Activated carbon (AC) is widely used for organic micro-pollutants (OMPs) removal, yet adsorbability evaluation remains challenging due to molecular diversity and adsorbent heterogeneity, especially given the limitations of traditional assessment metrics such as hydrophobicity (logD). This study proposed a machine learning (ML)-driven assessment strategy by aligning the adsorbability of various AC adsorbents with a hypothetical “Standard AC” to evaluate the adsorbabilities across 56 OMPs. XGBoost, RF, and ET models achieved high prediction accuracy on the test set (R² = 0.88–0.98, RMSE = 0.17–0.38, MAE = 0.13–0.27), and were further validated against a published experimental dataset. Interpretable ML analysis identified a logD threshold of ≈ 2, at which the dominant adsorption mechanisms transitioned from hydrophobic interactions for OMPs with higher hydrophobicity to π-π interactions, hydrogen bonding, and pore-filling for those with lower hydrophobicity. Adsorbability increased with molecular weight, as flexible molecules (rotatable bond ratio > 0.012) overcame steric hindrance in micropores, enhancing pore-filling efficiency through improved accessibility. By introducing a standardized, data-driven adsorbability reference and elucidating the intrinsic interplay between molecular properties and adsorption mechanisms, this study offers a robust framework for knowledge-informed treatability evaluation and a practical benchmark to guide adsorption process design.