AbstractBackgroundPerseverations during a verbal fluency task can be immediate or delayed. There is a general agreement that delayed perseveration may occur due to impaired working memory (WM). However, experiments to establish this have been complex and time consuming. It is also not known if perseverations are caused by similar mechanisms of impairment in WM in healthy controls and in the cognitively impaired. On a semantic fluency task, switching between subcategories produces an automatic self‐generated interference. Impairment in WM capacity may lead to perseveration when switchback to a previous subcategory occurs. The number of switches preceding this switchback may also be considered to extend the WM load and result in perseveration. We hypothesize that individuals with WM impairment (aMCI and AD) are likely to perseverate only after few initial switches. In contrast, a healthy control (HC), with a relatively intact WM, may have fewer perseverations for similar number of switches and may be able to sustain more switches quantitatively.ObjectiveTo study how clusters and switches during verbal fluency tasks can help understand WM contributions to perseveration in patients with cognitive impairment as opposed to cognitively normal controls.MethodWe studied three groups, aMCI (n = 30), AD (n = 30) and HC (n = 60) based on their performance on a semantic fluency task along with tests of executive function. Our primary analysis, focussed on number of switches in between subcategories of index word and first perseverative error. Additionally, average word distance between the two was analysed. Finally, relationship between perseveration, switching and switch back was also studied.ResultsForty‐eight percent of perseverative errors occurred after the third switch for individuals with cognitive decline. In contrast, thirty‐five percent of cognitively healthy perseverated after three switches and forty‐three percent after five. Further, perseverations in HC group, were found to be more attributable to longer subcategory lists exceeding the normal word list span, even within the same cluster. Finally, at a given distance between index word and perseveration, cognitively impaired performed hierarchically worse than HC.ConclusionThese findings suggest that switching across subcategories creates self‐generated interference which further amplifies working memory impairment among the cognitively impaired.