A new paper on arXiv examines the cognitive cost of reading text produced by large language models. While prior work has focused on linguistic differences between human and machine writing, the authors say the implications for readers' mental effort remain underexplored. They propose reading time as a direct measure of that effort.

The study is described as a reading-time comparison between human-written and LLM-generated texts. The abstract, however, is cut off before presenting any results, so the direction or size of the effect is not yet available. The paper's contribution at this stage appears to be framing reading time as a useful metric for evaluating LLM output.

Because only the abstract was available, this digest cannot confirm the study's conclusions. The key point is the authors' emphasis on cognitive load as a separate dimension from style or grammar—one that may matter for how readers actually process machine-generated content.