Actionable companion · Cognitive load & attention

Learning with agency: Effects of Virtual Pedagogical Agents and Instructional Content Type on Learners’ Experience in Online Learning: Evidence from Behavioral, Eye-Tracking, and EEG Data

AI-assisted editorial implementation guide, separate from the original research. Proposed actions are not claims or recommendations attributed to the source authors. No independent peer review of this guide is asserted.

Read original paper Search paper agent CC BY 4.0

Evidence and context

Title and abstract screening; stage labels describe reported settings, not verified numerical ages. Design labels require full-text appraisal. Reported stage(s): Higher education. Verify the participant population and task before transferring the findings.

Original authors’ abstract, reproduced under the linked source license:

Online learning has become an important means of acquiring knowledge. Virtual pedagogical agents have gradually evolved from auxiliary tools into intelligent instructional media capable of affective interaction. However, how these agents affect learning outcomes remains unclear. Integrating social agency theory and cognitive load theory, this study developed a model in which virtual pedagogical agent type was the independent variable; learning experience was the dependent variable; psychological distance, emotional attachment, and attentional engagement were mediators; and instructional content type was the moderator. A 2 (virtual pedagogical agent: human-like vs. non-human) × 2 (instructional content type: traditional knowledge vs. interest-oriented popular science) mixed design was used. Questionnaires, eye tracking, and electroencephalography (EEG) were combined to assess the learning processes of 40 university students. The results showed that the interaction between agent type and instructional content type significantly affected learning experience. Specifically, in learning contexts featuring traditional knowledge content, human-like virtual agents were associated with higher learning satisfaction, greater emotional attachment, and lower psychological distance. In contrast, during interest-oriented content learning, non-human agents achieved superior knowledge recall scores, along with higher learning satisfaction, stronger emotional attachment, and lower psychological distance. Furthermore, eye-tracking data showed that attentional allocation was also affected by this combination. EEG data showed that FCz spectral power varied with it as well, however, because power rose simultaneously in theta, alpha, beta, and gamma; this broadband pattern is not an unambiguous index of deeper processing and may partly reflect a general increase in signal amplitude. In addition, the moderated-mediation analysis revealed that psychological distance and emotional attachment mediated the relationship between virtual-agent type and learning experience, while knowledge type moderated these mediating paths. The practical implications for online learning are discussed.

What kind of evidence is this?

Conversational & pedagogical agents · Survey / observational study

Stage: Higher education. Screening labels describe reported context and study-design language. They are not a risk-of-bias rating or proof of effectiveness for an age band.

Study methods · source passage

Forty university students volunteered to participate, including 11 men and 29 women. The gender distribution was 6 men and 14 women in the traditional knowledge group and 5 men and 15 women in the interest-oriented popular science group, with similar gender ratios between the two groups. All participants had experience with online learning and were physically and psychologically healthy. All participants were university students majoring in social-science-related fields with comparable educational levels. No participants reported conditions such as dyslexia or attention-deficit hyperactivity disorder (ADHD), or other traits that could impair information processing and learning abilities. None reported a psychological disorder, cold, fever, or other condition that could affect experimental performance. All had normal or corrected-to-normal vision, no color blindness or color-vision deficiency, and could clearly view the computer display. Participants read and signed written informed-consent forms and received a small cash payment after the experiment. The study was approved by the Academic Ethics Committee of Guangdong University of Technology (Approval No. GDUTXS20250015; approved on 21 February 2025). The sample size was determined by the throughput of the combined eye-tracking and EEG protocol, which requires approximately one hour of laboratory time per participant, and by the availability of the recording facility.

Limitations and uncertainty · source passage

Integrating social agency theory and cognitive load theory, this study combined questionnaires, eye tracking, and EEG to examine how the interaction between virtual pedagogical agent type and instructional content type affects learning experience in online learning. A consistent interaction emerged. For learning satisfaction, psychological distance, and emotional attachment, human-like agents produced better outcomes with traditional knowledge content than with interest-oriented popular science content. Non-human agents showed the opposite pattern: interest-oriented popular science content produced greater satisfaction and emotional attachment and less psychological distance. Knowledge recall supported this pattern only for the non-human agent. Recall was significantly higher for interest-oriented popular science content than for traditional knowledge content, whereas recall did not differ by instructional content type with the human-like agent. The mediation analyses clarified how this interaction affected satisfaction. Psychological distance and emotional attachment produced significant moderated mediation effects between agent type and learning satisfaction. Thus, the agent-content combination first changed learners’ psychological distance from and emotional connection to the agent, which then affected satisfaction. This finding supports the socioemotional pathway proposed by social agency theory and indicates that the relation between agent-design features and instructional content, rather than either in isolation, is what bears on trust and psychological closeness. By contrast, the moderated mediation effect through attentional engagement was not significant. FCz spectral power also showed a significant interaction. With the human-like agent, power did not differ by instructional content type. With the non-human agent, FCz spectral power was significantly higher for interest-oriented popular science content than for traditional knowledge content in every frequency band. Because power rose simultaneously in all four bands, including alpha, this broadband pattern is not an unambiguous index of deeper processing and may partly reflect a general difference in signal amplitude. It is reported here as a difference in FCz spectral power that accompanied the behavioral interaction, and no stronger inference is drawn from it.

Design the system before the pilot

Define whether the agent asks questions, gives hints, explains or assesses. Test hallucinated feedback, excessive praise, misleading confidence and responses when the learner is confused. A conversational interface alone does not establish autonomous agency.

Limit simultaneous instructions and unnecessary dialogue. Compare task performance and a suitable mental-effort measure; easier-feeling work is not automatically better learning.
Measure engagement separately from achievement. Check whether novelty-driven participation persists and whether it translates to independent performance.
Use common rubrics, blind scoring where feasible and a fresh unaided reasoning task. Document permissible assistance and avoid treating AI-detector scores as proof of misconduct.

A 90-day district pilot

Days 1–15 · Appraise. Have the curriculum lead and a teacher read the methods and limitations. Record the study design, sample, comparator, outcomes and evidence gaps. Define a local question that the study can help inform.
Days 16–30 · Design. Choose one disciplinary task. Require a first attempt, a documented AI consultation, a critique of the output and an independent explanation. Align the comparison group and scoring rubric with the same objective. Name an accountable owner and set the baseline, success threshold and budget before implementation.
Days 31–60 · Test. Start with a small voluntary cohort, preserve an accessible non-AI route and require teacher review. Use synthetic or de-identified examples; approve data handling and tool terms through district processes.
Days 61–90 · Decide. Blind-score reasoning and accuracy on a new task without AI. Distinguish model examination scores, student perceptions and actual learner gains; only the last directly tests educational benefit. Compare with baseline, document uncertainty and stop or revise if learning worsens, access is unequal or critical errors persist.

The board decision brief

Report the learning problem, evidence quality, participating population, total cost including review time, measured benefit, errors, access differences and remaining uncertainties. Make a dated recommendation to stop, revise or expand. A positive perception survey alone does not establish learning improvement.

Attribution and reuse

Shasha Liu; Xiaoqi Shen; Tiansheng Xia. Effects of Virtual Pedagogical Agents and Instructional Content Type on Learners’ Experience in Online Learning: Evidence from Behavioral, Eye-Tracking, and EEG Data. Behavioral Sciences (2026-09-19). https://doi.org/10.3390/bs16091691

Companion prepared 2026-10-30. Original authorship and DOI are preserved. The abstract is an attributed source extract; the implementation plan is new AI-assisted editorial material, offered under the same CC BY 4.0 license. This guide has not undergone independent peer review.