Across 400 responses, the clearest pattern is that the models do not behave as if there is one universal "best store." Instead, they map recommendations tightly to the user's social role, aesthetic identity, and practical shopping constraints. A luxury-banker or physician profile is commonly routed to retailers like MR PORTER, Mytheresa, or Breuninger, while a Berlin streetwear student is routed to SSENSE, END. Clothing, or local scene-coded stores such as Overkill, HHV, Asphaltgold, or BSTN. The contrast is explicit in the responses: ChatGPT tells a Frankfurt luxury user, "I’d pick MR PORTER as the default," while Gemini tells a fashion-vintage photographer that "Grailed is the gold standard for archival fashion and vintage." That split captures the study well: classic-professional contexts get polished, trusted multi-brand luxury; creative and hype-driven contexts get curation, drops, resale, and niche discovery.

A second major axis is ethics and transparency. When the user is sustainability-led, vegan, or explicitly anti-fast-fashion, the answer set changes materially. Instead of broad marketplaces, the models favor ASKET, Avocadostore, Patagonia, hessnatur, Armedangels, Greenality, or repair/resale ecosystems. This is especially visible in minimalist and eco-conscious profiles, where words like "full traceability," "impact receipt," "GOTS," "repair," and "circularity" recur. ChatGPT often operationalizes this by giving a workflow such as "Use Vinted as your baseline … and ASKET to lock in your uniform," whereas Gemini more often frames the same choice as part of the person’s worldview and identity.

Model behavior also differs in tone. ChatGPT tends to answer like a pragmatic shopping adviser: it frequently names a "best overall" option, then two alternatives, and often cites return windows, shipping policies, or customs handling. Gemini is more likely to write as a style-matching curator, leaning into locality, scene awareness, and cultural positioning. Importantly, language differences in the panel come from the profiles, not the model: almost all responses are German because most profiles are German-speaking, while the small English-language subset mainly appears for English-speaking expat profiles in Berlin and follows the same underlying recommendation logic.