Recent advances in artificial intelligence (AI) have revolutionized the creation and interpretation of visual content, increasingly influencing decision support systems in business. In our study, we examine whether AI-generated descriptive metadata—provided alongside images generated by DALL·E 3 from OpenAI—exhibits systematic linguistic disparities across three central social dimensions: attractiveness, gender, and occupational stereotypes. By integrating algorithmic bias and linguistic framing theories, we argue that this metadata offers a window into how the AI “thinks” about images and reveals biases that might not be evident from the images alone. Our findings indicate significant variations in the use of key linguistic elements such as concrete nouns, dynamic verbs, evaluative adjectives, adverbs, and personal pronouns. These insights are critical for ensuring that decision support systems rely on fair and accurate representations.