Most users notice that AI systems differ. The differences are usually described impressionistically: one model feels warmer, another more direct, a third more cautious, a fourth more willing to push back. Capability benchmarks measure little of this variance, because most leading AIs perform comparably on standardized tests. The variance lives in interaction style. We introduce APMI (Adaptive Personality Match Index), a four-dimensional taxonomy for typing both humans and AI systems by interaction style, enabling direct comparison-based matching rather than recommendation-based selection. The framework grew out of earlier work applying Myers-Briggs-style instruments to AI models (KH candor captures four behaviors existing evaluations do not measure — epistemic honesty, willingness to update on evidence, social position-holding under pressure, and product-layer transparency. Matching compares a user's APMI plus preferences against an AI's APMI plus style and candor. We apply the framework to a panel of ten frontier AI systems and report observed clustering, candor variance, run-to-run instability, and stable refusal patterns. The central claim of this paper is direct: *matching users to AI systems by interaction-style fit produces meaningfully different outcomes from selection by capability alone, and the relevant fit is measurable.* The framework produces three categories of match per user — Complement, Comfort, and Stretch — each suited to different user contexts. APMI is not a measure of AI intelligence, a complete personality model, a predictor of output correctness, or a guarantee of stability across model versions. It is a focused taxonomy for describing and matching interaction behavior, intended to be refined through validation against real user outcomes. **Keywords:** AI evaluation, personality typology, human-AI interaction, behavioral testing, candor, AI matching, interaction style
Building similarity graph...
Analyzing shared references across papers
Loading...
K H
D H
Hayashi Eye Hospital
Building similarity graph...
Analyzing shared references across papers
Loading...
H et al. (Wed,) studied this question.
www.synapsesocial.com/papers/69fd7fa1bfa21ec5bbf0824d — DOI: https://doi.org/10.5281/zenodo.20058919
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: