A body of text may be compared with one or more user-selected text portions to rank a plurality of text portions of the body of text, such as for predicting which of the text portions are likely to be annotated by users. As one example, the text of a content item may be compared with excerpts of other content items that have been highlighted or otherwise annotated by a plurality of users. Based at least in part on the comparison, some implementations identify one or more portions of text of the content item that are likely to be selected or highlighted by users that access the content item. In some examples, a classifier may be trained based on popular highlights determined for a plurality of content items. The classifier may be applied to a body of text to determine portions that users are likely to consider profound or interesting.