Metacognitive reasoning is central to decision-making. For every decision, we can also judge our trust in that decision, or our level of "confidence." The mechanisms and representations underlying reasoning about confidence remain debated. We test whether children rely on "processing fluency" to infer their own confidence: do decisions that come quickly and easily lead to high confidence, while decisions that are slow and effortful result in low confidence? Using children's verbal disfluency--fillers (e.g., "umm," "uhh"), hedges (e.g., "I think," "maybe"), and pauses in speech--as an observable index of processing fluency, we assess whether children's reports of confidence are a read-out of their verbal disfluency. Five-to-eight-year-olds answered semantic questions about animals and performed perceptual comparisons, then reported their confidence in their answers in a two-alternative forced-choice confidence judgment task. Verbal disfluency predicted both answer accuracy and children's reports of confidence: children produced more fillers, more hedges, and longer speech onsets during incorrect trials and during low confidence trials. But we also found a dissociation between fluency and confidence. When examining trials where accuracy and confidence diverge (i.e., correct but low confidence or incorrect but high confidence trials), we observe no reliable relationship between confidence and fillers and hedges, and children take "longer" to begin answering on high confidence trials. We conclude that--in 5-8-year-old-children--fluency is a reliable tracker of "accuracy" but not confidence, and that fluency is only predictive of metacognitive judgments in children when confidence and accuracy are aligned.