In this study, a formal and textual stylistic approach was employed to investigate specific prompts in journalistic and legal genres that were provided to Generative Pre-trained Transformer-4 (ChatGPT-4) for generating prompts. ChatGPT-4 generates three prompts: minimal, tone-specified, and constraint-based. Each one of these prompts includes a prompt, an output, and a stylistic analysis. The purpose is to determine how prompt design influences stylistic variation, evaluate the AI's capacity for self-analysis, and highlight the importance of prompt engineering. Leech and Short's (2007) stylistic checklist of lexical, grammatical categories, and cohesion has been used as a theoretical framework for analysing the AI outputs. In the process of data analysis, a qualitative research method was used to evaluate ChatGPT-4’s ability to generate content within the two selected genres to perform a stylistic analysis, and evaluate its outputs against recognised standards of the stylisticians’ toolkit (Leech and Short 2007, Leech etal 1989). In the analysis, the following linguistic features are used: lexical features (nouns: abstract and concrete, verbs, adjectives), grammatical features: sentence (complexity, length, elements, types, structure), tense, and voice; Textual features: punctuation, specialised vocabulary, emotive lexis, word choice, and tone. The findings show that the legal genre had longer and more syntactically complex sentences, more lexical abstraction, a higher frequency of present-tense verbs, specialised vocabulary, and formal word choice. The journalistic genre, on the other hand, is characterised by a greater use of pasttense verbs, shorter syntactic structures, and a reportorial tone. Although declarative sentences and an active voice were used in both genres, legal prompts were distinct in that they contain a greater variety of authoritative tones and foregrounding punctuation. While AI is capable of identifying surface-level features, prioritising tone and genre over grammatical detail, it was less accurate and inadequate. It disregards detailed linguistic categorisations by methodically analysing lexical choice, syntactic complexity, and textual cohesion; human analysis demonstrates superior accuracy. Ultimately, ChatGPT-4 often replicates the style of stylistic critique, whereas stylisticians validate their conclusions through explicit linguistic evidence, particularly in abstract domains where linguistic validity predominates.