An often overlooked benefit of Google Scholar is that the tool often shares interesting studies based on your own work and interests This is how I discovered this meta-analysis.

The results summarized:

• Feedback supports learning from text.
• Feedback is particularly effective if provided directly after reading.
• If provided after reading, elaborate and correct-answer-feedback are most effective.
• Computer-delivered feedback is more supportive than non-computer-delivered feedback.

Or as the researchers Swart et al. summarize their study in their own conclusion:

The results of our meta-analysis show that providing feedback during or directly after reading has a positive effect on learning from text. Promoting learning from text appears to be most effective when feedback is presented after reading a text and at least contains the correct answer or is elaborate. Feedback during reading is less effective, probably because it hinders the natural reading process, thereby placing too heavy demands on working memory (i.e. split-attention effect). Additionally, computer-delivered feedback is more beneficial for learning from text than non-computer-delivered feedback. When developing or choosing (educational technologies for) instructional strategies to support learning from text, one should keep in mind that it seems best to minimally interrupt the reading process (i.e. placing minimal load on the limited working memory capacity of the reader), and to help the reader evaluate and, if necessary, revise the mental model of a text with the help of questions or tasks and subsequent elaborate computer-delivered feedback directly after reading.

Abstract of the study:

The aim of the present meta-analysis was to examine the effects of feedback on learning from text in conventional readers (ranging from primary school students to university students). Combining 104 contrasts of conditions of reading texts with and without feedback, including 6,124 participants, using the random effects model resulted in a positive effect of feedback on learning from text (g+ = 0.35). Moderator analyses showed that feedback is particularly effective if provided directly after reading, but less so when provided during reading. If feedback is provided directly after reading, elaborate feedback and knowledge-of-correct-response feedback were more effective than knowledge-of-response feedback. If feedback is provided during reading, no differences are found between the effects of different types of feedback. Additionally, computer-delivered feedback is more beneficial for learning from text than non-computer-delivered feedback. Implications for optimizing conditions to support learning from text are discussed.