The present study examined whether socio-motivational effects from working with a Teachable Agent (TA) might transfer from the formative learning phase to a summative test situation. Forty-nine students (9-10 years old) performed a digital pretest of math skills, then played a TA-based educational math game in school over a period of eight weeks. Thereafter, the students were divided into two groups, matched according to their pretest scores, and randomly assigned one of two posttest conditions: either with the TA present, or without the TA. Results showed that low-performers on the pretest improved significantly more on the posttest than did high-performers, but only when tested with the TA. We reason that low-performers might be more susceptible to a supportive social context – as provided by their TA – for performing well in a test situation.