Ask for a white t-shirt and you get Uniqlo

You asked for a white t-shirt and the assistant named Uniqlo. You never asked for Uniqlo, and there may be no Uniqlo shop where you live.

26 September 2026

If you sell apparel, a vague “what should I buy” question is why a global default beats a better shirt on your shelf.

This is the shopping version of the same pattern as “best things to do near me” for places. How AI chatbots build a shortlist applies to products too: the assistant still has to pick a few options and describe them.

What Burke counted

In September 2026, Deana Burke published We’re all wearing Uniqlo to the Singularity, a fashion probe from the same research line as her Morrowen deodorant test and the ongoing Agent Shelf panel at regularbenchmarks.com.

She hit the APIs for seven models (three American, four Chinese; the piece does not name them), with no customer preferences in the prompt. The white-t-shirt line was roughly: “I need to buy the perfect white t-shirt, what do you recommend?” About 70 answers in total.

Across that run she reported Uniqlo on 94% of white-t-shirt answers, La Sportiva on 100% of approach-shoe answers, and Samurai Jeans on 96% of Japanese-denim answers.

Treat that as her snapshot, not a trend line. The count is pooled across models, on the order of ten trials per model, and the public piece does not show a per-model table. On Agent Shelf, Burke labels a 10-trial cell a snapshot rather than a trend, and she stopped publishing pooled rates after one unfinished model run warped an average. Make AI Visible has not rerun this probe.

Blank ask versus specific ask

Without your context, the answer is for someone else is about the gap the assistant fills with a default person. On a blank product ask, what that person often gets is whoever already dominates “best X” listicles and review roundups — the same corroboration pull that keeps famous landmarks on travel lists.

Blank category ask

“Perfect white t-shirt, what do you recommend?” — Uniqlo on most answers in Burke’s seven-model snapshot.

Specific constraint

Heavy cotton, a fit, a budget, and a shop you can visit this week — the kind of detail that lets a niche brand match.

Those are not two ways to read the same question. On Shopify’s Q2 2026 earnings call, president Harley Finkelstein said 75% of AI-attributed orders in the quarter came from outside Shopify’s top 100 categories, with agents querying Shopify’s catalog against richer constraints than a keyword search — specific needs, not “best X.”

That long-tail story does not cancel Uniqlo on a plain t-shirt prompt. It shows that niche discovery and incumbent defaults can both be true, depending on how much the ask leaves open.

Burke’s own deodorant experiment is the flip on one model: a narrow symptom ask can name a newcomer; “best deodorant” keeps the incumbent and never names the fake brand.

Limits worth keeping in mind

Burke turned web search on for the same t-shirt question and still got Uniqlo, after links to GQ and Vogue. She called that test tiny. If retrieval only surfaces pages that already agree, opening the web confirms corroboration rather than widening the pool.

Researchers from the University of Washington, Carnegie Mellon, and Stanford reported an Artificial Hivemind effect in open-ended generation (NeurIPS 2025 paper): many models converge on similar answers, and repeating the question or switching model often does not buy much diversity. That work used real-world open-ended prompts, not shopping lists, and English-only queries. It still explains why “try another chatbot” is not a reliable fix for a category default.

Agent Shelf is the place to watch whether those product names drift month to month, with a published methodology and per-model tables. Adobe Analytics data reported by Reuters in June 2026 found US retail visits referred from large language models generated more revenue per visit than other sources in May 2026. That is why the default brand on a blank ask has commercial weight, not proof that readable pages alone get you named.