A recent study by researchers at Berlin’s Cluster of Excellence Science of Intelligence reveals critical insights into the design of robotic tutors and AI-driven educational assistants. Published in Communications Psychology, the research challenges the assumption that highly personalized automated feedback is always optimal. Instead, it demonstrates that the timing and cognitive load of robotic support significantly influence learning outcomes. Led by first author Helene Ackermann, the team examined how ninety adult learners responded to three distinct feedback protocols delivered by a humanoid robot during a spatial puzzle task. The robot provided immediate corrective feedback after each attempt, ranging from simple error notifications to detailed, personalized hints that referenced the learner’s prior steps and self-reported emotional states. The findings indicate a clear trade-off. While personalized feedback correlated with improved overall task performance, it simultaneously increased cognitive load immediately following an error. Learners presented with highly specific, multi-step hints struggled to process the information in real time, which temporarily hindered their next response. Conversely, simpler, task-focused feedback proved more effective for immediate recovery after mistakes. Individual differences further complicated the effectiveness of automated support. Participants with higher baseline cognitive ability benefited less from additional hints, likely because they could independently navigate the task without external guidance. Interestingly, learners who reported higher levels of boredom showed marked improvement when receiving direct, task-focused feedback, suggesting such prompts successfully redirected their attention. These results underscore that effective educational technology cannot rely on static personalization alone. The research emphasizes that AI tutors must dynamically assess a learner’s situational cognitive and emotional states to determine when to offer detailed guidance and when to withhold it. As humanoid robots and AI assistants move from experimental settings into classrooms and workplaces, the study establishes that intelligent support is defined not by the volume of information provided, but by the precision of its delivery. Future systems must prioritize adaptive timing and real-time affective monitoring to avoid overwhelming users while maximizing long-term comprehension.