BackgroundPatient portal secure messaging can support age‐friendly dementia care, yet little is known about care partner use of the portal and how message concerns relate to age‐friendly issues.MethodsWe conducted a two‐part observational study. We first assessed the feasibility of automating care partner identification from patient portal messages by developing and testing a natural language processing (NLP) rule‐based classification system from portal messages of 1973 unique patients 65 and older. Second, two independent reviewers manually coded a randomly selected sample of portal messages for 987 persons with dementia to identify the frequency of expressed needs from the 4M domains of an Age‐Friendly Health System (medications, mentation, mobility, and what matters).ResultsA total of 267 (13.53%) of 1973 messages sent from older adults' portal accounts were identified through manual coding as sent by a nonpatient author. The NLP model performance to identify nonpatient authors demonstrated an AUC of 0.90. Most messages sent from the accounts of persons with dementia contained content relevant to the 4Ms (60%, 601/987), with the breakdown as follows: medications—36% (357/987), mobility—10% (101/987), mentation—16% (153/987), and what matters (aligning care with specific health goals and care preferences)—21%, 207/987.ConclusionsPatient portal messaging offers an avenue to identify care partners and meet the informational needs of persons with dementia and their care partners.