Synthetic relationships with social pedagogical agents in education: A scoping literature review

Sebas Nouwen^1*Janienke Sturm²Uwe Matzat¹Wijnand IJsselsteijn¹

• ¹Eindhoven University of Technology, Eindhoven, Netherlands
• ²Fontys University of Applied Sciences, Eindhoven, Netherlands

Introduction: As AI in education increasingly takes the form of Social Pedagogical Agents (SPAs), learners begin to relate to these systems in human-like ways. Concepts such as social presence, affective support, trust, and rapport are widely studied but scattered across domains, labels, and measures. This scoping review maps how "synthetic relationships" between students and SPAs are described in empirical educational research and how SPA design and interaction features relate to students' learning experiences and, when reported, outcomes. Method: Following PRISMA guidelines for scoping reviews, database searches (Scopus, Web of Science, ACM Digital Library, IEEE Xplore, ERIC, PsycINFO) were conducted in August 2025 for peer-reviewed English-language studies (2010–2025) in educational settings involving SPAs with social or relational capabilities and indicators of relationship quality. We excluded studies on physical robots, noneducational contexts, purely functional outcomes, or nonempirical/grey literature. After screening 216 unique records, 29 studies were retained and thematically analyzed to identify recurring design levers, relational constructs, and links to learning. Results: Studies showed substantial terminological and methodological diversity, with short, single-session experiments predominating. Eight recurring design levers shaping relationship quality emerged: (1) voice and appearance, (2) congruence and alignment, (3) empathy expression, (4) social safety and disclosure, (5) transparent personalization, (6) refutational explanations, (7) memory and consistency, and (8) interaction duration. Across studies, SPAs enhanced four key aspects of relationship quality: (a) social presence, (b) affective support, (c) trust, and (d) rapport. Learning gains were strongest when these indicators aligned with sound pedagogy. Discussion: Synthetic relationships with SPAs function mainly as enabling conditions that enhance the learning climate—motivation, effort, and honesty—rather than directly improving test scores. We group the eight design levers into four broader categories: expressive (how the agent signals), relational (how it connects), transparency (what it communicates and why), and structural (how interaction unfolds). These, paired with the four relational indicators, form a practical framework for optimizing student–SPA relationships. Future work should prioritize multi-session and longitudinal designs, standardize relational measures, and address ethical concerns around purpose, transparency, influence limits, and shared oversight. When coupled with sound pedagogy, SPAs show strong potential to augment teaching and enrich learning.