroup comparisons showed no significant differences in mean scores between the groups for the first (Z=-0.083, p = 0.934), second (Z=-1.895, p = 0.058), and fourth measurements (Z=-1.404, p = 0.160). However, at the third measurement, the AI group demonstrated significantly higher mean scores compared to the booklet group (Z=-2.995, p = 0.003), with a large effect size (r = 0.546).

Table 5 presents the analysis examining the temporal change in the mean scores of mothers in the groups on BAI. Accordingly, no statistically significant differences were found between the mean scores of the first, second, third, and fourth measurements for mothers in both groups (AI: f = 4.800, p = 0.187; booklet: f = 3.985, p = 0.263). Finally, the Mann-Whitney U test, used for intergroup comparisons, revealed no significant differences between the mean scores of the mothers in the groups for Measurement 1 (Z = -1.364, p = 0.173), Measurement 2 (Z = -0.781, p = 0.435), Measurement 3 (Z = -0.937, p = 0.349), and Measurement 4 (Z = -1.426, p = 0.154).

Table 5 Change in Beck anxiety inventory between measurement times

Discussion and conclusion

Discussion

This study evaluates the impact of AI technologies on breastfeeding self-efficacy, success, and maternal anxiety within the context of breastfeeding counseling, essential for maternal and child health. Numerous studies in the literature address the use or appropriateness of AI-based technologies in training for different patient populations [46,47,48,49,50,51]. Campbell et al. [47] reported that ChatGPT when used to train individuals with thyroid disease, provided appropriate responses to patient questions without requiring specific commands. Similarly, AI-based models, including ChatGPT, are increasingly being used to counsel patients on health-related issues [46, 48, 52, 53]. In this study, breastfeeding counseling training content was delivered to mothers through a chatbot created by computer engineers using AI technology. In the literature, both technology-based and non-technology-based training techniques have been used in breastfeeding education to improve mothers’ breastfeeding success and self-efficacy [54,55,56]. However, this study is the first to apply modern technology such as AI in breastfeeding counseling, which is a significant contribution to the field.

Breastfeeding self-efficacy is a critical factor that reflects mothers’ confidence in their ability to breastfeed and plays an important role in increasing breastfeeding rates. Therefore, for breastfeeding continuation and success, it is an essential factor [57, 58], and practices aimed at increasing mothers’ breastfeeding self-efficacy are of great importance. Studies have shown that breastfeeding education using various techniques improves breastfeeding self-efficacy [59, 60]. Among these techniques, technology-based interventions have been reported to be effective in increasing breastfeeding self-efficacy in numerous studies [55, 56, 58, 61]. A randomized controlled trial comparing mobile education and home care found that postpartum mothers in the experimental group had higher breastfeeding self-efficacy than those in the control group [55]. Similarly, in a randomized controlled trial of 40 mothers of infants younger than three months, a smartphone-based educational application had a significant positive effect on breastfeeding self-efficacy [56]. In their quasi-experimental study, Rahimparvar et al. [61] evaluated the effect of a mobile training program on mothers’ breastfeeding self-efficacy in the postpartum period and observed that the mothers’ breastfeeding self-efficacy in the experimental group was higher than the control group 4 and 8 weeks after birth. Seddighi et al. [62] also found that a mobile phone application positively influenced breastfeeding efficacy in a study of 198 mothers. However, while Rahimparvar et al. [61] and Seddighi et al. [62] highlight an important point regarding interactivity with the text message method they use in their studies, they differ from this study in that these messages are sent at specific times rather than when individuals need them. The AI-based chatbot in this study provided continuous support by responding to mothers’ questions anytime, unlike at predetermined intervals. Results showed that while breastfeeding self-efficacy scores improved in both groups over time, the AI group achieved a higher level of self-efficacy at the final measurement with a moderate-to-large effect size (r = 0.404) (Table 3). This suggests that the AI-based intervention had a more substantial impact compared to traditional informational support. Furthermore, within-group analyses showed moderate effect sizes (W = 0.182 for AI; W = 0.217 for booklet), indicating that both interventions were effective in gradually increasing mothers’ confidence in breastfeeding throughout the early postpartum period. This result unveils the overall effectiveness of breastfeeding counseling, as evidenced by score improvements in both groups from baseline.

Breastfeeding assessment is an ongoing process from the start of breastfeeding until discharge, helping mothers enhance their skills and build confidence in meeting their baby’s needs [63, 64]. A study demonstrated that higher breastfeeding self-efficacy was associated with greater breastfeeding success [65]. Similarly, Öztürk et al. [66] found a positive and significant relationship between breastfeeding success scores and breastfeeding self-efficacy scores. Kılıç et al. [67] investigated the effect of virtual reality (VR) breastfeeding training on breastfeeding success and self-efficacy and found that mothers who received VR-based training at 4 and 24 h after cesarean delivery had higher breastfeeding success and self-efficacy than those who received standard breastfeeding training. The results of the present study are consistent with the literature, as breastfeeding success was higher in both groups. The results showed that the counseling provided in both groups had a moderate effect on breastfeeding success (Table 4), with moderate effect sizes (W = 0.342 for AI; W = 0.340 for booklet). This highlights the positive influence of structured educational interventions on maternal breastfeeding behaviors. However, the third measurement, taken at seven days postpartum, indicated that the AI group had higher breastfeeding success than the booklet group, and the difference associated with a large effect size (r = 0.546). This suggests that mothers in the AI group may have used the AI-based mobile application more effectively. This finding underscores the potential of AI-based tools to more effectively enhance early breastfeeding technique, possibly due to their immediacy, accessibility, and tailored content. The lack of a significant difference between the two groups at the final measurement, which was conducted on postpartum day 10, may be related to mothers in both groups adapting to the postpartum period and improving their ability to manage the breastfeeding process and to the natural process of maternal adaptation. It is possible that, by this time, mothers in both groups had become more familiar with the breastfeeding process and developed their own coping strategies, regardless of the type of support received. This may have contributed to a leveling of outcomes between the groups, highlighting the importance of early and intensive support in the initial postpartum days. The AI interventions offer early advantages in terms of accessibility and responsiveness, but their sustained impact might require reinforcement or combination with emotional and social support mechanisms to maintain long-term effectiveness.

A meta-analysis found that 8.5% of mothers experience anxiety disorders during the postpartum period [68]. Research suggests that maternal anxiety during this time negatively affects breastfeeding initiation and continuation, as well as exclusive breastfeeding [69]. Melo et al. [11] explored the relationship between breastfeeding self-efficacy and both trait and state anxiety, observing that postpartum women with lower levels of trait anxiety had higher breastfeeding self-efficacy. Mothers with lower state anxiety scores at 60 days postpartum demonstrated higher levels of breastfeeding self-efficacy [11]. A study found that planned training for mothers of infants in neonatal intensive care units significantly reduced maternal anxiety levels [70]. Hence, mothers need to be supported through counseling services postpartum. While several studies have shown that mothers with high breastfeeding self-efficacy are less likely to experience postpartum psychological problems [71,72,73], our study observed that breastfeeding counseling provided through both the AI-based chatbot and the booklet did not affect the anxiety levels of mothers. This finding may be attributed to the fact that both groups received counseling without any human interaction and, therefore, these methods were insufficient to address mothers’ emotional needs. Emotional regulation, especially in the vulnerable postpartum period, requires personalized and relational support [74], which may be difficult to achieve through text-based formats like chatbots. Another possible explanation is the limited follow-up period of 10 days postpartum, which may have been too short to observe measurable changes in anxiety. Previous research suggests that psychological outcomes such as anxiety often require sustained exposure to supportive interventions to demonstrate measurable change [75, 76]. Therefore, future studies should consider extending the intervention and follow-up period to more accurately assess long-term psychological effects and better address the complex emotional needs of new mothers. These findings suggest that while informational support can enhance breastfeeding self-efficacy and technique, reducing postpartum anxiety may necessitate more holistic approaches that incorporate emotional and social support mechanisms.

This study is the first to examine the impact of an AI-powered chatbot on breastfeeding self-efficacy, breastfeeding success, and anxiety of mothers in the postpartum period. While this represents a significant contribution, several limitations must be acknowledged. First, although the chatbot developed in the study provided informational support through counseling, it did not address social or emotional regulation, which may limit its overall effectiveness. Second, mothers were able to access the chatbot at any time and receive immediate responses, contributing to the continuity of care; however, the lack of a true control group reduces the overall power of the study, as only two groups were compared. Third, given that breastfeeding success tends to increase with continued breastfeeding [77], the inability to conduct long-term follow-up limits the understanding of the sustained impact of the intervention. Finally, the duration of the intervention was limited to 10 days, which may be too short to observe significant changes in psychological outcomes such as anxiety.

Conclusion

The findings of this study demonstrate the potential of AI-based applications to improve breastfeeding counseling. Health professionals can integrate AI-supported tools, such as pre-trained chatbots, into routine breastfeeding education programs to enhance mothers’ breastfeeding self-efficacy and success. These tools provide personalized, easily accessible, and continuous support, which can help address common challenges faced by mothers in the postpartum period. Additionally, the use of AI-based applications can reduce the workload of healthcare staff by complementing traditional methods, such as printed educational materials or in-person counseling. By providing timely, consistent, and evidence-based responses to frequently asked questions, AI-based counseling tools can reduce the demand for routine consultations, thereby alleviating some of the workload on nurses, midwives, and other maternal health professionals. Such technologies should be viewed not as replacements for human care, but as valuable supplements that can enhance service delivery, improve efficiency, and expand access—especially during early postpartum days when mothers may have multiple, time-sensitive informational needs. This type of AI-supported intervention may be particularly suitable for use in primary healthcare centers, postpartum home-visiting programs, and digital health platforms. It may offer the greatest benefit to first-time mothers, those with limited access to professional support, or women living in rural or underserved areas. Future implementations should consider combining AI-based counseling with other supportive interventions to optimize outcomes and ensure comprehensive maternal and childcare. Moreover, longer intervention and follow-up periods are recommended to better assess the stability and long-term effects of AI-based counseling tools.

Practice implications

This study examined the effects of an AI-based chatbot and a booklet on breastfeeding success, self-efficacy, and anxiety levels in breastfeeding counseling for mothers. The findings of this study suggest that AI-based breastfeeding counseling can serve as a practical tool to support postpartum mothers, particularly in contexts where face-to-face counseling services are limited or unavailable. The immediate accessibility of informational support offered by the chatbot may help bridge gaps in care during the critical early days of breastfeeding. Clinically, the moderate to large effect sizes observed in breastfeeding self-efficacy and LATCH scores indicate that structured informational interventions—whether delivered through AI or traditional formats—can positively influence maternal behaviors. However, the lack of significant improvement in anxiety levels highlights the need for more holistic approaches that integrate emotional and interpersonal support. These findings emphasize the potential for combining technological tools with human-centered care models to enhance both practical and psychological outcomes in maternal health and highlight the importance of nurses reaching mothers using innovative technologies and continuing to provide counseling services. Further randomized controlled trials are needed to promote breastfeeding and evaluate the effectiveness of individualized innovative breastfeeding interventions.