Caregivers in Argentina and Paraguay were recruited online through targeted online advertisements and the social media channels of two mental health professionals (one from each country). Recruitment and group participation ran from January to November 2025. The advertisements stated that the program was designed specifically for caregivers of children between the ages of 3 and 12 years who exhibited disruptive behaviors, such as frequent tantrums, defiance, aggression, rule breaking, and difficulty following instructions. Interested caregivers were directed to a website to learn more about the program and complete registration. Registration and payment were conducted entirely online. Upon registration, caregivers received access to the ParenteAI platform. In addition to the digital platform, participants were enrolled in a live therapist-led parenting group session that met weekly via Zoom. The first three cohorts participated in 12 weekly sessions, but the seven subsequent cohorts received an 8-week group format. The group length was reduced to make it more manageable for the caregivers and because Pat was able to provide information about the remaining modules. Surveys were administered twice, once at week 4 (midintervention) and again at week 8 (the final session). Four children who were slightly over the age of 12 years were admitted with the caregivers’ understanding that the program was aimed at younger children and that adaptations were going to be discussed with them.

The intervention consisted of a hybrid model that combined live group meetings with a digital platform (ParenteAI) with an AI-based CA (Pat).

This study was approved by the Institutional Review Board of the Universidad Santo Tomás in Chile (approval #23136643/2023). The analysis was conducted using secondary deidentified data collected as part of the original ParenteAI program implementation. At the time of registration, caregivers agreed to the ParenteAI Terms of Service, which included a clause stating that anonymized data could be used for research purposes. Additional informed consent was obtained at the time of survey completion. Participants were not compensated and could withdraw at any time.

This study was a secondary analysis of real-world data. Descriptive statistics were used to report NPSs and the relative contribution of the intervention components. A thematic analysis of participants’ responses was conducted following the steps proposed by Braun and Clarke [37]: (1) familiarization with the data, (2) generation of initial codes, (3) searching for themes, (4) reviewing themes, (5) defining and naming themes, and (6) producing a report. Responses to the NPS question (“Why did you provide this rating?”) and the question regarding positive aspects were combined. With respect to the open-ended questions on the relative contribution of the intervention components, participants’ responses were grouped into three categories based on (1) attributing change to interactions with Pat, (2) attributing change to participation in the group, and (3) complementarity of the interventions.

AI was used as a supportive tool to generate an initial set of proposed codes based on participants’ responses. This proposal was thoroughly reviewed by three of the paper’s authors, who evaluated its conceptual relevance, refined operational definitions, removed redundancies, deepened analytical nuances, verified the relevance of potential themes, and generated the final list of themes.

In addition, following the guidelines of the coding reliability approach described by Boyatzis [38] and later operationalized and further developed in applied research contexts by Guest et al [39], participants’ responses were independently analyzed by two evaluators. Code frequencies and interrater reliability (IRR) were calculated using the Cohen κ coefficient. Discrepancies were discussed and resolved by consensus among the three authors. All themes were initially developed in Spanish and subsequently translated into English by the authors.

In total, 88 primary caregivers participated in the study. They were mostly female (n=76, 86%), aged 34-65 years (mean 44.31, SD 5.73), and had a child between 3 and 14 years old (mean age 7.98, SD 2.45 years). The children were 80% (n=70) male and 20% (n=18) female. Only 3% (n=3) of the caregivers identified as Indigenous or belonging to Native people, while 84% (n=74) identified as White or of European descent. More than half of the caregivers reported being married (n=49, 56%). Furthermore, 38% (n=33) had a university degree, while 42% (n=37) had a postgraduate degree. See Table 1 for participants’ demographic characteristics.

The mean NPS was 9.32 (SD 1.31; range 2-10), and the majority of caregivers were promoters (Figure 3). The NPS was calculated as the percentage of promoters (scores 9-10) minus the percentage of detractors (scores 0-6), following standard NPS methodology. Of 117 responses, 94 (80.3%) were classified as promoters (scores 9-10), 19 (16.2%) as passives (scores 7-8), and 4 (3.4%) as detractors (scores 0-6). The resulting mean NPS was 76.92 (range 80.34-3.42), which is considered excellent. When asked why they provided such NPSs, most responses (n=211, 56.3%) were about the intervention in general (which included group sessions and the platform), and a considerable portion (n=164, 43.7%) were specific to Pat. The IRR for the overall program or Pat was 0.87.

Overall, caregivers’ responses revealed a positive sentiment about combining therapists with Pat. As shown in Table 2, the sentiment of the majority of responses was positive (172/181, 95.0%), a small portion of responses were neutral (8/181, 4.4%), and only 1 response was coded as negative (1/181, 0.6%). A total of 10 themes emerged, and the IRR was 0.90.After assessing the IRR, discrepancies between raters were resolved through consensus, resulting in a final set of themes.

There were 202 coded segments in total. Regarding participants’ perceptions of the overall program, four major themes emerged (Table 3): useful strategies (73/202, 36.1%), professional and peer support (65/202, 32.2%), positive changes (52/202, 25.7%), and overall satisfaction and recommendation (12/202, 5.9%). The most frequent theme was “useful strategies,” highlighting the value placed on PMT strategies and participants’ appreciation for clear, structured, and applicable tools to address daily parenting challenges. For example, one caregiver noted:

The “professional and peer support” theme highlighted the value placed on ongoing guidance from professionals and shared experiences with other caregivers. The “positive changes” theme captured caregivers’ reports of improvements in family communication, reflection, and their children’s behavior, as reflected in statements such as:

Finally, the “overall satisfaction and recommendation” theme reflected caregivers’ general endorsement of the program.

Participants were asked about Pat, and most comments were positive in nature (Table 4). The most frequent positive theme was “24/7 accessibility” (69/164, 42.1%), reflecting users’ appreciation for the constant availability of the platform. The second-most common positive theme was “useful strategies” (54/164, 32.9%), emphasizing that Pat provides concrete, evidence-based tools for managing parenting challenges. The third positive theme, “emotional support and containment” (33/164, 20.1%), captured caregivers’ sense of empathy and understanding from Pat.

Neutral comments accounted for a smaller portion of the responses. The most frequent neutral theme was “preference for human accompaniment” (6/164, 3.7%), with some caregivers expressing that live professional meetings felt more personal and specific. Another neutral theme, “limitations at critical moments” (1/164, 0.6%), reflected the perspective of a user who found it difficult to access help from the platform, when needed.

Finally, the only negative theme, “difficulty trusting the app” (1/164, 0.6%), captured skepticism or hesitation toward relying on Pat compared to human professionals. For example, one participant said:

Regarding participants’ recommendations for improvement, four main themes emerged: user experience and technological functionalities (56/122, 45.9%), overall satisfaction without recommendations (29/122, 23.8%), content adaptation (26/122, 21.3%), and resources and support materials (11/122, 9.0 %). The IRR was 0.84.

The “user experience and technological functionalities” theme included suggestions to enhance accessibility, usability, and interactivity. For instance, caregivers said:

Such features were incorporated for the final cohorts, but initially, the web app supported only text.

The “overall satisfaction without recommendations” theme reflected participants’ approval of the program without suggesting changes.

The “content adaptation” theme included recommendations to tailor Pat’s content to caregivers’ specific contexts, children’s developmental stages, and family dynamics. Suggestions included each parent having a unique username when both parents of a child participated so that Pat could understand the unique dynamics of each parent and provide tailored guidance.

The “resources and support materials” theme captured requests for complementary resources and printable materials to support learning and follow-up, such as adding visual resources.

Caregivers reported attributing, on average, 61% of the progress to Pat and 46% to the group sessions. When asked to estimate what percentage of their child’s progress they attributed to Pat compared to the live group sessions (see Multimedia Appendix 1 for a graphical representation), percentages exceeded 100% because several caregivers assigned overlapping credit to both components. Their qualitative explanations for these attributions revealed three main categories: (1) favor Pat, (2) favor group sessions, and (3) favor a combination of both Pat and group sessions (Table 5). The largest proportion of responses favored Pat (45/106, 42.5%), primarily due to its practical tools and specific strategies, its emotional support and personalization, and its constant accessibility. The caregivers were satisfied with the group experience, but the accessibility and immediacy of Pat were valued even more:

A second group of caregivers (44/106, 41.5%) emphasized the importance of the group sessions, highlighting the clear and reliable professional support and shared learning with other caregivers they provided. Caregivers expressed they most valued the presence of the therapists, as well as the value in interacting with and listening to other caregivers, which they found reinforced theories learned. Finally, a smaller subset (17/106, 16%) of caregivers emphasized the complementarity between both delivery approaches, describing how the combination of Pat and live group sessions yielded the greatest benefit, as they were able to learn from Pat and then discuss and expand what was learned in the group.

This study aimed to understand caregivers’ experiences with a hybrid PMT program that combined live, therapist-led group sessions with asynchronous support from Pat, an AI-based CA. Overall, caregivers reported high satisfaction, reflected in an NPS of 76.92, which is considered excellent. Qualitative analyses revealed overwhelmingly positive feedback about the program overall and specifically about Pat. Regarding the overall program, caregivers highlighted its practical tools, the support and sense of community fostered by the group format, and positive changes observed. Comments specific to Pat emphasized its constant availability; its sense of empathy, containment, and companionship; and the useful strategies it taught.

Caregivers reported more benefits from Pat than from the live sessions, with approximately 61% of the improvement being assigned to Pat and 46% to the live group sessions. These percentages suggest that both components played significant and overlapping roles in promoting change. However, caregivers leaned more toward Pat, and most responses highlighted the importance of the in-the-moment support provided by Pat. Although these findings should be considered with caution as this was a survey, taken together, the findings highlight the positive acceptance of caregivers of the hybrid model and the complementary benefits of incorporating AI to enhance the support provided to caregivers of children with disruptive behaviors.

The findings of this preliminary study have several implications for clinicians, program developers, and health systems. For clinicians, hybrid models that combine live PMT through videoconferencing with AI support can enhance scalability and continuity of care, allowing families to receive personalized guidance from their homes and receive support between sessions without increasing therapist workload. AI tools, such as Pat, can function as engagement boosters (ie, reminding, reinforcing, and tracking caregivers’ use of skills). Although this study did not analyze engagement, other studies on group interventions with adults with depression have shown that AI could enhance the engagement with the treatment and improve outcomes [22]. For developers, including accessible resources (eg, apps) and voice features to avoid typing will be essential to ensure smoother engagement with the CAs. Finally, for health systems, integrating AI-supported interventions could lead to greater accessibility and cost efficiency, reducing logistical barriers, such as time, transportation, and provider shortages, that often limit families’ participation in evidence-based parenting programs.

Several limitations should be considered when interpreting the findings reported here. First, the study relied on real-world, cross-sectional, and self-report data, which may be influenced by recall or social desirability biases. Second, there was no control group or pre-/postmeasurement of child behavior change, preventing conclusions about child behavior outcomes or causal effects. Third, the sample likely reflects a degree of selection bias, as participating caregivers were digitally literate, motivated, and had consistent internet access, and we do not know the impressions of those who did not complete the surveys. Additionally, the percentages attributed to Pat and the live sessions represent subjective perceptions rather than objective estimates of causal contribution. Finally, because the study was conducted in Argentina and Paraguay, findings may be context bound and not fully generalizable to other cultural or health care settings. To note, most evidence-based parenting interventions (including PMT) have been developed and tested primarily in high-income, English-speaking contexts, often with non-Hispanic White samples [42]. This study provides a unique contribution to the field by including a Latin American population that often lacks mental health resources and is understudied.

Future research should build on these findings by using RCT designs to examine the causal effects of hybrid, AI-supported PMT compared to traditional formats. A key next step is to quantify the therapeutic alliance that may develop between caregivers and AI agents, given caregivers’ reports of empathy and support from Pat. Finally, future evaluations should incorporate child-level outcomes (eg, behavioral improvements) and family-level outcomes (eg, reductions in parenting stress and improvements in family functioning) to determine the broader impact of this approach on both caregivers and children.

In this preliminary investigation, caregivers reported positive experiences with a hybrid program combining live videoconferencing group PMT sessions plus an AI. Delivering PMT in group format and through videoconferencing increases the scalability of the workforce and increases accessibility to the resources. Furthermore, the integration of AI (Pat) expanded the benefits of traditional PMT by continuing the delivery of skills and providing personalized, on-demand support between sessions. Caregivers valued Pat’s immediacy and practicality, noting that “Pat helps us with specific questions about how to talk to our child or how to handle situations,” as well as its constant availability, sharing that “having Pat 24/7 makes you feel supported at all times.” Together, these components created a synergistic model in which human expertise and empathy were complemented by AI-driven accessibility and continuity, reducing barriers, such as time, cost, and clinician availability. This hybrid approach appears to make PMT more flexible, scalable, and sustainable without sacrificing the relational quality that remains central to therapeutic success.