The Effectiveness of Artificial Intelligence-Based Pet Therapy in Improving the Care of Patients: A Systematic Review

Abstract

Animal-assisted interventions can support emotional well-being and social engagement, but their use may be limited by allergies, infection risks, and animal-handling requirements. As a scalable alternative, artificial intelligence-based pet therapy, including socially assistive and robotic pets, has been introduced in health care and community settings. This systematic review, conducted in accordance with PRISMA guidelines, searched PubMed, CINAHL Ultimate/MEDLINE, Scopus, Web of Science, and SAGE for studies published between 2014 and 2025 that evaluated the reported effects of AI-based pet therapy on patient outcomes. Of the 584 records identified, 27 studies met the inclusion criteria after duplicate removal and screening. Most studies involved older adults with dementia, although children, veterans, and community-dwelling adults were also represented. Across studies, AI-based pet therapy was associated with reduced agitation, anxiety, stress, and pain, as well as improved mood, communication, and social engagement. PARO was the most frequently studied robotic pet, and interventions were typically delivered 1–3 times weekly for 30–60 min over 4–12 weeks. Overall, AI-based pet therapy appears to be a promising complementary non-pharmacological approach, particularly for people with dementia and hospitalized children, although stronger evidence from larger, more standardized, and longer-term studies is still needed.

1. Introduction

Animal-assisted intervention is a broad term that includes various approaches, such as animal-assisted therapy, animal-assisted activities, and animal-assisted education. Animal-assisted therapy (AAT), provided by a formally trained professional, involves a goal-oriented therapeutic intervention [1]. AAT has become widespread, with programs targeting different pathologies and populations [2]. This targeted intervention, based on positive human–animal interactions, helps individuals cope with or recover from health issues by fostering relaxation, reducing emotional distress, and providing a sense of purpose [3]. Animal-assisted interventions are particularly valuable in mental health treatment, especially when combined with therapies such as cognitive behavioral therapy, as they can disrupt harmful thoughts, including suicidal ideation, and prevent self-harm [4]. Reported benefits include reduced negative affect, improved mood, increased empathy, and better therapeutic alliance, especially for individuals who struggle with traditional therapy engagement [5,6,7]. Alternative therapies such as equine-assisted services have also gained increasing recognition for their reported effects in addressing conditions such as post-traumatic stress disorder [8]. AAT is also believed to enhance empathy and reduce behavioral issues [9].

Further evidence supports the potential value of traditional animal-assisted approaches. One study examined the effects of AAT in elderly patients with schizophrenia over a 12-month period [10]. Weekly sessions involving dogs and cats aimed to improve mobility, communication, and self-care, and the study reported significant improvements in social functioning, daily activities, and overall well-being compared with baseline conditions [10]. In recent years, AAT has continued to grow because of its physical, psychological, and social benefits, helping individuals improve quality of life through interactions with various therapy animals [11]. Dogs, cats, and horses are frequently highlighted as particularly beneficial, with dogs standing out because of their long-standing bond with humans, emotional responsiveness, and ability to understand social cues, making them a common choice for enhancing mental health and well-being in therapy [12].

Despite these benefits, traditional animal-assisted interventions also present challenges. Concerns have arisen due to unpredictable factors such as allergies, fear of animals, and the risk of spreading hospital-related infections [13]. In response to these limitations, robotic pet interventions have been proposed as safer alternatives [14]. Artificial intelligence (AI) technologies, including chatbots, virtual assistants, and socially assistive robots, are increasingly used to support individuals with compromised social or emotional well-being [15]. Biomimetic neural networks offer an effective AI mechanism and have demonstrated strong results in various challenges [16], while biomimetic robots enhance understanding of animal behavior, cognition, and movement [17].

Therefore, it is important to examine whether AI-based pet therapy can provide benefits similar to traditional animal-assisted interventions while reducing some of the practical and safety-related limitations associated with live animals. As robotic pets and socially assistive robots are increasingly introduced into health and care settings, clearer evidence is needed on their effects on patient outcomes, their suitability for different populations, and their potential role as complementary therapeutic tools. This review addresses this need by synthesizing current evidence on the effectiveness of AI-based pet therapy in improving patient care.

The aim of this systematic review is to evaluate the effectiveness of artificial intelligence–based pet therapy, including robotic pets and socially assistive robots, in improving patient outcomes across healthcare and care-related settings. Specifically, the review aimed to synthesize evidence on the effects of AI-based pet therapy on emotional well-being, agitation, anxiety, stress, pain, communication, social engagement, and activity participation, while also considering intervention characteristics, target populations, and acceptability, and methodological quality of the included studies. The research question guiding the review was: How effective is AI-based pet therapy in improving patient outcomes?

2. Materials and Methods

This study was conducted as a systematic review to synthesize reported outcome patterns of AI-based pet therapy in improving patient outcomes across clinical and community settings. The review was reported in accordance with PRISMA, and the completed PRISMA 2020 checklist is provided in Supplementary Materials [18]. PRISMA-guided transparent reporting, but it did not substitute for the underlying review methodology. The review protocol was registered through the Open Science Framework (OSF) [19]. The review was not designed as a living review; accordingly, the search was conducted once within the predefined time frame and was not repeated thereafter.

2.1. Search Strategy

A systematic literature search was performed in PubMed, CINAHL Ultimate/MEDLINE, Scopus, Web of Science, and SAGE. The interfaces used were PubMed via PubMed, CINAHL Ultimate via EBSCOhost, MEDLINE via EBSCOhost, Scopus via Scopus, Web of Science via Web of Science, and SAGE via Sage Journals. No grey literature sources were searched. The search strategy was adapted to each database and combined keywords and synonyms related to AI-based pet therapy and patient outcomes using Boolean operators (AND, OR). The search string used was: (“artificial intelligence pet*” OR “artificial intelligence animal*” OR “artificial pet-assisted therapy” OR “artificial animal-assisted therapy” OR “artificial pet-assisted intervention” OR “artificial animal-assisted intervention” OR “robotic animal-assisted therapy” OR “robotic pet-assisted therapy” OR “robot pet*” OR “robot-animal*”) AND (“health status” OR “medical status” OR “disease symptom*” OR “health care” OR “patient care” OR “health outcome*”). The search was limited to articles published in English between 2014 and 2025.

Screening was managed in Rayyan and followed a staged PRISMA-based procedure. First, all records retrieved from the databases were combined and checked for duplicates, which were removed before relevance screening. Screening then proceeded in two substantive stages: title-and-abstract screening followed by full-text assessment of potentially eligible studies. During title-and-abstract screening, decisions were based primarily on the title and abstract, with keywords and bibliographic information consulted where available; full-text screening involved assessment of the entire article. Screening was conducted independently by two reviewers in each round, and disagreements were resolved through discussion, with referral to an additional reviewer when necessary. Uncertain records were retained for full-text review to reduce the risk of premature exclusion.

2.2. Eligibility Criteria

Studies were eligible for inclusion if they examined AI-based pet therapy in patient or care-related contexts, addressed patient outcomes or patient-relevant effects, and were published in English between 2014 and 2025. Both primary studies and review articles were considered where they were relevant to the review question. Records were excluded if they were duplicates, published outside the predefined time frame, not relevant to AI-based pet therapy or related robotic animal-assisted interventions, not focused on patient care or patient-targeted populations, did not address patient-related outcomes or care-related effects, or provided inadequate content at full-text review to answer the review question.

2.3. Data Extraction

Data extraction was conducted by one reviewer using a standardized extraction form developed in advance on the basis of the review question and the structure of the study-characteristics table. The following data were extracted from each included study: bibliographic information; study design and aim; population characteristics; intervention characteristics, including robot type, delivery format, and session details where available; and outcome-related findings, including patient-related effects and relevant implementation, usability, acceptability, or ethical observations. Extraction was performed conservatively, with unreported information recorded as “not reported,” and extracted data were checked for internal consistency before synthesis. No masking procedure or duplicate extraction was used (

Table 1. Inclusion and exclusion criteria.

Inclusion Criteria
Intervention or process or topic	AI-based pet therapy
Population/problem targeted	Patient
Type of study	No limit
Outcomes	Patient outcome
Language	No limit
Time frame	10 years
Exclusion criteria
Inadequate content, inadequate outcome, over 10 years old, duplicates

).

Because this study was based exclusively on published literature, no ethical approval was required. No generative artificial intelligence was used in the design, screening, extraction, analysis, or interpretation of the review.

2.4. Critical Appraisal

Quality appraisal and data management included qualitative elements using the JBI Critical Appraisal Checklists (2019). This method provided a systematic way to evaluate methodological quality across the included study designs and helped ensure consistent extraction and interpretation of evidence. Each study was independently assessed by two reviewers, who then compared their ratings. Any disagreements were resolved through discussion or, when required, by a third reviewer. The relevant JBI checklist was selected according to study design, including randomized controlled trials [20], quasi experimental studies [21], systematic reviews [22], qualitative studies [23].

3. Results

Across all the databases searched, a total of 584 records were identified. The highest number of hits was retrieved from PubMed (n = 567), followed by Scopus and SAGE (n = 6 each), MEDLINE (n = 3), and CINAHL Ultimate (n = 2) (

Table 2. Number of hits.

Keywords	PubMed	CINAHL Ultimate	MEDLINE	Scopus	SAGE
(“artificial intelligence pet*” OR “artificial intelligence animal*” OR “artificial pet-assisted therapy” OR “artificial animal-assisted therapy” OR “artificial pet-assisted intervention” OR “artificial animal-assisted intervention” OR “robotic animal-assisted therapy” OR “robotic pet-assisted therapy” OR “robot pet*” OR “robot-animal*”) AND (“health status” OR “medical status” OR “disease symptom*” OR “health care” OR “patient care” OR “health outcome*”)	567	2	3	6	6

).

We removed 57 duplicates. We excluded 482 articles based on their title and abstract due to irrelevance. We then reviewed the remaining articles in full and excluded a further 18 due to irrelevance (Figure 1).

3.1. Study Characteristics and Intervention Modalities

The included studies used diverse designs, including systematic and scoping reviews, randomized controlled trials, quasi-experimental studies, pilot studies, qualitative studies, and ethical or implementation-focused analyses. Most studies focused on older adults and people with dementia in nursing homes, residential aged care, or long-term care facilities, although some examined children in hospital or school settings, healthy adults, healthcare professionals, or pet owners. The robotic interventions also varied widely. The most commonly studied robots were animal-like or pet robots, especially PARO, Joy for All robotic cats and dogs, LOVOT, AIBO, MiRo-E, and other companion robots, while fewer studies examined humanoid or telepresence robots such as NAO, Huggable, and Enabot. Intervention modalities included individual and group sessions, structured therapeutic activities, free play, tactile interaction, companionship, cognitive and sensory stimulation, communication facilitation, distraction during medical procedures, and remote interaction with pets. Session length ranged from brief 5 min interactions to 30–60 min sessions, delivered from single encounters to repeated programmes lasting several weeks or months. Overall, the interventions were heterogeneous, but most shared a common aim: to use social, responsive, and often animal-like robotic interaction to promote emotional well-being, communication, engagement, and social participation.

3.2. Critical Evaluation of Articles

This section presents the methodological appraisal of the included articles using the appropriate JBI critical appraisal tools according to study design. The evaluation focuses on key sources of bias, including study design, participant selection, comparability of groups, outcome measurement, follow-up, statistical analysis, and transparency of reporting. The aim is to assess the reliability of the included evidence and identify the main methodological strengths and limitations that should be considered when interpreting the findings.

Table 3. Critical assessment of RCT.

Checklist Question	Geva et al. [24]	Logan et al. [27]	Petersen et al. [25]	Valentí Soler et al. [26]
Q1 1	yes	yes	yes	yes
Q2 2	unclear	unclear	unclear	unclear
Q3 3	yes	no	yes	no
Q4 4	no	no	no	no
Q5 5	no	no	no	no
Q6 6	yes	no	no	yes
Q7 7	no	no	no	yes
Q8 8	yes	yes	yes	yes
Q9 9	yes	yes	yes	yes
Q10 10	yes	no	unclear	yes
Q11 11	unclear	unclear	unclear	unclear
Q12 12	yes	yes	yes	yes
Q13 13	yes	yes	yes	unclear

¹ Was true randomization used for assignment of participants to treatment groups? ² Was allocation to treatment groups concealed? ³ Were treatment groups similar at the baseline? ⁴ Were participants blind to treatment assignment? ⁵ Were those delivering the treatment blind to treatment assignment? ⁶ Were treatment groups treated identically other than the intervention of interest? ⁷ Were outcome assessors blind to treatment assignment? ⁸ Were outcomes measured in the same way for treatment groups? ⁹ Were outcomes measured in a reliable way ¹⁰ Was follow up complete and if not, were differences between groups in terms of their follow up adequately described and analysed? ¹¹ Were participants analysed in the groups to which they were randomized? ¹² Was appropriate statistical analysis used? ¹³ Was the trial design appropriate and any deviations from the standard RCT design (individual randomization, parallel groups) accounted for in the conduct and analysis of the trial?

presents the critical appraisal of the included randomized controlled trials [24,25,26,27] using the revised JBI Critical Appraisal Tool for the Assessment of Risk of Bias for Randomized Controlled Trials (RCTs) [20]. This tool uses signaling questions to evaluate risk of bias across key domains of validity, including internal validity, such as selection and allocation procedures, performance bias, detection and measurement bias, and attrition bias, as well as statistical conclusion validity, including appropriateness of statistical analyses and whether participants were analysed in the groups to which they were randomized. Overall, the trials demonstrated strengths in the reliability and consistency of outcome measurement, with all studies meeting the criteria for Q8 and Q9, and most studies using appropriate statistical analysis (Q12) and accounting for trial design considerations (Q13).

However, several methodological limitations were identified across the included studies. Allocation concealment was consistently unclear across all four trials (Q2), indicating that the papers did not clearly describe how assignment was concealed before group allocation. Blinding was also a major limitation. Participant blinding was generally not achieved (Q4), which was expected given that participants could usually identify whether they were receiving an active robotic intervention, inactive robot, plush toy, tablet/avatar, dog-assisted therapy, or usual activities. Similarly, those delivering the intervention were not blinded in any study (Q5), as experimenters, nurses, staff, child life specialists, or therapists were aware of which intervention they were providing. Outcome assessor blinding was also limited or unclear (Q7), particularly where outcomes were self-reported or collected by staff who likely knew the treatment allocation.

Selection and baseline comparability were variably reported. Randomization was reported in all studies (Q1), but baseline similarity was mixed (Q3), with Logan et al. [27] and Valentí Soler et al. [26] showing limitations or uncertainty in baseline comparability. Follow-up and handling of missing data also varied (Q10). Geva et al. [24] and Valentí Soler et al. [26] appeared to have adequate follow-up, whereas Logan et al. [27] did not fully analyse exclusions or incomplete data, and Petersen et al. [25] provided unclear detail on attrition or missing data. Intention-to-treat analysis was unclear across all studies (Q11), with no clear statement that participants were analysed in the groups to which they were randomized.

Table 4. Critical assessment of systematic reviews (SR).

Checklist Question	Du et al. [28]	Abbott et al. [14]	Moerman et al. [30]	Scerri et al. [29]
Q1 1	yes	yes	yes	yes
Q2 2	yes	yes	yes	yes
Q3 3	yes	yes	yes	yes
Q4 4	no	yes	yes	yes
Q5 5	yes	yes	no	yes
Q6 6	yes	yes	yes	yes
Q7 7	yes	yes	unclear	yes
Q8 8	no	yes	yes	yes
Q9 9	yes	unclear	no	no
Q10 10	yes	yes	yes	yes
Q11 11	yes	yes	yes	yes

¹ Is the review question clearly and explicitly stated? ² Were the inclusion criteria appropriate for the review question? ³ Was the search strategy appropriate? ⁴ Were the sources and resources used to search for studies adequate? ⁵ Were the criteria for appraising studies appropriate? ⁶ Was critical appraisal conducted by two or more reviewers independently? ⁷ Were there methods to minimize errors in data extraction? ⁸ Were the methods used to combine studies appropriate? ⁹ Was the likelihood of publication bias assessed? ¹⁰ Were recommendations for policy and/or practice supported by the reported data? ¹¹ Were the specific directives for new research appropriate?

presents the critical appraisal of the included systematic reviews and research syntheses by Du et al. [28], Abbott et al. [14], Scerri et al. [29], and Moerman et al. [30] using the JBI Critical Appraisal Checklist for Systematic Reviews and Research Syntheses [22]. Overall, the reviews had clearly stated review questions, appropriate inclusion criteria, and suitable search strategies, as shown by positive ratings for Q1, Q2, and Q3. Abbott et al. [28], Scerri et al. [29], and Moerman et al. [30] used adequate sources and resources for searching studies, while Du et al. [28] was rated negatively for Q4, mainly because restrictions such as English-language and full-text inclusion may have limited the comprehensiveness of the search.

Several methodological strengths were identified. Du et al. [28] had a clear objective, included only randomized controlled trials, used independent screening, extraction, and appraisal, applied appropriate statistical methods, and assessed inconsistency and publication bias. Abbott et al. [14] had a registered PROSPERO protocol, a comprehensive search strategy, a mixed-methods approach, stakeholder involvement, transparent synthesis, and appropriate quality appraisal tools. Scerri et al. [29] was a well-conducted qualitative evidence synthesis using meta-ethnography. It had a clear aim, appropriate inclusion criteria, a systematic database search, independent appraisal by two reviewers, and suitable synthesis methods based on Noblit and Hare’s meta-ethnography approach. Moerman et al. [30] also provided a clear review focus and used an appropriate narrative synthesis because the included studies were few and heterogeneous.

However, some limitations were present across the reviews. Du et al. [28] did not register a protocol, included studies of generally low quality, reported poor blinding, detected publication bias, and had high heterogeneity, which limits confidence in the findings. Abbott et al. [14] included quantitative studies that were generally small, short-term, and often at unclear or high risk of bias; blinding was difficult, outcome measures may not have fully captured benefits, long-term evidence was limited, and publication bias was not clearly assessed. Scerri et al. [29] was limited by an English-language search and the inclusion of only eight articles, which may restrict the breadth of evidence. Publication bias assessment was not applicable because it was a qualitative synthesis. Moerman et al. [30] showed the most notable methodological weaknesses: the authors did not clearly use a standard critical appraisal tool, it was unclear whether data extraction was completed independently by two reviewers, publication bias was not assessed, and some included evidence came from conference papers or abstracts with limited methodological detail.

Table 5. Critical assessment of quasi-experimental studies.

Checklist Question	Lane et al. [31]	Harris-Gersten et al. [36]	Barber et al. [34]	Dinesen et al. [35]	Sung et al. [32]	Klumpe et al. [33]
Q1 1	yes	yes	yes	yes	yes	yes
Q2 2	no	no	yes	no	no	yes
Q3 3	no	yes	yes	unclear	no	yes
Q4 4	unclear	unclear	yes	unclear	unclear	yes
Q5 5	yes	no	unclear	yes	yes	yes
Q6 6	yes	yes	yes	yes	yes	yes
Q7 7	unclear	unclear	unclear	unclear	yes	yes
Q8 8	unclear	unclear	unclear	yes	unclear	unclear
Q9 9	yes	no	yes	yes	yes	yes

¹ Is it clear in the study what is the “cause” and what is the “effect” (i.e., there is no confusion about which variable comes first)? ² Was there a control group? ³ Were participants included in any comparisons similar? ⁴ Were the participants included in any comparisons receiving similar treatment/care, other than the exposure or intervention of interest? ⁵ Were there multiple measurements of the outcome, both pre and post the intervention/exposure? ⁶ Were the outcomes of participants included in any comparisons measured in the same way? ⁷ Were outcomes measured in a reliable way? ⁸ Was follow-up complete and if not, were differences between groups in terms of their follow-up adequately described and analyzed? ⁹ Was appropriate statistical analysis used?

presents the critical appraisal of the included quasi-experimental studies [31,32,33,34,35,36] using the JBI checklist for quasi-experimental studies [21]. Overall, most studies clearly identified the intervention or exposure and the measured outcomes, indicating adequate temporal clarity for Q1. However, the main methodological weakness across this group was study design. Several studies lacked a control or comparison group, particularly Lane et al. [31], Dinesen et al. [35], Harris-Gersten et al. [36], and Sung et al. [32], which limits the ability to attribute observed changes directly to robotic intervention. In addition, participant comparability was often weak or unclear because several studies used small, convenient, or highly specific samples. Lane et al. [31] had important methodological limitations, including no randomization, no control group, convenience sampling, possible observer bias, locally developed outcome measures, and incomplete follow-up data. Harris-Gersten et al. [36] also had a weak design due to the absence of a control group, unclear comparability of care apart from the intervention, limited statistical analysis, and unclear reliability of outcome measurement. Dinesen et al. [35] was exploratory and lacked a control group, while measurement reliability was partly limited because some outcomes were rated by healthcare professionals rather than directly by residents. Sung et al. [32] used a one-group pretest–posttest design with a small convenience sample and incomplete participation, as only 12 of 16 participants completed the study. Barber et al. [34] and Klumpe et al. [33] were methodologically stronger because they used within-subject or repeated-measures designs, which improved comparability between conditions. However, Barber et al. [34] had uncleared pre- and post-measurements for some outcomes, mixed reliability due to one scale showing low internal consistency, and unclear reporting of follow-up or missing data. Klumpe et al. [33] used a crossover design in which the same participants interacted with both real dogs and AIBO robotic dogs, strengthening internal comparability. Nevertheless, the study had unclear completeness of data because of missing or incomplete questionnaire data and limited availability of AIBO internal data. Its small and homogeneous sample also reduced generalizability.

Table 6. Critical assessment of qualitative studies.

Checklist Question	Jung et al. [37]	Li et al. [45]	Pike et al. [41]	Yuan et al. [39]	Ihamäki and Heljakka [40]	Koh et al. [43]	Koh et al. [44]	Pu et al. [38]	Jang [42]
Q1 1	unclear	unclear	yes	unclear	unclear	yes	yes	unclear	unclear
Q2 2	yes	yes	yes	yes	yes	yes	yes	yes	yes
Q3 3	yes	yes	yes	yes	yes	yes	yes	yes	yes
Q4 4	yes	yes	yes	yes	yes	yes	yes	yes	yes
Q5 5	unclear	yes	yes	yes	yes	yes	yes	yes	yes
Q6 6	no	unclear	yes	no	unclear	unclear	yes	unclear	no
Q7 7	no	unclear	yes	unclear	unclear	unclear	yes	unclear	no
Q8 8	no	yes	no	yes	yes	yes	yes	yes	unclear
Q9 9	yes	yes	yes	yes	unclear	yes	yes	yes	yes
Q10 10	yes	yes	yes	yes	yes	yes	yes	yes	yes

¹ Is there congruity between the stated philosophical perspective and the research methodology? ² Is there congruity between the research methodology and the research question or objectives? ³ Is there congruity between the research methodology and the methods used to collect data? ⁴ Is there congruity between the research methodology and the representation and analysis of data? ⁵ Is there congruity between the research methodology and the interpretation of results? ⁶ Is there a statement locating the researcher culturally or theoretically? ⁷ Is the influence of the researcher on the research, and vice versa, addressed? ⁸ Are participants, and their voices, adequately represented? ⁹ Is the research ethical according to current criteria, and is there evidence of ethical approval by an appropriate body? ¹⁰ Do the conclusions drawn in the research report flow from the analysis, or interpretation, of the data?

presents the critical appraisal of the qualitative studies [37,38,39,40,41,42,43,44,45] using the JBI Critical Appraisal Checklist for Qualitative Research [23]. Overall, the studies showed good alignment between the research methodology, research questions, data collection, data analysis, interpretation of results, and conclusions, as most studies were rated positively for Q2–Q5 and Q10. The main weaknesses were related to researcher positioning and reflexivity. Several studies did not clearly state the researchers’ philosophical, cultural, or theoretical position, which led to unclear or negative ratings for Q1 and Q6. Researcher influence on the research process was also often poorly reported, especially in Jung et al. [37], Pu et al. [38], Yuan et al. [39], Ihamäki and Heljakka [40], Jang [42], Koh et al. [43], and Li et al. [45].

Participant representation was generally adequate. Pu et al. [38] clearly represented participants’ voices through direct quotations, while most other studies also included participant or stakeholder perspectives. However, Jung et al. [37] lacked residents’ voices, Pike et al. [41] had a small sample with incomplete participation, and Jang [42] only partly represented participant voices because findings were mainly presented through factor types and statement rankings rather than rich direct quotations. Ethical reporting was mostly adequate, although it was unclear in Ihamäki and Heljakka [40]. Pu et al. [38] and Jang [42] both reported ethical approval.

Abdi et al. [46], Koh et al. [47], Papadopoulos et al. [48], and Preuß & Legal [49] were not included in the formal JBI critical appraisal because their study designs did not match for critical appraisal tools Abdi et al. [46], Koh et al. [47], and Papadopoulos et al. [48] were scoping reviews, which are primarily designed to map the breadth and nature of evidence rather than assess intervention effectiveness or synthesize findings in the same way as systematic reviews. Preuß & Legal [49] was an ethical analysis and literature review, not an empirical study or systematic evidence synthesis. Therefore, these articles were used to provide contextual, theoretical, ethical, and background information, but they were not formally assessed using the JBI critical appraisal tools selected for randomized controlled trials, quasi-experimental studies, qualitative studies, and systematic reviews/research syntheses.

The JBI critical appraisal results were used as an indicator of methodological quality and risk of bias across the included studies. Therefore, no separate risk-of-bias tool was applied.

To provide a structured overview of the evidence included,

Table 7. The characteristics of included studies.

Authors (Year)	Type of Study	Aim/Objectives	Population and Sample	Type of AI/Robotic Therapy	Intervention/Description of Therapy	Key Findings
Abdi et al. (2017) [46]	Scoping review	Examines the literature on the use of SAR (socially assistive robot) in elderly care and aims to establish the roles this technology may play in the future.	61 final publications—describing 33 studies and including 1574 participants and 11 robots.	The types of AI/robotic therapy identified include affective therapy.	The interventions involved the use of social assistive robots (SAR) to support elderly care through various roles (e.g., improving mood and reducing agitation, enhancing cognitive function via training exercises and games, providing companionship, and in some cases influencing physiological measures like blood pressure). These interventions were delivered through robot-led activities or interactions.	Socially assistive robots (SAR) show promise in elderly care across five roles: affective therapy, cognitive training, social facilitation, companionship, and physiological therapy. Cognitive training and social facilitation had the most consistent positive effects. Group interactions were generally more effective than one-on-one. However, some benefits could be matched by simpler alternatives like soft toys or computer programs. Study quality varied, and further rigorous research is needed to confirm SAR’s clinical value and cost-effectiveness.
Abbott et al. (2019) [14]	Systematic review of qualitative and quantitative research	Collect evidence on the experiences of staff, residents, and relatives with robotic pets and their impact on the health and well-being of older people in nursing homes.	Nineteen studies—10 qualitative, 2 mixed methods, and 7 randomized trials	Five types of robopet were identified (robotic seal Paro, robotic cat JustoCat, robotic cat NeCoRo, robotic dog Aibo, robotic teddy bear CuDDler.	Most studies assessed participants’ effects/experiences during robotic-pet sessions delivered one-to-one or in groups, led by therapists/researchers or self-led. Sessions typically ran 10–40 min, 2–3 times per week, over 4 weeks to 4 months.	Robotic pets show mixed effects. Loneliness fell but not significantly (SMD −0.51). In dementia, Paro significantly reduced agitation (SMD −0.32). Depression and quality of life showed no significant change (depression SMD 0.09). Paro boosted verbal/physical engagement and beat a stuffed toy; anxiety mostly null, medication findings mixed, apathy inconsistent, and sleep unaffected.
Barber et al. (2020) [34]	Pre-post test	To investigate the potential effects of social interaction levels and biophilic beliefs on participants’ evaluations of two potential therapeutic additions, animals and robots, specifically dogs and the biomimetic robot MiRo-E.	Thirty-four individuals participated in the study: 18 boys and 16 girls aged 11 to 12 years. Participants were a voluntary sample from a cohort of 7th grade students in a regular secondary school in West Sussex, United Kingdom.	MiRo-E is a biomimetic, mammal-like robot from Consequential Robotics built for education and HRI research. It is designed to be appealing and runs a brain-inspired, three-layer control system—from reflexive behaviors to higher-level cognitive loops.	Participants first completed an online questionnaire, then entered a room with a therapy dog, a therapy robot, and a guide, and stood at the center of a grid. They had a five-minute, video-recorded free-play session (timed by a stopwatch) with the option to stop anytime; the researcher announced start and end. Afterward, they returned to the lobby to complete a post-session evaluation questionnaire.	Participants preferred the live therapy dog over the therapy robot (p < 0.001; two had no preference). Most strongly enjoyed both (dog: n = 32; robot: n = 21), but satisfaction was higher after the dog (M = 4.85 ± 0.61) than the robot (M = 4.32 ± 1.03; Z = −2.42, p = 0.02). Positive-emotion words were similar at baseline (dog: 8.00 ± 4.02; robot: 8.08 ± 5.72; p > 0.99), unchanged after the robot (8.84 ± 4.02; p = 0.27), but decreased after the dog session.
Dinesen et al. (2022) [35]	Quasi-experimental research	This study examined how the social robot LOVOT interacts with people with dementia and the experiences of healthcare professionals who use it to engage with this group.	42 people with dementia, of whom 30 were assigned to group meetings and 12 to individual meetings	LOVOT is a 4.2 kg social robot by Groove X (28 × 43 × 26 cm) with AI for real-time, human-like movement. It has body-wide touch and distance sensors that detect warm/cold contact, respond to stimulation, and can even “fall asleep.”	The interaction between people with dementia and LOVOT was tested in individual and group sessions. The former lasted four weeks, while the latter lasted 12 weeks.	LOVOT did not produce clinically significant improvements in overall well-being for people with dementia, but it did boost immediate mood: facial expressions were more positive after sessions (individual or group). The effect was short-lived. Staff described LOVOT as fun and calming and noted it encouraged more communication and interaction with others and with staff.
Du et al. (2023) [28]	Systematic review of randomized clinical trials with meta-analysis	Compare and rank the effectiveness of animal-assisted therapy and robot-assisted therapy in treating dementia.	Nineteen randomized controlled trials (RCTs)	Robotic animals such as PARO	Interventions (with therapy dogs or robotic pets like PARO) combined hands-on contact (petting, feeding, grooming, play, talking) with simple cognitive tasks, motor/sensory exercises, and group social interaction. Sessions typically ran 30–60 min, delivered about 2–3 times per week.	Prior meta-analyses suggest animal-assisted therapy does not improve cognition, agitation, or quality of life, but may reduce depression—though these findings are weakened by study omissions and mixing designs. For robotic pets, evidence consistently shows reduced agitation, with mixed effects on depression and quality of life. A network meta-analysis found no significant difference between animal-assisted and robotic pet therapy in reducing agitation.
Geva et al. (2022) [24]	Quantitative methodology, Randomized experimental study	To determine whether the interactive features of the robotic seal PARO contribute to reducing pain and stress; to compare touch with the robot switched ON versus OFF	60 healthy adults. Participants were randomly allocated into an interactive group (ON) or a non-interactive group (OFF)	Social assistive robot PARO—a plush robotic baby seal with dual processors, microphones, 12 tactile sensors, whiskers and actuators; when petted it moves its tail, opens its eyes and vocalizes like a seal pup	After calibrating heat stimuli, participants received mild and strong heat-pain stimulation on their forearm while touching PARO; the robot was either interactive (ON) or turned off. A 5 min familiarization session preceded the test; pain and stress were measured with a visual analogue scale before and during touch.	Only in the interactive group did stress decrease significantly (baseline 2.9 ± 2.5 vs. touch 1.8 ± 2.1). Mild pain decreased significantly only in the interactive group (Δ = −1.3 VAS), while strong pain decreased in both groups. Greater perceived pleasantness and willingness to meet PARO again correlated with larger pain reductions. The authors conclude that the robot’s interactive qualities contribute to pain and stress relief.
Harris-Gersten et al. (2023) [36]	Programmatic evaluation used within-subjects, pre-post design	The purpose of the study is to examine the usability (frequency of use, frequency of reminders from caregivers) and acceptability (usefulness, satisfaction) of social robot pets among veterans with dementia.	The study sample consisted of veterans aged 65 and older living in the community and their informal caregivers enrolled in the COACH (Caring for Older Adults and Caregivers at Home) program operated by Durham Veterans Affairs (VA). Twenty veterans were included.	Social robot pets (provided by the Triangle J Area Agency on Aging)	Families received a 10–15 min at-home orientation from COACH staff; caregivers were taught basic operation and encouraged to prompt and support veterans’ use based on tolerance. After 3 months, caregivers completed a structured phone interview using the Social Robot Pet Intervention Usability and Acceptability Tool.	Most veterans used the robot pet frequently or daily (80%); one did not use it over 3 months. Reported benefits were modest on average (M = 1.35, SD = 0.8). Caregivers largely found the pet very helpful for veterans (n = 13) and themselves (n = 12); 80% said it somewhat/greatly reduced difficult symptoms. Satisfaction was high—89% (n = 18) met the threshold and would recommend a social robot pet for dementia care.
Ihamäki & Heljakka (2021) [40]	Qualitative exploratory intervention study	The study examined whether a commercial robot dog can spark positive emotions, social connection, and wellbeing in older adults during intergenerational group sessions, capturing their immediate, firsthand responses in a day-center activity with preschoolers.	(n = 10) elderly people (ages 65–80 years, 3 males, 7 female)	A low-cost interactive robotic pet (by Hasbro), called “Joy for All Companion Pets”. The pet-like robot Golden Pup from the sence was selected for the play intervention.”	Intergenerational 80 min day-center session with the Golden Pup robot: 20 min pet reminiscence, 30 min facilitated free play, brief emotion interviews at the end; entire session video-recorded.	Golden Pup interactions lifted mood and emotional wellbeing, with touch and play sparking social interaction, reminiscence, and storytelling. The robot facilitated communication among older adults and between them, preschoolers, and caregivers. Overall, such robot pets show promise as meaningful companions that enhance social connection and wellbeing in elderly care.
Jang (2020) [42]	Q methodology	The objective of this study is to understand the subjectivity of medical professionals in their perception of pet robots, describe the characteristics of each subjectivity type and understand the categorization of pet robots.	Twenty medical professionals (doctors or nurses) currently employed in a clinical setting	Interactive pet robots used in robot-assisted animal therapy.	Participants were asked to categorize 56 statements on pet robots. Statements on robot pets were categorized on a 14-point scale. Then an interview was conducted with the subject on the statements on the polar ends.	The key findings of the study revealed three distinct perceptions among medical professionals towards pet robots: those who value the emotional benefits, those who emphasize the ease of pet management, and those who focus on the convenience brought by technological advancements. Most participants had a positive attitude towards using robot pets in clinical settings, seeing them as useful tools for emotional support and patient care.
Jung et al. (2017) [37]	Qualitative exploratory study	The aim of this study is to inform the development of animal like robot companions that can understand and respond to human touch.	9 Dutch health-care providers from two facilities—4 without robot experience (layman) and 5 with Paro experience (experts).	Robot Paro	In this study, health-care providers were interviewed to explore their perceptions and experiences regarding the use of the animal like robot Paro in dementia care. The intervention involved introducing Paro to people with dementia as a social companion robot designed to engage patients through tactile interaction and auditory feedback.	Health-care providers generally have a positive attitude toward using Paro in dementia care, noting its potential to reduce patient stress and stimulate communication. The study also highlighted that patients primarily use gentle, affectionate touch gestures, while some negative touches tend to be accidental.
Klumpe et al. (2025) [33]	Quantitative methodology, cross-over design study	To compare social bonding between humans and real dogs versus a robot dog (AIBO). Assess urinary oxytocin (OXT) changes, self-reported attachment and evaluations of companionship.	19 female psychology undergraduates from one US university.	Robotic dog AIBO ERS-1000: has multiple microphones, cameras, pressure sensors on the paws and touch sensors on head/back/chin. It can walk, sit, lie down, shake hands, dance; uses reinforcement learning and responds to visual, voice and tactile cues.	Participants were randomized to begin with either the dog or AIBO. Each month included twelve weekly interaction sessions (petting, play and commands). Urine samples were taken at the start, middle and end of each phase. Participants completed attachment questionnaires and rated companionship; AIBOs were reset before first use and customized by participants.	Mixed-effects analyses showed that OXT increased during interactions with dogs but decreased with AIBO. Participants reported significantly stronger attachment to dogs and rated them as better companions. The study highlights that current robot dogs do not yet evoke social bonds to the same extent as real dogs, and future designs may need to leverage unique robotic features rather than mimicking dogs.
Koh et al. (2021) [47]	Scoping review	The aim of this review is to synthesize evidence on the delivery and impact of low-cost, familiarly and realistically designed interactive robotic pets for older adults and people with dementia.	9 studies	Low-cost and realistically and familiarly designed robotic pets (i.e., the Joy for All robotic cat and dog)	Low-cost Joy for All cats/dogs were used with older adults (including dementia) mainly at home or in long-term care for 2 weeks–6 months. Use was typically individual and self-directed with minimal staff/caregiver facilitation, though some studies added group or structured sessions.	The key findings indicate that robotic pets can effectively reduce loneliness and improve emotional well-being in older adults, especially those with dementia. They also promote social engagement and decrease agitation and anxiety. These benefits suggest that robotic pets are a valuable tool in supportive care.
Koh et al. (2022) [43]	A descriptive qualitative study	Study aims to explore the determinants of implementing pet robots for dementia care in nursing homes, from the perspectives of healthcare professionals and organizational leaders.	A total of 22 participants from eight nursing homes	Robotic pets equipped with interactive and responsive features designed to simulate real animal behavior.	Semi-structured interviews with healthcare professionals and organizational leaders were conducted. The intervention involves using pet robots as substitutes for live pet therapy in nursing homes to improve the psychosocial health of residents with dementia.	Key findings include that successful implementation of pet robots depends on their realistic design, affordability, and ease of cleaning; external factors like regulations and funding; alignment with nursing home care priorities; varied staff attitudes towards technology; and the importance of proper assessment and care planning.
Koh et al. (2023) [44]	A secondary qualitative analysis of data generated from in-depth, semi-structured interviews	To explore care professionals’ and organizational leaders’ ethical intuitions before and when implementing pet robots in nursing homes for routine dementia care.	2 care professionals and organizational leaders from eight nursing homes in Ireland.	Pet robots	Implementation of pet robots for nursing home residents with dementia	Ethical themes fell into three layers: individual/relational (respect autonomy, tackle social isolation, manage psychological impacts), organizational (reduce caregiver burden, adapt workflows, varied openness to tech), and societal (beliefs about dementia care plus justice concerns over affordability and access).
Lane et al. (2016) [31]	Pilot study; there was no randomization nor use of control groups in research design	To evaluate the effectiveness of the Paro robot as a nonpharmacological intervention for managing dementia-related mood and behavior issues in veterans in a VA long-term care facility.	23 veteran residents of a Veterans Affairs (VA) geropsychiatric long-term care facility	The Paro robot	The introduction and use of the Paro robot (a robotic pet) with veterans in a nursing home setting, observed during three periods—pre-Paro, during Paro, and post-Paro—to assess changes in mood and behavior related to interaction with the robot.	The study found that using the Paro robot increased positive mood and behaviors and reduced negative indicators in veterans with dementia. It was most effective with residents who were calm and approachable, suggesting that Paro is a promising nonpharmacological option for dementia care in VA long-term care facilities.
Li et al. (2025) [45]	Qualitative study using semi-structured interviews	To explore how telepresence via the mobile robot Enabot shapes users’ experiences and relationships with pets and others. Examine embodied projection (seeing the robot as an extension of oneself) and how this affects interactions.	22 Chinese pet owners aged 21–34.	Enabot (EBO), a home-based telepresence robot	Participants used Enabot in their homes to remotely observe and interact with their pets via a smartphone app. Interviews focused on telepresence, embodied projection and social impacts.	Researchers developed a typology of “embodied projection”: when users viewed the robot as an extension of themselves, they actively drew their pets’ attention via the robot. This reshaped the human–pet relationship and users treated the robot as a tool for self-care (e.g., soothing guilt or anxiety when away). The study shows telepresence robots are not merely functional devices but mediate emotional connections and personal wellbeing.
Logan et al. (2019) [27]	Pilot randomized controlled trial (RCT) with a between-groups, open-label design	To (1) describe the introduction of SR technology into the pediatric inpatient setting through an innovative partnership among a pediatric teaching hospital, robotics development, and computational behavioral science laboratories, and (2) present feasibility and acceptability data.	54 children medically or surgically hospitalized children ages 3 to 10 years	Social robot (SR) teddy bear and a tablet-based avatar of the bear; The Huggable robot was a plush teddy bear with a soft exterior, powered by an Android smartphone embedded inside.	Randomized trial with children (3–10 years): 30 min sessions in one of three arms—tele-operated Huggable robot, tablet-based avatar, or static plush bear. Child life specialists guided interactions; robot and avatar were “Wizard-of-Oz” controlled to appear autonomous.	The social robot intervention was feasible and well accepted, with most participants completing the study. Children exposed to the robot reported more positive affect and showed greater joy and agreeableness compared to those given a plush bear. Child life specialists also identified several potential emotional benefits of using social robots in pediatric care.
Moerman et al. (2019) [30]	Systematic state-of-the-art review	To inventory how socially assistive robots (SARs) are used in hospital settings to support children’s wellbeing during medical treatment and assess their effects.	10 publications on 8 studies; six different robots (1 humanoid and 5 pet-like) were used with hospitalized children.	Various SARs—humanoid and animal-like robots (e.g., bear- or seal-shaped robots) designed to provide interactive support, distraction and emotional comfort.	Robots were used to distract children during medical procedures, offer emotional support during illness and enhance wellbeing during hospital stays. Interventions included interactive play, conversation or pet-like companionship.	Studies reported positive effects: robots aided distraction and engagement, reduced stress or pain, increased relaxation, smiling and communication. Some adolescents in a psychiatric ward felt unsafe. The authors conclude that SARs may positively influence children’s wellbeing, but more research is needed to measure effect sizes and integrate robots into clinical routines.
Papadopoulos et al. (2018) [48]	A scoping review	To provide an overview of the existing evidence related to the views of nurses and other health and social care workers about the use of assistive humanoid and animal-like robots	19 articles; 4 used mixed methodology; 10 used qualitative methodology (semi-structured interviews or focus groups) and 5 used a quantitative methodology (survey or administration of a structured questionnaire)	Paro: an animal-like robot which takes the form of a Canadian harp seal. JustoCat: an interactive robotic cat with washable fur; Guide: a humanoid robot. Cafero: a humanoid robot Kompai: a humanoid robot Care-O-bot 3: a humanoid robotic assistant	In the 19 articles reviewed, researchers examined the views, attitudes, and perceptions of health and social care workers toward socially assistive humanoid and animal-like robots. 11 studies involved participants interacting directly with robots; 8 studies collected views without direct interaction, using surveys, questionnaires or scenarios.	Staff with hands-on robot experience were more positive; those without were unsure of clinical value. Views depended on perceived capability, reliability, and whether robots were companions or team members. Concerns included adjustment time for older adults, reduced human connection, and technical glitches. Most reported little workflow change, though some found work more engaging due to patient benefits; a few disliked sharing their workspace with robots.
Petersen et al. (2017) [25]	Randomized clinical trial	To assess the effectiveness of the PARO robotic pet, an FDA approved biofeedback device, in treating dementia-related symptoms.	61 patients with mild to moderate dementia, with 77% females, average 83.4 years in age	PARO robotic pet	The intervention involved older adults with mild to moderate dementia engaging with the PARO robotic pet for 20 min, three times a week, over a three-month period.	Participants in the PARO group showed higher pulse oximetry and GSR levels, and lower pulse rates, indicating reduced physiological stress. Their RAID and CSDD scores improved, reflecting decreased anxiety and depression. Additionally, the PARO group required less pain and psychoactive medication compared to the control group, suggesting that the robotic pet had a therapeutic effect on emotional and physical well-being.
Pike et al. (2020) [41]	Qualitative multiple case study; Qualitative semi-structured interviews.	To explore the effects and acceptability of a commercially available robot companion cat for people living at home with dementia, considering its impact on mood, behavior, and communication.	12 participants (people living at home with dementia symptoms, mostly female, older adults).	Ageless Innovation’ Joy for All Companion Cat: a non-programmable, interactive robot cat with movement, purring, and meowing responses to touch/light.	Participants were given a robot cat to keep at home. Initial interview after 2 weeks and follow-up at 3 months; use of photo elicitation with participants and families to explore experiences.	Robot cats, when accepted, improved communication, reduced withdrawal, provided distraction from repetitive behaviors, promoted calmness, and enhanced carer wellbeing. Benefits are most evident in moderate to severe dementia. Some rejections are due to dislike of cats or resistance to interventions.
Preuß & Legal (2017) [49]	Ethical analysis and literature review	To compare benefits and concerns of live animals versus robotic animal companions for elderly people in domestic smart home (AAL) settings, with ethical considerations.	N/A	Various animal shaped social robots (e.g., PARO seal, AIBO dog, NeCoRo/JustoCat cat, CuDDler polar bear, robotic rabbits) with interactive capabilities, sensors, and potential health.	Review of existing research and ethical considerations regarding replacing or supplementing live animals with robotic animal companions in domestic AAL settings.	Robotic animals can provide social interaction, reduce loneliness, improve mood, and avoid many drawbacks of live animals. Additional benefits include health monitoring via sensors. Concerns include loss of genuine human–animal bond, infantilization, deception, privacy risks, and cultural acceptance.
Pu et al. (2019) [38]	A qualitative study	To explore how people with mild to moderate dementia and chronic pain perceive PARO as an alternative intervention to manage their pain and mood	11 participants with dementia and chronic pain from three residential aged care facilities	PARO robotic pet	They interacted with PARO for 30 min, 5 days a week over a 6-week period. Participants then completed individual semi-structured interviews at the end of intervention	The PARO intervention is a promising intervention to improve positive emotions and there is some anecdotal evidence that pain may be decreased from the perspectives of people living with chronic pain and dementia. Long-term care staff may incorporate PARO therapy into daily dementia care.
Scerri et al. (2020) [29]	Systematic literature review, Meta-ethnography	To explore formal caregivers’ perceptions and experiences of using pet robots for people with dementia in long-term care, and factors influencing their use.	8 qualitative studies	Animal-like social robots (e.g., PARO seal, robotic cats, AIBO dog, JustoCat).	Use of pet robots in care routines and sessions varied by study (individual/group activities, tactile interaction, conversation facilitation).	Pet robots can reduce agitation, promote social interaction, increase comfort and engagement. Barriers include infection control, costs, stigma, ethical concerns. Successful adoption requires appropriate introduction, staff involvement, and ongoing support.
Sung et al. (2014) [32]	Pilot study using a one-group pretest–posttest design.	To evaluate the effect of robot-assisted therapy using (pet robot) PARO on social interaction and activity participation among institutionalized older adults.	16 older adults recruited, 12 completed in a residential care facility in Taiwan.	PARO therapeutic seal robot.	30 min group sessions twice weekly for 4 weeks, led by trained nurse. Interaction, tactile stimulation, facilitated conversation, cleaning PARO at session end.	Significant improvement in communication/interaction (p = 0.003) and activity participation (p = 0.008) after intervention. Supports feasibility of integrating PARO therapy in routine activities.
Valenti Soler et al. (2015) [26]	Quantitative methodology, Randomized controlled clinical trial, parallel groups	To compare effects of therapy sessions using humanoid robot (NAO), pet robot (PARO), and real dogs on behaviour, apathy, and quality of life in advanced dementia.	Nursing home: 101–110 patients (moderate/severe dementia). Day care: 17–20 patients. Various dementia diagnoses.	PARO (animal-like robot), NAO (humanoid social robot).	Therapy sessions twice weekly for 3 months. Structured cognitive, physical, and sensory activities. Robots/animals are integrated into standard therapy format.	NAO and PARO groups showed improvement in apathy (nursing home, Phase 1). PARO improved quality of life scores (Phase 2). NAO group saw cognitive decline in MMSE but not sMMSE. some benefits in NPI irritability (day care). Effects varied by dementia severity and setting.
Yuan et al. (2022) [39]	Qualitative (semi-structured interviews)	To identify the benefits and challenges of implementing different types of social robots in real-world aged care practice from care staff perspectives	11 staff members working a different location across six Australian residential aged care organizations (care manager, lifestyle coordinator, therapists, care assistants, director)	Two types: robopets (Paro, JFA robotic cats and dogs) and humanoid robot Nao	Three participants have used both Nao and robopets and the other participants have used either Nao or robopets.	Robots in aged care can enhance both staff and older adults’ wellbeing, but their success depends on addressing training, ethical, and relational challenges within the care triad.

summarizes the main characteristics of the selected studies. The table includes information on the authors and year of publication, type of study, study aims, population and sample, type of AI or robotic therapy, intervention characteristics, and key findings. Presenting these features in a comparative format supports a clearer understanding of the scope, methodological diversity, and principal outcomes of the studies included in this review.

Most studies focused on enhancing social inclusion, improving mood, and alleviating anxiety, pain and stress. They also addressed issues of implementation and ethics. Sixteen different robots were used in the studies, six of which did not specify the robot used. The most frequently mentioned robots were the PARO robot (n = 13) [14,24,25,26,28,29,31,32,37,38,39,48,49], and the JustoCat robot [14,29,48,49]. The AIBO [14,29,33,49] and Joy for All cat/dog [39,40,41,47] robots were each used in four studies. The interventions were mainly structured group sessions (one to three times per week for 30–60 min over four to 12 weeks), with ad hoc/individual uses or one-time presentations being the exceptions. The largest populations in the included studies were patients with dementia (n = 244–256) and children (n = 122). Most studies were based on qualitative approaches (n = 10) or literature reviews (n = 8) (Table 7).

3.3. Thematic Synthesis of Included Studies

3.3.1. Study Selection

The included evidence consisted of a broad range of study designs, including systematic reviews, scoping reviews, randomized controlled trials, quasi-experimental studies, pilot studies, qualitative studies, mixed-methods research, and ethical or implementation-focused analyses [14,24,25,26,27,28,29,30,31,32,34,35,36,37,38,39,40,41,42,43,44,45,46,47,48,49]. The studies were selected because they addressed the use, impact, acceptability, or implementation of social robots, robotic pets, humanoid robots, or telepresence robots in health, care, therapeutic, or social-support contexts [14,28,29,30,39,43,44,46,48,49].

3.3.2. Characteristics of Included Studies

The included studies varied considerably in methodological design and strength of evidence [14,24,25,26,27,28,29,30,31,32,34,35,36,37,38,39,40,41,42,43,44,45,46,47,48,49]. Reviews summarized existing evidence on socially assistive robots, robotic pets, and robot-assisted therapy in older adult, dementia, or pediatric care [14,28,29,30,46,47,48]. Quantitative studies included randomized controlled trials, crossover designs, and quasi-experimental or pre–post studies that measured outcomes such as agitation, depression, quality of life, social interaction, activity participation, stress, pain, and physiological responses [24,25,26,27,31,32,33,34,35,36]. Qualitative studies explored lived experiences, staff perspectives, ethical views, implementation determinants, and user perceptions of robotic pets and social robots [37,39,40,41,43,44,45].

3.3.3. Target Populations and Settings

Most studies focused on older adults, particularly people with dementia living in nursing homes, residential aged care, long-term care facilities, or community dementia-care contexts [14,25,26,29,31,32,35,36,38,39,41,43,44,46,47]. Several studies included healthcare professionals, care staff, organisational leaders, or caregivers to examine implementation, acceptability, and professional perceptions of robot use in care settings [29,36,37,39,42,43,44,48]. A smaller group of studies focused on children, including hospitalized children, children undergoing medical treatment, and school-aged children interacting with therapy dogs or biomimetic robots [27,30,34]. Other populations included healthy adults exposed to PARO during experimental pain testing and young adult pet owners using telepresence robots to interact with pets [24,45].

3.3.4. Intervention Characteristics

The most common intervention modality was the use of animal-like robotic pets, particularly PARO, Joy for All robotic cats and dogs, LOVOT, AIBO, MiRo-E, Golden Pup, NeCoRo, JustoCat, and CuDDler [14,24,25,26,29,31,32,33,34,35,38,40,41,46,47]. Some studies also examined humanoid or telepresence robots, including NAO, Huggable, and Enabot [26,27,30,39,45]. Interventions were delivered individually or in groups and included tactile interaction, free play, companionship, structured therapeutic activities, cognitive or sensory stimulation, communication facilitation, emotional support, distraction during medical procedures, and remote pet interaction [14,26,27,30,32,34,35,38,40,41,45]. Session duration varied widely, ranging from brief 5 min interactions to repeated 30–60 min sessions delivered over several weeks or months [25,26,27,32,34,35,38,40].

3.3.5. Main Patient-Related Outcomes

Across the studies, the most frequently reported benefits were improved mood, increased social engagement, enhanced communication, greater activity participation, reduced loneliness, and improved emotional well-being [14,30,31,32,35,38,40,41,46,47]. In dementia care, robotic pets were associated with reduced agitation, increased comfort, improved interaction, and greater engagement, although effects on depression, anxiety, quality of life, and long-term well-being were inconsistent [14,25,26,28,29,35,38,47]. In children, social robots supported distraction, engagement, positive affect, communication, and emotional support during hospital care or play-based interaction [27,30,34]. In experimental settings, interactive PARO was associated with reductions in stress and mild pain, suggesting that robot responsiveness may contribute to therapeutic outcomes [24].

3.3.6. Adverse, Mixed, or Null Findings

Although many studies reported positive outcome patterns, several findings were mixed or limited [14,25,26,28,33,34,35,36]. Some studies found no clinically significant improvement in overall well-being, quality of life, sleep, anxiety, or depression, despite positive short-term effects on mood or engagement [14,28,35]. Real animals were often preferred over robot animals, and social bonding with robotic dogs appeared weaker than bonding with real dogs in some comparative studies [33,34]. Some participants rejected robots, found them strange, disliked the animal form, or experienced emotional overstimulation during interaction [30,35,41]. Methodological limitations were also common, including small samples, short intervention periods, lack of control groups, heterogeneous outcomes, unclear long-term effects, and variable study quality [14,28,30,31,32,35,36,46,47]. Therefore, the overall evidence suggests promising but still preliminary support for robotic interventions, with stronger research needed to confirm effectiveness, safety, acceptability, and implementation value [28,29,30,39,43,44,46].

4. Discussion

This review synthesized evidence on animal-assisted interventions (AAI) and robot-assisted interventions (RAI), with particular attention to robotic pets and socially assistive robots used to support emotional well-being, social engagement, communication, and care-related outcomes. Overall, the findings suggest that RAI may offer meaningful short-term benefits across different populations and settings, especially where living animals are difficult to implement. However, the evidence also shows important variation in outcomes, study quality, intervention design, and user acceptability. The following sections discuss these findings in relation to existing evidence, population-specific effects, possible mechanisms of engagement, current limitations, and future development needs.

4.1. Comparison Between AAI and RAI

Use of our review’s findings in comparison with broader evidence helps situate the relative strengths and limits of AAI versus RAI interventions. Both AAI and RAI share the capacity to improve mood, reduce agitation, and foster social engagement. Evidence on robopets and socially assistive robots shows consistent reductions in agitation and improvements in interaction and emotional expression, broadly aligning with reviews that point to shared mechanisms with AAI, including social connection, meaningful activity, and positive effect [14,28,35,50]. However, live-animal–based interventions often demonstrate larger, more durable, and physiologically grounded effects. A meta-analysis of AAT found moderate effect sizes on outcomes including behavioral problems, emotional wellbeing, medical difficulties, and autism-related symptoms [51]. In older adults, dog-assisted therapy shows a moderate, significant reduction in depressive symptoms, while robotic-pet interventions generally do not [13]. In dementia care, both AAI and robotic-pet therapy reduce agitation, with limited and inconsistent effects on depression, cognition, and quality of life, although AAI tend to marginally outperform RAI [28]. A recent RCT comparing dog-assisted and robot-dog-assisted therapy in children with autism or Down syndrome showed that children in dog-assisted therapy had significantly greater improvements in emotional attunement and emotion regulation compared with both the robot-dog and no-treatment control groups [52]. Taking together, these findings suggest that while RAI can achieve many of the same short-term outcomes as AAI, they should be seen as complementary rather than substitutive approaches. From a practical and ethical standpoint, robots offer clear advantages where animals are impractical, as they eliminate concerns about allergies, hygiene, animal welfare, handling resources, or safety risks. Reviews and conceptual work emphasize that RAI offers scalability, control, and hygiene advantages but lacks the emotional depth of AAT [50,53].

4.2. Effects Across Populations and Contexts

Across populations, robotic pets provide meaningful emotional and social benefits, with effects varying by context. Among people with dementia, studies consistently report reduced agitation or stress and improved mood and social engagement. Benefits can be short-term, but broader outcomes, such as reduced loneliness and higher caregiver satisfaction, support their nonpharmacological value [14,25,28,35,36,47]. Implementation hinges on affordability, hygiene, and organizational support due to stigma and maintenance demands [29,43]. Among older adults without cognitive impairment, robotic companions such as PARO and Golden Pup enhance mood, communication, and social interaction. In some settings, simple alternatives, such as plush toys or digital programs, produce similar benefits, depending on preferences and care context [32,40,46]. In children, robots provide joy, relaxation, and engagement. Children often prefer live dogs, yet robots deliver emotional benefits in structured clinical settings, reducing stress or pain and promoting communication [27,34]. Occasional discomfort among adolescents highlights the need for developmentally sensitive integration [30]. Clinicians generally perceive robotic pets as a cost-effective intervention that supports patient well-being and provides caregiver relief, although they also highlight technical limitations and potential concerns regarding diminished human interaction. Ethical priorities include autonomy, equitable access, and human-centered care. Future designs may strengthen bonds via distinct robotic features [33,37,39,42,44,48].

4.3. Mechanisms of Interaction and Engagement

In healthy adults, interactions with PARO reduce stress and mild pain, with enjoyment moderating effects [24]. Overall, robotic pets may support emotional well-being and social engagement across different groups when implemented thoughtfully, particularly when their use is aligned with user needs, ethical considerations, and their intended complementary, nonpharmacological role. In healthy adults, the effects of PARO may not be attributable solely to tactile stimulation but could also relate to the robot’s perceived interactivity and social responsiveness. Earlier experimental work similarly suggested that touching PARO, compared with its mere presence, was associated with reduced pain perception and increased happiness, with stronger hypoallergic effects observed among participants who reported more positive experiences and a greater perceived ability to communicate with the robot [24,54]. Findings in cognitively intact older adults likewise suggest that potential benefits may depend not only on exposure to PARO itself, but also on the quality of engagement, as participants generally perceived the robot as easy to use and potentially useful for themselves and others, while more active engagement was associated with greater positive effect following the interaction [55]. At the same time, the broader PARO literature warrants cautious interpretation, as systematic reviews point to potentially beneficial effects on quality of life and biopsychological outcomes, but also highlight substantial heterogeneity and predominantly low-to-moderate study quality [46,56]. This may partly help explain why robotic companions, despite showing promising short-term emotional benefits, may still elicit less pronounced social bonding than live animals, as a recent crossover study found stronger attachment and greater oxytocin increases with real dogs than with AIBO in healthy young adults [33].

4.4. Identified Gaps and Ongoing Challenges of the Included Studies

Across the included studies, several common limitations were identified by the original authors. Many studies used small, non-representative, or self-selecting samples, often drawn from specific populations such as older adults or individuals with dementia, which limits generalizability. Short study durations and brief exposure to robots frequently introduced novelty effects, reducing insight into long-term outcomes. Many relied on self-reported data or unblinded designs, raising the risk of response and observer bias. Outcome measures were sometimes insensitive or inconsistent, and methodological rigor varied, with low-quality RCTs, limited theoretical grounding, and heterogeneous designs. Several studies focused on caregiver or staff perspectives rather than direct users, omitting key viewpoints. Additionally, cultural and contextual factors were often overlooked, and many reviews included English-language studies only, which may have excluded relevant international evidence.

4.5. Future Perspectives and Development Pathways

Our analysis revealed further limitations present across the included studies. High heterogeneity across robot types, populations, designs, and measures limits comparability and synthesis. Small samples and short-term or context-specific interventions further constrain generalizability. Collectively, these factors highlight the need for larger, longer-term, and more standardized studies to better understand the effectiveness and acceptability of AI-based animal therapy across settings and populations.

4.6. Limitations of the Present Review

Our review should be interpreted considering several limitations. The search was limited to English-language publications from 2014 to 2025 and did not include grey literature, so some relevant evidence may not have been captured. The review protocol was not prospectively registered, which may affect transparency. Data extraction was conducted by one reviewer and checked for consistency, but duplicate extraction could have further strengthened the process. The included studies were also diverse in terms of populations, robot types, intervention formats, settings, and outcome measures, which made direct comparison difficult and prevented meta-analysis. In addition, the inclusion of both primary studies and review articles provided a broad overview but may have introduced some overlap in evidence. Finally, some included studies had small samples, short follow-up periods, or limited reporting of methodological details, so the findings should be understood as promising but still developing evidence.

5. Conclusions

This systematic review suggests that AI-based pet therapy, including robotic pets and socially assistive robots, may be a promising complementary non-pharmacological approach for improving patient care. Across the included studies, robotic interventions were most commonly used with older adults and people with dementia, but evidence also included children, veterans, healthy adults, healthcare professionals, and community-dwelling users. The most frequently studied robotic pet was PARO, while other interventions included LOVOT, Joy for All robotic pets, AIBO, MiRo-E, NAO, Huggable, and Enabot. Interventions varied widely in format, ranging from brief free-play encounters to structured individual or group sessions delivered over several weeks or months.

Overall, the findings indicate that AI-based pet therapy can support emotional well-being, communication, social engagement, and activity participation. In dementia care, robotic pets were associated with reduced agitation or stress, improved mood, and increased interaction, although effects on depression, cognition, quality of life, and long-term well-being were inconsistent. In children and experimental settings, robotic interventions showed potential for reducing stress or pain and supporting engagement, although live animals were often preferred and may produce stronger social bonding.

Despite these promising findings, the strength of evidence remains limited. Many included studies had small samples, short intervention periods, heterogeneous designs, unclear or incomplete reporting, and limited follow-up. Several studies lacked control groups or used outcome measures that were difficult to compare across settings. These limitations reduce confidence in the durability and generalizability of the reported effects.

AI-based pet therapy should therefore be considered a supportive and complementary intervention rather than a replacement for human care or live animal-assisted therapy. Its greatest value may be in settings where live animals are impractical because of hygiene, allergies, safety, animal welfare, or resource limitations. Future research should use larger samples, longer follow-up, standardized outcome measures, and stronger study designs to clarify effectiveness, acceptability, safety, ethical implications, and implementation pathways across different patient populations and care environments.