AI recommends a K-beauty that Korea doesn't buy.

Ask an AI what K-beauty to buy and it recommends a different country's chart. Counting brand mentions in the answers two AI engines gave to a fixed set of K-beauty questions on 2026-07-27, COSRX came back most — 124 mentions — followed by Anua (92), Dr.Jart+ (72) and Etude (68). But line that leaderboard up against what Olive Young Korea actually sold the same week and the two barely touch: four of AI's top five brands aren't on Korea's top 100, and the brand Korea buys most doesn't appear in the AI answers at all.

Which brands does AI name most?

The internet-famous ones. The most-cited are COSRX (124), Anua (92), Dr.Jart+ (72), Etude (68) and Purito (42) — brands with deep English-language review cultures, the names that recur in Reddit routines and YouTube explainers. This is a small measurement, two engines answering ten questions, so read it as a sample of what AI reaches for, not a census. But the shape is clear: AI's K-beauty canon is the one that travels through English-language content.

Do those brands actually sell in Korea?

Mostly not on this week's chart. COSRX, AI's #1, holds zero of Olive Young Korea's top 100 slots this week; Dr.Jart+, Etude and Purito are absent too. The brands that appear on both sides are the exceptions — Anua (AI #2, Korea #21), Torriden (AI #7, Korea #10) and SKIN1004 (AI #6, Korea #41) — names that are both talked about online and bought at home. For most of the AI leaderboard, though, mention volume and shelf position are unrelated: being recommended by an AI is not the same as topping the chart it was trained to describe.

What does Korea buy that AI ignores?

Its volume brands. MEDIHEAL — Korea's #1 this week, with 8 of the top 100 slots — doesn't appear on the AI leaderboard at all, and neither do CLIO, Abib or numbuzin, all high on Korea's chart. These are the mask-and-basics brands from earlier in this desk's reporting: bought constantly, packaged in sets, rarely the subject of an English-language routine video. AI doesn't recommend them because the content it reads doesn't discuss them — not because Koreans aren't buying them.

What this does and doesn't tell you.

It compares mentions to shelf position, and both are proxies. The AI figure counts brand mentions in two engines' answers to ten fixed questions, matched against our chart-brand dictionary — a directional sample, not a full audit, and it will move run to run as we archive each measurement. AI mentions are not sales, and one retailer's chart is not all of Korea. What survives those limits is the gap itself: the brands AI surfaces when asked about K-beauty are, for the most part, a different set from the brands Korea is actually buying — which is exactly the distance a chart is built to measure, and the reason "an AI recommended it" and "it's a best-seller" are two different claims.