Abstract Background The need persists for sensitive, low‐cost, and high‐access cognitive markers to complement biomarker information in the earliest stages of Alzheimer’s disease. Sensitive cognitive markers may be derived from information at the item level of neuropsychological tests. We investigated if item‐level metrics add information beyond total score in the relationship of semantic fluency—generating as many words as possible of a category in a given time frame—to cortical thinning across eight years in middle‐aged to older adults without dementia. Method Participants were part of the SMART‐MR cohort (n = 251, age = 59.6±8.6, 19.5% women), designed to investigate brain changes in individuals with arterial disease, which puts them at high risk for dementia. Global cortical thickness was derived from MRI scans at two visits approximately eight years apart. Item‐level metrics were derived from linguistic databases: mean log lexical frequency (how often a word occurs in our language) and mean age of acquisition (AoA; at what age a word was learned) of words generated on a 2‐minute semantic (animal) fluency task (Fig 1). We calculated residualized item‐level scores representing the deviation in the item‐level metric from what is expected based on the total score to account for collinearity. Linear regression models—adjusted for age, sex/gender, education, and history of stroke—analyzed the relationship between total score or item‐level metrics as determinants and cortical thickness at follow‐up as the outcome, adjusting for baseline level. Result Total score alone was not related to cortical thinning at follow‐up (β = ‐.023[‐.001;.000], p =.466). However, a higher‐than‐expected mean lexical frequency (β = ‐.065[‐.054;‐.002], p =.033) and lower‐than‐expected mean AoA (β =.066[.001;.019], p =.033) were related to more cortical thinning at follow‐up. An alternative analytic approach to capture change, using the difference in cortical thickness between visits without baseline adjustment, yielded similar results. Conclusion Alternative item‐level metrics of semantic fluency, as opposed to the traditional total score, related to future neurodegeneration in high‐risk individuals without a dementia diagnosis. The results seem robust across two different item‐level metrics and two different analytic approaches. Valuable item‐level information could be leveraged for accurate identification of the earliest stages of clinical Alzheimer’s disease, in addition to biomarker evidence, for intervention and timely diagnosis.