Skincare ingredients, followed back to the evidence

Every ingredient carries a level saying how far human studies have actually got. Trials that found nothing are recorded the same way.

PopularSunscreen Retinol Niacinamide

An evidence grade on every ingredient
Sources are the papers themselves
Six languages

Skin concernsStart from a concern and follow it to the ingredients.19 IngredientsTopical ingredients, each with an evidence level.64 MedicinesThings that are not cosmetics in Korea. Ask a doctor or pharmacist.10 ProceduresProcedures done at a clinic. Not something you use at home.59 SurgeryOperations performed at a medical clinic.32

Ingredients

See all ingredients →

Strong evidenceUV filter

Sunscreen

An ingredient class — not a single line on an ingredient list

An Australian trial randomised 1,621 people and followed them 4.5 years. Squamous cell tumour incidence was 39% lower (1,115 vs 1,832 per 100,000); basal cell carcinoma showed no difference. This is one of the few entries on this site carrying strong evidence.

Moderate evidenceRetinoid⚠ In pregnancy Do not use

Retinol

In a vehicle-controlled trial, 0.4% retinol applied for 24 weeks moved the fine-wrinkle score by −1.64 against −0.08 for vehicle. Thirty-six people, and it was forearms, not faces. In pregnancy, guidance says do not use it.

Moderate evidenceAntioxidant

Niacinamide

In a 12-week split-face study of 50 people, 5% niacinamide reduced fine lines, yellowing, red blotchiness and pigmented spots. The control was not nothing but the same moisturiser, and the manufacturer paid for the trial.

Medicines

See all medicines →

Things that are not cosmetics in Korea. Ask a doctor or pharmacist.

Moderate evidencePrescription only

TretinoinA prescription medicine in Korea. In a 1991 trial of 251 people, 79% of those using 0.05% for 24 weeks improved against 48% on vehicle. But in a 29-person trial in 2022, active and vehicle both improved 30–40% and could not be told apart.

Strong evidenceOver the counter — ask the pharmacist

Benzoyl peroxideA network meta-analysis of 179 randomised trials and about 35,000 observations found that the most effective topical option for mild-to-moderate acne was a retinoid combined with benzoyl peroxide, reducing lesion counts by 26.16% more than placebo. In Korea, single-ingredient products are available over the counter.

Moderate evidencePrescription only

AdapaleneAll 12 products registered in Korea are prescription-only. In a 67-person split-face trial, adapalene 0.3% with benzoyl peroxide 2.5% reduced scar counts by 15.5% over 24 weeks while the vehicle side increased by 14.4%.

Moderate evidenceOver the counter — ask the pharmacist

Azelaic acidAll six products registered in Korea are over the counter. In a Cochrane review, azelaic acid beat placebo on participants' own assessments (three trials pooled, risk ratio 1.52), and a phase 3 trial of a 15% foam also beat vehicle. The review noted that most included studies were at high or unclear risk of bias.

Moderate evidencePrescription only

EflornithineIn a vehicle-controlled, double-blind trial alongside laser (31 people), complete or almost complete hair removal reached 93.5% and all three endpoints were significant. In a separate 22-person trial of maintenance after IPL, regrowth fell 14% at one month and 17% at six — but the three-month point was not significant.

Strong evidencePrescription only

FinasterideIn male pattern hair loss, two trials in 1,553 men and a Japanese trial in 414 all beat placebo. Hair counts, investigator assessment and a blinded expert photographic panel all pointed the same way, and the placebo groups went on losing hair. But in a trial of 137 postmenopausal women there was no effect. And in a meta-analysis pooling 15 placebo-controlled trials and 4,495 subjects, the relative risk of sexual adverse effects was 1.66 (95% CI 1.20–2.30).

Moderate evidencePrescription only

HydroquinoneIt is the name that appears most often as a comparator on this site, and not one trial here was designed to measure hydroquinone. All three were designed to measure something else, with hydroquinone as the control arm. What that arm did: 4% cream lowered the melasma score (mMASI) by an average of 33% over 90 days (95% CI 23–42%). And it did not differ significantly from thiamidol, cysteamine or tranexamic acid.

Moderate evidencePrescription only

IsotretinoinWe could not put a trial comparing its effect size against placebo on this page. This drug was established in the 1980s, and almost every randomised trial since asks how much, not whether. And on that question two reviews came apart: reading the same four trials, a 2023 review wrote that low dose is preferable, while a 2026 meta-analysis wrote that neither regimen was favoured (p = 0.15). And this page gives one of its badges to pregnancy — even with a prevention programme, there were 3.4 pregnancies per 1,000 courses.

Strong evidencePrescription only

SpironolactonePooling five placebo-controlled randomised trials in 563 patients, the odds of investigator-assessed acne improvement were 6.59 times higher (95% CI 3.50–12.43), with almost no disagreement between trials (I² = 0%). But the pooled patient-rated assessment was not significant (OR 5.22, CI 0.62–44.24, I² = 85%). And the primary outcome of the 410-patient British SAFA trial was a patient-reported quality-of-life score at 12 weeks, where the difference of 1.27 (CI 0.07–2.46) barely cleared zero. At 24 weeks it was 3.77 (2.50–5.03). That trial was paid for by British public research funding, not by a manufacturer.

Strong evidenceOver the counter — ask the pharmacist

MinoxidilVehicle-controlled trials: two in women (381 over 48 weeks, 404 over 24) and two in men (352 over 16 weeks, 90 over 36). In all four, 5% beat vehicle on hair count. But 2% did not differ from vehicle on the users' own assessment, 10% was not better than 5%, and oral minoxidil did not beat topical.

Procedures

See all procedures →

Procedures done at a clinic. Not something you use at home.

The six below are ordered by how well established the evidence is — not by how large the effect is.

Moderate evidenceInjection

Botulinum toxinIn two phase 3 randomised, double-blind, placebo-controlled trials (730 participants in total), both the primary and key secondary endpoints were significantly better than placebo (P < .0001 for each). All 59 products registered in Korea are prescription-only.

Moderate evidenceInjection

Calcium hydroxylapatite (Radiesse)For filling nasolabial folds there are two 12-month randomised trials. But in a 2025 gene-expression trial, CaHA upregulated inflammatory rather than regenerative pathways and showed no evidence of tissue regeneration. And the fashionable hyperdiluted use has no randomised trials at all.

Moderate evidenceInjection

Polycaprolactone (PCL)One randomised trial shows it outlasting hyaluronic acid in nasolabial folds. But the decisive property of this material lies elsewhere: when it blocks a vessel, there is no way to dissolve it quickly. That a 2025 study exists on managing PCL embolism says something in itself.

Moderate evidenceInjection

Human acellular dermal matrix (hADM)Donated human skin is stripped of its cells, and the remaining dermal matrix is particulated and injected. The clinical evidence as a skin booster is a single 20-person split-face trial; the rest is cell, tissue and rat work. There is a separate 202-person trial for filling nasolabial folds, but that is a different product for a different purpose.

Moderate evidenceInjection

Hyaluronic acid fillerIn a double-blind split-face trial following 174 people for 15 months, new hyaluronic acid fillers were non-inferior to an established one. That pattern is characteristic: there are many randomised trials, almost all filler against filler, and they mostly conclude 'no difference'. But it has one property nothing else here has — it can be dissolved.

Moderate evidenceInjection

HyaluronidaseThe antidote used when filler blocks a blood vessel. In a review pooling 21 studies and 231 patients, resolution was 78.4% for skin necrosis and 45.2% for visual impairment, and starting within six hours significantly improved outcomes. But two meta-analyses point opposite ways on dose.

Devices

See all devices →

Procedures done with a device — light, radiofrequency, ultrasound.

The six below are ordered by how well established the evidence is — not by how large the effect is.

Moderate evidenceDevice-based

Monopolar radiofrequencyWe read four randomised trials of monopolar radiofrequency. All of them — including a multicentre study analysing 212 people — used another device as the control, and all of them concluded 'not inferior'. We found no trial comparing it against doing nothing.

Moderate evidenceDevice-based

Microfocused ultrasoundIn a 2023 systematic review of 16 studies, brow lift measured 0.47–1.7 mm and submental area fell 26–45 mm². In the same data, 92% were classed as improved by the treating physician — but only 42% of patients said so themselves. The gap between those two numbers is the honest size of this procedure.

Moderate evidenceDevice-based

Laser toning and picosecond toningIn a split-face trial pitting picosecond against Q-switched toning, there was no significant difference between them, and the melanin index did not improve significantly on either side. In a network meta-analysis of 39 studies and 1,394 participants, the top of the ranking was combination rather than laser alone, and second place went to oral tranexamic acid. Punctate leukoderma is a recognised complication.

Moderate evidenceDevice-based

Microneedle radiofrequencyIn a 30-person split-face trial for acne scars it was comparable to ablative fractional CO2 laser, and the CO2 side hurt more. But in a 20-person trial, alternating it with a non-ablative fractional laser was no better than the laser alone. A review of 16 studies and 481 patients called it 'likely effective' while noting that more randomised trials are needed to establish the right parameters.

Moderate evidenceDevice-based

Fractional laser (Fraxel, fractional CO2)The randomised acne scar trials we read are mostly 'something new against fractional CO2', and they almost always end in 'no significant difference in scar score'. In a 2025 picosecond comparison, mean ECCA scores did not differ (P = 0.209), and the microneedle radiofrequency comparison was similar. But pigment and erythema measures rose significantly on the CO2 side.

Moderate evidenceDevice-based

IPL (intense pulsed light)In a randomised trial following 227 people for two years, the control group received no treatment at all. Clearance of telangiectasia of 90% or more was reached by 66.36% with IPL against 0% in controls; total efficacy was 95.33% against 30.83%. In a photoaging trial, one side of the face was treated and the other left alone — and biopsy showed decreased melanin and increased collagen only on the treated side.

Surgery

See all surgeries →

Operations performed at a medical clinic.

This layer carries no evidence grades — placebo-controlled trials are not possible for cosmetic surgery. The six below are one per area.

EyesLocal

Double eyelid surgeryCreates a crease in the upper eyelid. The suture method and the incision method differ in how long recovery takes and in how much can be undone.

NoseSedation

RhinoplastyChanges the structure of cartilage and bone. The bridge, usually built with an implant, and the tip, shaped with your own cartilage, use different materials and carry different later problems.

BodyLocal

Scar revisionThe scar is cut out and closed again, to become a narrower scar or one lying in a better direction. It is not surgery that removes a scar. And the new scar can turn out worse — particularly if you tend to keloid. This site carries several non-surgical options, and in many cases those come first.

Jaw and contourGeneral

Facial bone contouringThe cheekbone or the lower jaw is cut and repositioned, or shaved, to change the width and line of the face. It is the hardest operation on this site to undo, and the slowest to reach its final shape.

FaceSedation

Buccal fat removalPart of the buccal fat pad inside the cheek is taken out through the mouth. No external scar, and a short recovery. The problem is that it cannot be undone, and that what it removes is exactly the volume that decreases on its own with age.

ScalpLocal

Hair transplantFollicles from the back of the head are moved to where hair has gone. The total number of hairs does not increase. And the hair you did not move keeps thinning after surgery.

How the levels are set here

Cosmetic trials typically run 20–40 participants for 4–12 weeks and are paid for by the manufacturer. That is why 'strong evidence' is rare on this site — sunscreen and prescription retinoids are most of it.

Strong evidence: Consistent across several randomised controlled trials in people, or a meta-analysis.
Moderate evidence: Human trials exist, but they are small or the results disagree.
Early research: Test-tube or animal work, or a small pilot study. Not a result from skin.
Ingredient data: A fact from an ingredient list or a regulatory document. Not an efficacy claim.
Not established: Widely claimed, but not confirmed in people.