Language changes are among the earliest indicators of cognitive decline in aging. Perplexity, a linguistic measure derived from information theory that quantifies speech predictability, has emerged as a potential marker for detecting early cognitive changes. However, its underlying neural substrates remain unclear. This study investigated the structural brain correlates of perplexity in 38 elderly participants (26 cognitively healthy, 12 with mild cognitive impairment) using magnetic resonance imaging (MRI). Perplexity was computed automatically from autobiographical interviews using single-word (1-g) and word-pair (2-g) models. Voxel-based morphometry analyses, adjusted for total intracranial volume, sex, and education, revealed distinct associations between perplexity measures and regional gray matter volume. Region-of-interest analyses confirmed significant positive correlations between 1-g perplexity and left middle temporal gyrus volume as well as between 2-g perplexity and left precuneus. These findings suggest that perplexity reflects both linguistic processing and autobiographical memory, as evidenced by its associations with language-relevant temporal regions and memory-related precuneus. This study provides initial insights into the neural basis of perplexity as a measure that captures both linguistic and content-related aspects of language production in cognitive aging.