This study examines how prompt-driven expansion affects the content and source-groundedness of AI-generated translations. It introduces “augmented translation”, translation prompted to exceed the source text in length, and examines changes in textual features and hallucinations across augmentation levels. The study analyzes 100 English tourism paragraphs from TripAdvisor and their Persian ChatGPT translations with 20%, 50%, and 100% augmentation. Using qualitative thematic and descriptive analysis, the study identifies five patterns of change: lexical and stylistic amplification, content alteration, pragmatic and relational shift, omission and reduction, and structural and syntactic shift. Lexical and stylistic amplification was most frequent, especially through adjective addition and persuasive enhancement. Counterintuitively, 20% augmentation showed the most content deletion, while 50% favored adjective-based elaboration and 100% increasingly produced unsupported claims and persuasive enhancement. Hallucinations also shifted across augmentation levels: lower augmentation produced calculation-related errors, whereas higher augmentation increasingly generated unsupported claims. Hallucinations also tended to emerge toward the later portions of augmented translations, suggesting that pressure to sustain additional content may push generation beyond source-grounded information. The findings highlight prompt design as a consequential factor in AI-assisted translation and reveal a tension between elaboration and informational reliability.