Can AI quantify the flexible self? Modeling relationships between social media behavior and self-concept

We examine whether social media posts contain information about how people understand and evaluate themselves. A six-month study combines three repeated questionnaires with donated Instagram archives from 48 emerging adults across 57 accounts. We study self-concept clarity, self-criticism, and dialectical self-views through two approaches: representational analyses that predict questionnaire scores from numerical descriptions of posts, and agentic analyses in which language-and-vision models judge the posted material. Preliminary evidence is clearest for self-concept clarity: three direct judges recover participant rankings with correlations of .51--.57, and content descriptors support prediction in a larger sample. In a small paired subset, finsta ('fake instagram') posts elicit lower clarity and higher self-criticism estimates than rinsta ('real instagram') posts. These results motivate further study of contextual self-presentation while exposing open methodological questions. Models often decline to estimate when evidence seems insufficient, sharply reducing comparable samples; ranking participants also proves easier than estimating their absolute scores. The findings support an exploratory evaluation framework, rather than validated individual assessments.