Each numbered study measures one piece of it: how models see products, which brands they prefer, and where they send the buyer.
Dated and sourced. Built from live shopping conversations, model outputs, and routing evidence.
Together they map the systems now sitting between shoppers and stores: what breaks, what gets favored, and what brands can actually control.
Benchmarks fashion-intelligence behavior across CLIP, SigLIP, and DINOv2: uncertainty, impostor detection, and collection cohesion. Also published as “A Vision Benchmark for Fashion AI”.
Maps how nine leading language models converge on the same designers, garments, and fashion narratives when taste is inferred from prompts alone. Also published as “What LLMs Repeatedly Choose”.
Follows ChatGPT and Gemini recommendations into the routing layer: official product pages, retailer leaks, and missing purchase paths.
30 ChatGPT shopping answers across ten jewelry buying intents. Brands were named 248 times; only 88 mentions (35.5%) linked the buyer to the brand’s own store. Who owns the path, who leaks it, and which intents are still wide open.
The recommendation systems hidden inside current models: the assumptions they make, the tastes they overfit to, and the sales they now influence.
Whether it distinguishes brand, season, and design family from the image itself.
The objects and creators that recur across models when taste is generated from text.
How likely a brand is to appear in AI-mediated shopping answers.
Where the model hallucinates authority instead of admitting uncertainty.
The latest study tracks what happens after AI names a brand: the buyer lands on the brand’s own page, leaks to a third-party seller, or gets no usable path at all.
