Annotation of segment boundaries in child speech presents a persistent challenge, particularly with large-scale datasets. While auto-segmentation methods have been developed for adult speech, little attention has been paid to evaluating their performance on child speech. This study evaluates factors contributing to performance of automatic segmentation by analyzing the annotations of two studies into Australian-English-speaking children. The first study assesses human-human reliability in 3- and 12-year-olds. The second compares manual and the Montreal Forced Aligner (MFA) segmentation in 4- to 11-year-olds. The results indicate that the MFA falls short of the human annotator, though discrepancies decrease as children grow older. Systematic discrepancies between the human annotator and the MFA suggest different criteria for placing segment boundaries. These results shed light on how to incorporate adult-based automatic aligners into semi-automatic (and thus: still partially manual) acoustic analysis of child speech.