A prerequisite for low-stakes activities to improve learning is to keep students engaged when confronted with challenging material. Comparing a personalized and non-personalized version of computerized practising, this study experimentally evaluates the relationships between student effort and ability across different dimensions of task perceptions. Students practise longer when the task is perceived to be not too difficult. Students assigned to a personalized version of the tool have a lower success rate while practising, but this does not translate to differences in practice intensity, task perceptions, or summative test scores. In a personalized practising environment, perceived interest and usefulness both have the potential to promote engagement, albeit in different ways. Students with a lower level of subject-specific ability find the tool more interesting and, particularly, consider the non-personalized tool difficult but useful. Instead, in the personalized condition, it is the group of students with higher prescores who value the tool as relatively useful. These results indicate that more students will remain engaged with adaptive practising if software takes into account such differences. Multiple approaches and algorithms may thus be necessary to optimally adapt practising to individual learners.