LMU Study Finds AI-Generated News Articles Less Understandable

News articles generated by artificial intelligence (AI) are less understandable to readers than those written by human journalists, according to a new study from Ludwig-Maximilians-Universität München (LMU). The research highlights that perceived difficulties in understanding AI-generated content stem primarily from how numbers are used and the specific word choices made, even when the AI output is edited by human journalists.

The study, published in Journalism: Theory, Practice, and Criticism, surveyed over three thousand online news consumers in the UK. Participants were asked to evaluate texts, half of which were automatically generated and the other half written by human journalists. The findings indicated that readers consistently found the automated articles significantly less comprehensible.

Respondents reported that AI-generated articles contained too much inappropriate, complicated, or unusual language. They also judged the handling of numbers and data in automated articles to be significantly poorer compared to human-written pieces. However, readers expressed equal satisfaction with the overall character, narrative structure, and flow of both automated and manually written articles.

Lead author Neil Thurman suggests that journalists and computer scientists should focus on reducing the quantity of numbers, better explaining potentially unfamiliar words, and increasing language that helps readers grasp the core of a story. This research is noted as the first to examine the relative understandability of both manually and automatically produced news articles and to pinpoint the reasons behind the perceived differences.