Assessment of various GPT Models versus the human text: a quantitative analysis of lexis and cohesion

The aim of the paper is to analyse select lexical parameters and text cohesion of academic texts automatically generated by several OpenAI LLM models (GPT) in order to (1) investigate the quality of GPT output relative to the original text; (2) compare the quality of various GPT models. The material used in the study comprised a fragment of a research article. The method of analysis involved NLP-based text analysis tools that focus on the examination of various lexical and text cohesion parameters. The study shows that the similarity between AI-generated texts and the text written by the human author is very high and that there is no single model which would achieve the highest values for all linguistic indices.