Monthly report · Format preview
COSRX
If you read one thing
- COSRX was named in 2 of the 6 buying questions that never mentioned the brand. Beauty of Joseon was named in all six, and SKIN1004 in four.
- Your own site was cited zero times on those six questions. Three competitors’ own pages were cited, and in two of those three the brand owning the page was also named. That is the clearest lever in this report, and the exception matters: Biodance was cited without being named, so a citation raises the odds rather than guaranteeing the outcome.
- Three of the ten questions returned no citations at all, and COSRX won all three. The model already thinks well of the brand. The gap is not reputation; it is absence from the questions where a customer has not heard of you yet.
Where you stand
10 questions were asked across 1 assistants, each in its own separate session with memory off. Four of the ten contained the word “COSRX”, so being mentioned back proves nothing. The six that did not are the ones that decide whether a new customer finds you, and they are what the headline figure counts.
2 of 6
Named in 2 of the 6 discovery questions. first reading, no comparison yet
Is that a real change? Unknown, and it would be dishonest to say otherwise. This is the first reading, so there is no previous month to compare against and no way yet to tell signal from variance. Treat every number here as a baseline rather than a score.
Repeating the same question in a fresh session does not always return the same brands. On a set of this size we treat a move of 1 questions or fewer as noise and say so rather than dressing it up as progress. Sustained direction over three months is the signal worth acting on.
- Across every question
- 6 of 10, the flattering number, and not the useful one
- Strongest assistant
- Only one assistant was measured in this reading, so there is no comparison
- Weakest assistant
- Not applicable to a single-assistant reading
- Category leader
- Beauty of Joseon, named in 6 of 6
- Gap to leader
- 4 questions
Question by question
Every question asked, and whether COSRX was named in the answer. The two groups are kept apart because a question that already contains your brand name proves nothing when the answer repeats it back.
| Question | ChatGPT | Sources |
|---|---|---|
| The six discovery questions: these decide the headline | ||
| Q1 · best korean skincare brand for oily acne-prone skin | Named | 2 |
| Q2 · best korean moisturiser for dry skin | Absent | 1 |
| Q3 · best korean sunscreen for daily use in humid weather | Absent | 2 |
| Q4 · which korean skincare brands are good for sensitive skin | Absent | 3 |
| Q7 · best korean skincare under $20 that actually works | Named | 3 |
| Q8 · fragrance free korean skincare for reactive skin | Absent | 2 |
| The four questions that already contained “COSRX”: every brand wins these | ||
| Q5 · cosrx or beauty of joseon, which is better for acne | Named | none |
| Q6 · is anua heartleaf toner better than cosrx snail mucin essence | Named | none |
| Q9 · is cosrx a good skincare brand | Named | none |
| Q10 · what are the best cosrx products to start with | Named | 1 |
The pattern worth seeing: COSRX is named on every question in the lower group and on two of six in the upper one. Being known is not the problem. Being found is.
Who gets recommended instead
Every brand named across the discovery questions, ranked, with movement since July.
Named once each: ETUDE SoonJung, Dr.G, ILLIYOON, AXIS-Y, Haruharu Wonder.
What changed, and why
Movement since July, and what is behind it where it can be established. Where it cannot, this section says so.
This is the first reading, so there is nothing to compare it against yet
Every figure in this report is a baseline. The value of month one is not the number. It is that from next month there is something to measure against, using the same questions asked the same way.
What we will be watching in month two: whether COSRX appears on Q2 (dry skin) or Q4 (sensitive skin), and whether cosrx.com starts being cited on any question that does not already contain the word COSRX. Right now neither happens.
One thing already worth acting on, before any trend exists
Three of the six discovery answers cited a brand’s own website: pages.anua.com, biodance.com and beautyofjoseon.com. In two of those three the brand that owned the cited page was also named in the answer. biodance.com was cited on the under-$20 question and Biodance was not among the brands named in it, so the pattern is strong but not a rule.
cosrx.com was cited exactly once across all ten questions, on Q10, which already contained the word COSRX. Your own pages are not being retrieved for the questions where a new customer is deciding. That is a content and structure problem, not a product problem, and it does not require a trend to justify acting on.
What the answers are built on
Every domain cited across the discovery questions this month, how often, and whether it moved. These are the pages deciding the answer.
| Source | Cited | vs July | Type |
|---|---|---|---|
allure.com | 2 | first | Editorial |
vogue.com | 2 | first | Editorial |
glow.com.au | 2 | first | Editorial |
reddit.com | 2 | first | Forum |
harpersbazaar.com.sg | 1 | first | Editorial, Singapore edition |
k-beautyskin.com | 1 | first | Niche |
pages.anua.com | 1 | first | Competitor’s own site |
biodance.com | 1 | first | Competitor’s own site |
beautyofjoseon.com | 1 | first | Competitor’s own site |
cosrx.com | 0 | first | Your site, not cited on any discovery question |
A Singapore publication is deciding a Singapore answer
The sunscreen question (Q3) cited harpersbazaar.com.sg, and the answer said in as many words that “Singapore beauty editors specifically highlight it”. This is a market-specific answer being decided by a market-specific publication.
That is unusually reachable. Local editorial is a smaller, cheaper target than international titles, and nobody in this category appears to be competing for it deliberately.
Editorial rankings are doing the deciding, not product claims
On Q2 the assistant justified its winner by pointing at a third-party ranking: “Allure also currently ranks it as…”. The citation is doing the work. A better product description on your own site would not have changed that answer; being in that ranking would have.
Four of the nine cited domains are editorial. Three are competitors’ own sites. One is a forum and one a niche blog. None are retailers, which matters: the retailer product cards that appeared (Stylevana, yesstyle.com, zalora.sg) were surfaced as places to buy, not as reasons to recommend.
What we cannot move
The questions where publishing will not help, and why. This is here every month.
Three of the ten questions returned no citations at all
Q5 and Q6 (both head-to-head comparisons) and Q9 (brand reputation) came back with zero sources cited. The assistant answered from what the model already holds, not from anything it went and read.
Publishing will not move those in the short term, and anyone who tells you otherwise is guessing. They can move when a model is retrained, which is outside anyone’s control and not on a schedule we can promise.
The encouraging half: COSRX won all three of those. “If your main goal is acne, I’d generally pick COSRX over Beauty of Joseon.” The model already holds a good opinion of the brand. The problem is not reputation. It is that COSRX is absent from the questions where a customer has not yet heard of you. Those are the addressable ones, and every action below targets one.
Competitor teardown
The three brands beating you most often, and specifically what is getting them named. Reported as fact from the answers; we do not evaluate their products.
Beauty of Joseon, named in 6 of 6
Where they win: every discovery question, including both of the two where COSRX also appears. On Q3 they took first and fourth place with two separate products.
What is carrying them: beautyofjoseon.com was cited directly on Q8, and harpersbazaar.com.sg on Q3. They are the only brand in this set appearing both through their own pages and through local editorial.
What is reachable: the local editorial route. Their own-site citation is harder to displace, but the Singapore publication route is open to anyone who writes for it.
SKIN1004, named in 4 of 6
Where they win: Q1, Q3, Q4 and Q7, including two questions where COSRX is also named, so they are direct competition rather than a different segment.
What is carrying them: no own-site citation at all. They appear through third-party editorial and forum sources only, which means their position rests on other people writing about them.
What is reachable: this is the most contestable position on the board. A brand ranked highly with no owned-page citation is being carried entirely by coverage, and coverage moves.
Anua, named in 3 of 6, and first on Q1
Where they win: takes “Best overall” on Q1, the single highest-intent question in the set, and one where COSRX places second.
What is carrying them: pages.anua.com was cited directly. That is a dedicated landing page, not their homepage, and it was retrieved for a category question that never mentioned Anua by name.
What is reachable: this is the clearest template in the whole report. A purpose-built, retrievable page for a specific use case got a brand named first on the question that matters most. That is a repeatable structure, not a fluke.
What to do
Seven actions, ranked by expected value. Each names the questions it targets and the evidence behind it. Carried-over actions are marked, with what happened.
Build a barrier-repair moisturiser page that can be cited
Q2 named AESTURA, ILLIYOON, Dr. Althea, Beauty of Joseon and Anua, and not COSRX. The answer justified its pick by pointing at a third-party ranking. COSRX has products that belong in that answer but no retrievable page framing them for the barrier-repair use case in the language the question actually uses.
Targets Q2 · evidence: allure.com cited, COSRX absent, no cosrx.com citation
Own the humid-climate sunscreen question locally
Beauty of Joseon took first and fourth on Q3, supported by the Singapore edition of Harper’s Bazaar. This is a market-specific answer decided by a market-specific publication. A page written for humid-climate daily wear, plus a push for local editorial coverage, targets the one question where geography is doing the deciding.
Targets Q3 · evidence: harpersbazaar.com.sg cited, COSRX absent
Make the sensitive-skin range legible as a range
Q4 returned eight brands, each with a named product line: “Madagascar Centella line”, “Atobarrier 365 line”, “R.E.D Blemish line”. The assistant is reasoning in lines, not individual products. COSRX was not named. Grouping the relevant SKUs into a clearly named, clearly described line is the structural change that makes a brand eligible for this answer.
Targets Q4 · evidence: 8 competitors named by product line, COSRX absent
Copy Anua’s page structure, not their content
pages.anua.com is a dedicated landing page, not a homepage or a product page, and it was retrieved for a category question that never named Anua. Two of the three own-site citations in this reading follow the same pattern. The structural lesson is that a focused page for one use case outperforms a general brand page for retrieval.
Targets Q1, Q2, Q4 · evidence: 3 of 6 answers cited a competitor’s own page
Write for the fragrance-free and reactive-skin vocabulary
Q8 named Pyunkang Yul, PURITO Seoul (three separate products), Beauty of Joseon and Haruharu Wonder. COSRX was absent despite having qualifying products. The cited sources were beautyofjoseon.com and k-beautyskin.com: one owned page and one niche site, neither of them a major title. This is a low-competition question.
Targets Q8 · evidence: niche sources cited, no major title involved
Defend Q7, which is the one you are winning
On the under-$20 question COSRX was named first, with the Advanced Snail 96 Mucin Power Essence at $12.24. This is the only discovery question where COSRX leads. Value positioning is working; the risk is that it becomes the only thing the brand is known for.
Targets Q7 · evidence: COSRX named first, biodance.com and vogue.com cited
Do not spend anything on the three questions with no citations
Q5, Q6 and Q9 returned no sources. COSRX won all three. There is nothing to optimise and nothing at risk. Recording this is as useful as recording an opportunity, because it stops budget going somewhere it cannot work.
Targets nothing, deliberately · evidence: zero citations on Q5, Q6, Q9
Page briefs
Two written this month. Each is specific enough to hand to whoever writes your site copy without a further conversation.
Barrier repair for damaged and dehydrated skin
Suggested: cosrx.com/pages/barrier-repair, a landing page, not a product page
- Target questions
- Q2 “best korean moisturiser for dry skin”, and partially Q4
- Job of the page
- State plainly which COSRX products are for a compromised moisture barrier, and why, in the words the question uses: dry”, “barrier”, “moisturiser”. Not “hydration system” or other internal vocabulary.
- Must contain
- A named line with 3–5 products in it. The key ingredient and its concentration. A plain sentence on who it is for and who it is not for. At least one comparison table, because the assistant reasons in comparisons.
- Must not do
- Lead with brand history. Use invented terminology. Make a claim you cannot support. Assistants increasingly surface caveats, and an unsupportable claim invites one.
- Length and shape
- 600–900 words. H2 per use case. One table.
- How we will know
- Q2 would have to name COSRX, or
cosrx.comwould have to appear as a citation on it. Fair to look after two months; retrieval is not instant.
Daily sunscreen in heat and humidity
Suggested: cosrx.com/pages/sunscreen-humid-climate
- Target questions
- Q3 “best korean sunscreen for daily use in humid weather”
- Job of the page
- Make COSRX eligible for a question currently decided by a Singapore publication. Written for the climate, not the ingredient list.
- Must contain
- Explicit reference to humid and tropical conditions. Finish and texture under sweat. Reapplication over makeup. SPF and PA rating stated as numbers.
- Must not do
- Claim a sweat-resistance standard the product has not been tested to. This is a regulated claim in several markets.
- Length and shape
- 500–800 words. Lead with the use case, not the product.
- How we will know
- COSRX named on Q3, or a Singapore publication citing the page. The second is the likelier first signal.
Written for you this month
One finished, ready to publish under your own name. Written against the questions above, not to a keyword.
What actually changes a Korean skincare routine for humid weather
Attached to the email this link came with · 1,050 words · publish under your own name
- Targets
- Q3, and supports Q2
- Where to publish
- Your own journal or blog, on your own domain. Not a third-party platform. The citation value belongs on a domain you control.
- Why this one
- Q3 is the question where a local publication is visibly deciding the answer, which makes it the most winnable of the questions COSRX is currently absent from. A useful article on the topic is also the most natural thing to pitch to that publication later.
- What it deliberately does not do
- It is not a product advert. Retrieval favours pages that answer the question, and an article that only sells will not be cited by anyone.
Evidence
Every question, on every assistant, with what came back and what it cited. Full verbatim answers are attached to the email this link came with, so nothing here has to be taken on trust.
Discovery questions: the six that never mention the brand
“If you have oily, acne-prone skin, my top Korean brand pick is Anua—especially if you want lightweight products that focus on calming inflammation, pores, and excess oil without aggressively stripping the skin.” COSRX placed second, described as “Best for targeted acne care”.
The winner was justified by a third-party ranking: “Allure also currently ranks it as…”. The citation is deciding the answer, not the product description.
“Singapore beauty editors specifically highlight it as suitable f[or]…”, a local publication directly deciding a local recommendation.
Every brand was named with a specific product line attached. Retailer cards surfaced: Stylevana, yesstyle.com, zalora.sg.
COSRX Advanced Snail 96 Mucin Power Essence listed first, at $12.24.
Questions that named COSRX in the question, excluded from the headline figure
“If your main goal is acne, I’d generally pick COSRX over Beauty of Joseon.”
“If you can only buy one, I’d generally choose COSRX Snail Mucin because it’s more useful for hydration and barrier support.”
“Yes. COSRX is generally a good skincare brand, especially if you like simple, ingredient-focused Korean skincare.” With the caveat: “COSRX isn’t automatically good for everyone.”
The only question in the entire set where cosrx.com was cited, and it is the one that already contained the brand name.
How this was measured
| Assistants | One assistant, on its standard consumer web app, free tier, Temporary Chat. A Core report covers four assistants; Pro and Global cover seven, each named individually. |
| Interface | The standard consumer web app in every case, not a developer API. This is what your customers actually use, and it returns local pricing, local retailers and local editorial that the APIs do not. |
| Questions | 10, fixed since August 2026, each asked in its own separate session |
| Session state | Memory off, no chat history, no custom instructions, signed out where the assistant allows it |
| Market | Singapore |
| Collected | 10 August 2026 |
Memory off matters. The same question asked on a signed-in account with memory on can put a different brand first and cite different sources. A reading taken inside an account tells you what AI says to that person, not to a stranger deciding what to buy.
AI answers vary between sessions and change over time. Every figure here is a reading taken on the dates above, not a permanent score. Competitor brands appear because an assistant named them; we report that as fact and do not evaluate their products. See Method & Disclaimer.
Your question set is fixed for at least a quarter so month-to-month figures compare like with like. If a question stops making sense for your category we will propose a change and tell you what it does to comparability before anything is swapped.
Read and signed off before it was sent. Reply to the email this came with and you will get an answer from the person who wrote it.
The free check that came before this report is published in full too, so the two can be read side by side: Free check · COSRX · August 2026.