Follow-up of an earlier study investigating the link between cognitive impairment and long-term speech recognition in noise. Sentence recognition in noise (MBAA2 test, SNR50); cognition (Montreal Cognitive Assessment MoCA, Victoria Stroop, and ECLA 16 + phonological awareness subtests 3 and 4); subjective benefit (Glasgow Benefit Inventory, GBI); and approaches to speech therapy were assessed three years post implantation in 26 adult subjects. Average MoCA scores were below population norms, with seven subjects scoring below 22: the normal threshold set for the study. Average Stroop and ECLA scores were within the normal range. Poor MoCA performance and poor verbal fluency were linked to poorer sentence-in-noise scores. Subjects with MoCA scores < 22 or verbal fluency scores < 6 were fifteen times more likely to have long-term SNR50 > 5 dB. GBI scores were significantly correlated with SNR50s. Cognitive scores remained stable over time, except for completion times for ECLA 3 and 4, which improved significantly. Speech therapy with cognitive goals showed better outcomes in a small subgroup. The speed of completing the reading tests was improved after prolonged implant use which may indicate improved processing speed, better lexical mapping, and improved working memory. An improved decision tree is proposed to guide rehabilitation in underperformers. The results confirmed that after three years of implant use, top-down cognitive deficits still had an impact on sentence recognition in noise, with poorer scores related to lower patient reported benefit. There may be some speech recognition benefits to targeting speech therapy towards objectives in both perceptual and cognitive domains, but more data is required.