Back to blog
AI Avatars6 min read

What Research Says About Pedagogical Agents in Learning

A careful look at digital learning characters, the research signal, and Learny's safety-minded product direction

Written by Learny team

Product note. Learny is in development. Product capabilities described here are planned or may evolve during early access.

Editorial illustration of three expressive digital learning characters
AI AvatarsLearny editorial

What is a pedagogical agent?

Researchers often use “pedagogical agent” for a visible digital character designed to support learning through explanation, prompts, feedback, or encouragement. The category includes many technologies and learning settings; it is broader than today's generative AI companions.

That distinction matters. Evidence about pedagogical agents can inform design questions, but it cannot be transferred automatically to every new AI product.

What the reviews found

A 2013 meta-analysis reviewed 43 studies involving 3,088 participants and reported a small but statistically significant positive effect on learning. The analysis found variation across learner groups and design choices, including stronger reported benefits for K–12 learners than for post-secondary learners.

A later umbrella review examined 17 systematic reviews and meta-analyses. It described small positive effects across learning, motivation, and some affective outcomes while emphasizing that embodiment, role, interaction, and research quality all matter.

What the evidence does not establish

These figures do not prove that any specific avatar, conversational model, or app improves learning. They also do not validate Learny, which requires its own product-specific testing and evidence.

A character can be distracting, unclear, or poorly matched to a task. The relevant question is not whether an avatar exists, but whether its behavior supports a well-defined learning objective.

Learny's early-access design direction

Learny is designing AI characters as approachable guides for conversation and short practice. Planned behaviors include clear explanations, gentle prompts, and topic-aware responses within a safety-minded experience.

We are not claiming that the characters remember a learner across sessions or deliver proven outcomes. The early-access stage is where we intend to test clarity, usefulness, boundaries, and how families understand the companion's role.

Research & context

Sources and further reading

Schroeder, Adesope, and Gilbert — How Effective are Pedagogical Agents for Learning?Schroeder, Davis, and Yang — Designing and Learning With Pedagogical Agents
Inside Learny

Discover the thinking behind Learny, then join early access to follow the product as it develops.

See the learning loopExplore how guided conversation, focused practice, rewards, and parent context connectExplore learning pathsSee the topics Learny is being designed to support through early accessReview our safety approachRead about parent visibility, privacy, and safer boundaries for AI-supported learning
Join early access
Early access

A more personal way to learn is taking shape.

Request early access and we will let you know when a place becomes available. No payment is taken.

Request early access Read about safety