Why highlighting doesn't work

In 2013, John Dunlosky and colleagues published a large review in Psychological Science in the Public Interest rating common study techniques by how well the evidence supported them. Highlighting and underlining - probably the single most common study behaviour there is - came out near the bottom, rated low utility across nearly every condition they tested.

What the review actually found

Highlighting isn't harmful. It just does almost nothing beyond what reading alone would do. Students who highlight while reading remember about the same amount as students who just read, and highlighting can even backfire on material that requires drawing connections across a passage, because it encourages treating each highlighted sentence as an isolated important fact rather than part of a larger idea.

The techniques the same review rated highly - practice testing and distributed practice, the ideas behind the testing effect and spacing - require you to do something with the material afterward. Highlighting requires nothing afterward. It's a marking behaviour performed during reading, and it ends the moment reading ends.

Why it feels useful anyway

Highlighting produces a visible artifact - a page that now looks "worked on" - and it requires you to make small in-the-moment judgments about importance, which feels like engagement. Neither of those things is the same as building a retrievable memory. You can highlight a sentence perfectly and still have no ability to produce its content from memory a day later, because nothing about highlighting asks you to try.

What to do instead of highlighting

If you want a mark to review later, write a question in the margin instead of a highlight on the text - a question you'll have to answer, not a sentence you'll re-read.
Turn the material into cards after the first read, rather than marking it up during the first read. The card-writing step does the work highlighting was pretending to do.
If you're going to mark anything during a first pass, mark what confuses you, not what seems important. Confusion is something you can act on later. "This seemed key" mostly is not.