Knowing that advanced AI devices like ChatGPT nowadays are able to generate generally linguistically correct text products, this paper focuses on ChatGPT language capabilities related to the cohesion of text products generated by it. The paper provides a comparative qualitative and quantitative analysis of grammatical and lexical correctness of ChatGPT text products including spelling and punctuation in three languages: English, German and Polish. The research material consists of 480 ChatGPT answers to a metaphorical question concerning ‘colours of love’ including 100 expressions, 5 different explanations and 1 short story, ‘a verbal picture’, for 10 of them, separately in each of the analysed languages. The choice of the topic was motivated by the intent to offer an insight into how ChatGPT deals linguistically with a task that requires unusual but at the same time contextually appropriate solutions. The aim of the paper is to identify linguistic deficiencies in the analysed ChatGPT text products and to discuss the limits of chain-of-thought (ToC) in terms of the opportunities of improvement its linguistic agility. Section 1 briefly presents ChatGPTs capabilities and draws attention to its limitations. Section 2 deals with the textuality criteria highlighting cohesion as the feature which is essential for the considerations in this paper. Section 3 provides the materials and methods. Section 4 presents the results with the discussion. The results show that the impact of ToC can still be recognized in overused linguistic constructions.