Introduction:
Long COVID (LC) is one of the major challenges of modern medicine. The lack of standardized diagnostic criteria, nonspecific symptoms, their temporal variability and chronic nature significantly hinder diagnosis and classification. This creates the need for advanced technological tools to better understand the pathophysiology and identify individual disease phenotypes. Artificial intelligence, particularly machine learning (ML), is a promising approach. This review synthesises current evidence on ML applications in LC phenotyping and outlines core ML concepts relevant to this research context.

Material and methods:
A search of the PubMed database identified 47 articles published through July 9, 2024. After manual screening of titles and abstracts, 17 studies published in English were included in the final analysis.

Results:
We present an overview of ML applications in LC research, focusing on the identification of phenotypes, subphenotypes and risk factors associated with LC occurrence and severity. The identified phenotypes depict LC as a multisystem condition, in which symptom clusters often involve multiple organ systems rather than a single organ.

Conclusions:
ML represents a useful tool for identifying LC patients and classifying them into distinct subphenotypes characterized by different clinical manifestations. This approach improves the precision of diagnostic pathways and supports individualized therapeutic strategies. ML approaches may also improve clinical trial design by facilitating analysis of large, heterogeneous datasets – a particular advantage given LC's variable clinical course.