This exploratory study sheds new light on students' perceptions of online feedback types for a complex writing task, summary writing from spoken input in a foreign language (L2), and investigates how these correlate with their actual learning to write. Students tend to favour clear-cut, instructivist rather than constructivist feedback, and guided self-evaluation through model solutions in online learning environments. However, the former type is too limited to tackle all dimensions of advanced writing. Constructivist feedback, in the form of guided modelling, allows addressing the higher-order concerns involved in summary writing. In addition, it is widely acknowledged that activating the zone of proximal development (ZPD) through cognitive involvement is beneficial to learning. To investigate students' learning from both types of feedback, a one-group pre-post-test intervention study was set up. Students attending a course on summary writing in L2 within a bachelor programme in Applied Languages (n = 38) followed an individual online learning module containing both instructivist fill-the-gap exercises and model solutions with constructivist guiding questions for self-assessment. The students' actual learning gain was measured through pre- and post-tests, and compared with their perceived learning gain, as expressed in self-evaluation. The comparison reveals a dichotomy between the students' observed learning curve and an underestimation of their own progress. This dichotomy was found to originate in a mismatch between their expectations towards the online learning module and the characteristics of the constructivist feedback conveyed. This mismatch can be attributed to three key factors: (1) evaluation, (2) linguistic focus, and (3) learner motivation.