The basic idea has been repeated since Dunlosky et al. (2013): highlighting is bad if you want to use it as a study method. This new meta-analysis by Ponce et al. makes another sound as it distinguishes between student- and teacher-generated highlighting on the one hand and highlighting for memorization versus comprehension on the other.

Starting the results of this meta-analysis with the latter:

• “…on average, learners who used highlighting as a learning strategy (experimental groups) improved memory (e.g., recall of main ideas) by 0.36 standard deviations compared to learners who only read or studied the text (control groups). This mean effect size is within Hattie’s “hinge point” range, indicating that, on average, highlighting as a learning strategy has a significant and relevant impact on supporting students to remember important ideas in the learning material.”
• “…on average, learners who used highlighting as a learning strategy did not show improvements in comprehension compared to learners who only read or studied the text.”

But also:

The results from the meta-analyses also demonstrated that when college students used highlighting as a learning strategy, they benefited significantly but K-12 students did not, with effect sizes of 0.39 and 0.24, respectively (Hypothesis 2). These findings demonstrate that less experienced learners may not know what to highlight, so training will be more necessary for younger students. Rickards and Denner (1979) argue that children compared with more mature students are still developing meta-comprehension skills so such deficiency may lead them to select details rather than main ideas to highlight. In a more recent review, Miyatsu et al. (2018) also have noted that students who are younger than college age generally do not benefit from learner-generated highlighting. Studies with K-12 students on highlighting have employed expository texts; however, children are more familiar with narrative texts, which present important differences with expository texts not only in terms of content but also in their organizational structure. Such unfamiliarity with text structure is an important limitation for the appropriate use of highlighting as learning strategy (Meyer & Poon. 2001; Ponce et al. 2018a).

Moving to the first distinction:

Regarding instructor-provided highlighting, this meta-analysis reveals that when high-quality highlighting is provided, on average, both college and K-12 students benefit greatly, with effect sizes of 0.41 and 0.48, respectively (Hypothesis 7). Training on how to use highlighting as a learning strategy showed encouraging results. Despite the small number of studies, it shows that training has a significant and higher than average effect size on learning from text, with an average effect size of 1.02 (Hypothesis 3).

Abstract of the meta-analysis: