Quantifying sustained attention in real-world environments remains a persistent challenge in electroencephalography (EEG)-based biosensing, particularly when acquisition is constrained to a single, low-cost biosensor channel. Conventional ratio-based attention indices, namely, the theta/beta ratio (TBR) and the theta/alpha ratio (TAR), typically fail to register brief or low-amplitude attentional shifts, and this insensitivity is exacerbated by the unfavorable signal-to-noise conditions that accompany wearable recordings. To overcome these limitations, in the present work, we formalize the gamma progressive index (GPI), an EEG-derived indicator that emphasizes the higher-frequency end of the gamma band through a normalized, monotonically increasing linear weighting function. By concentrating the weighting on those gamma-band frequencies most closely tied to focused cognitive effort and by subsequently normalizing the weighted sum against each participant’s own baseline, we obtain the GPI that yields a stable and computationally economical measure that can be evaluated in real time on the onboard processor of a portable single-channel EEG headband. EEG recordings were obtained from three adults as they interacted with generative AI chatbots in a naturalistic setting. During these interactions, the GPI tracked increases in cognitive engagement more reliably than did the TBR or TAR, exhibiting a wider dynamic range and sharper temporal contrast. In head-to-head comparisons, the GPI outperformed both the TBR and the TAR in detecting short-duration attentional fluctuations and in preserving signal integrity in noisy recordings typical of wearable EEG systems. The proposed GPI therefore offers a robust, scalable, and computationally lightweight pathway for the ecologically valid monitoring of attention in everyday educational contexts. To the best of our knowledge, no prior study has combined a progressively weighted gamma-band spectral estimate, a within-subject baseline normalization, and an explicit on-chip single-channel implementation in a single mathematically specified metric; the present work supplies this combination as the journal extension of our earlier conference report.