Author keywords attached to academic papers are often used in intellectual structure analysis. However, the length and selection criteria for keywords vary across publications and, even some publishers do not require keywords for their articles. To explore the opportunity to overcome such keyword inconsistency issues, this study compared author keywords from papers focused on the digital divide with those extracted using the language models, BERT and ChatGPT. Preliminary findings reveal structural variations across the keyword networks and suggest a potential need to revisit keyword-based research. Future research will expand the scope of the dataset and conduct an in-depth analysis of keyword patterns across the language models.