This qualitative research aims to explore alternatives to multiple-choice online reading comprehension tests, which may have high construct validity, but which also have low ecological validity and negative backwash effects at the system level. Our aim is to investigate an approach that might capture and evaluate some of the complex cognitive processes that are involved in authentic web-based research tasks. Eight fluent readers participated and for each participant, files were generated based on all search terms used, all visited URLs, and the text of the final essay. Each participant?s evidence is evaluated using Latent Semantic Analysis to produce an indication of five factors: the degree of match between participant?s search goal and the given task, participant?s search goal and an expert?s Internet search goal, participant?s written task output with the given task, participant?s written task output with an expert?s written task output and the overall coherence of written task output.