Three actives with real evidence: retinoids, vitamin C and niacinamide

Published 6 August 2026Zane Hitchcox, editor

What counts as evidence here, and why so little clears the bar

The bar is a vehicle-controlled human trial: the active in a base, against the identical base without it. Almost every moisturiser improves the look of skin for a few hours, so without that control there is no telling whether the named ingredient did anything the cream it arrived in would not.

Very few cosmetic ingredients have been through that, and even the good end of this literature is thin. When a 2023 systematic review in the Journal of Cosmetic Dermatology went looking for every randomised controlled trial of topical vitamin C in photoaging and melasma, it found seven publications and 139 volunteers.¹⁵ That is fewer people than one mid-sized blood-pressure trial — and still far more than exists for most of the botanicals on a shelf.

Retinoids work through a receptor, and the cosmetic ones have to be converted first

Retinoids are the vitamin A family, and they act by binding nuclear receptors — retinoic acid receptors and retinoid X receptors — which then alter transcription in keratinocytes and fibroblasts. That is the mechanism described in a 2019 review in Postępy Dermatologii i Alergologii, which also sets out why activity is expressed in "retinol equivalents": the family members differ in potency, so a common unit is needed.⁵ Of the vitamin A forms a cosmetic may contain, none binds the receptor as supplied. The one that does is all-trans-retinoic acid, tretinoin, and everything on a skincare label is a precursor to it.

That single fact organises the whole category. Retinyl esters must be converted to retinol; retinol to retinaldehyde; retinaldehyde to retinoic acid. Each step depends on enzymes in the skin, each is incomplete, and potency falls at every stage. In a 1995 University of Michigan study in the Journal of Investigative Dermatology, up to 1.6% retinol under four-day occlusion thickened the epidermis about as much as 0.025% retinoic acid — but it raised epidermal retinyl ester content 240-fold, while retinoic acid itself stayed undetectable or at trace levels.² Retinol is a reservoir that leaks a little active, not the active.

The clinical evidence for tretinoin is the oldest and the strongest. The 1988 JAMA trial from the same Michigan department gave photoaged patients tretinoin on one forearm and vehicle on the other for sixteen weeks, with a separate facial comparison: on the face, 14 of 15 tretinoin recipients improved and none of the vehicle group did.¹ Thirty people is a small trial by any medical standard. It has been replicated many times since, which is why the finding has held.

Cosmetic retinol has its own controlled evidence, and it is a great deal shakier than the shelf implies. A 2007 Archives of Dermatology trial put 0.4% retinol lotion on one arm and vehicle on the other in thirty-six residents of senior citizens' facilities, mean age 87, for twenty-four weeks, and fine wrinkling improved against vehicle.³ Note what that is not: arm skin in the very elderly, applied by study staff.

The closest thing to a test of an ordinary cosmetic is a 2009 study in the Journal of Dermatological Treatment from Tohoku University: fifty-seven middle-aged Japanese women, 0.075% retinol on one half of the face and its vehicle on the other, nightly, for twenty-six weeks. Half were judged improved on fine wrinkling against a quarter on the vehicle side.²⁴ That is a face, a strength a real product might carry, and half a year — and its result is a count of people judged improved, not a measured size of change.

Set against that, the one systematic review to ask the question of over-the-counter retinol products is unkind. Published in the Journal of Clinical and Aesthetic Dermatology in 2021, it found nine randomised, double-blind, vehicle-controlled trials of retinol on facial ageing. Four showed no significant difference from vehicle at all; the reviewer judged the other five to show weak evidence on fine lines only, with flaws serious enough that their positive results "should not inform clinical decision-making but rather may serve as tools for advertising".²⁵ That is one author's appraisal, not a pooled estimate — but four null trials is a fact about the literature, and not one the category advertises.

Retinaldehyde sits between the two. A 1998 Journal of the American Academy of Dermatology study gave 125 patients with facial photodamage 0.05% retinaldehyde, 0.05% retinoic acid, or the retinaldehyde vehicle, and read crow's-feet replicas by optical profilometry. Both actives beat the vehicle at week 18, both effects were smaller by week 44, and retinoic acid irritated more.⁴ The published report gives no direct statistical comparison between the two actives, so the trial establishes that retinaldehyde beats its vehicle — not that it matches the drug.

Timescale, irritation and the pregnancy caution

Look at the clocks. The shortest of these trials ran sixteen weeks; the two retinol trials ran twenty-four and twenty-six; the retinaldehyde trial ran forty-four. A product promising retinoid results in a fortnight is promising something none of them was built to detect.

Irritation is not a side issue, it is the dose-limiting factor. The New Zealand data sheet for tretinoin 0.05% cream lists peeling and erythema as the expected early effects, with transient warmth or stinging, and instructs that excessive reactions be met by reducing frequency or pausing treatment. It also requires minimised sun exposure and routine high-SPF sunscreen, because treated skin is more susceptible to ultraviolet.⁷ Cosmetic retinol is milder, but the same dose-response applies.

Concentration is also a legal question in one large market. Commission Regulation (EU) 2024/996 caps vitamin A at 0.3% retinol equivalent in most leave-on and rinse-off products and 0.05% in body lotion, with the warning "Contains Vitamin A. Consider your daily intake before use." — driven by total dietary vitamin A intake, not by irritation.⁸ That cap binds Europe and nothing else; what a jar sold in Auckland must comply with is a question for the agency that actually regulates cosmetics in New Zealand, which is not the one most people assume.

On pregnancy the position should be stated plainly. Tretinoin cream carries pregnancy category D in New Zealand, and the data sheet says it is not recommended during pregnancy or in women of childbearing potential.⁷ The best human evidence is a 2015 British Journal of Dermatology meta-analysis of 654 first-trimester topical-retinoid exposures against 1,375 unexposed controls: no significant increase in major congenital malformations (odds ratio 1.22, 95% CI 0.65–2.29), miscarriage, stillbirth, low birthweight or prematurity — and authors who wrote that the power was "not adequate to justify the use of topical retinoids during pregnancy".⁶ Both are true at once: an inadvertent exposure is not a catastrophe, and that is not a clearance. Anyone pregnant or planning to be should take this decision with a doctor, not with an article.

With vitamin C, the formulation is the active ingredient

L-ascorbic acid is the form the skin uses and the form hardest to keep. It oxidises on exposure to air, light and metal ions, and once oxidised it is a different molecule. The most useful paper on this is not a clinical trial: a 2001 Dermatologic Surgery study from Duke put L-ascorbic acid and three derivatives on pig skin and measured what got in. It must be formulated below pH 3.5 to enter the skin at all, absorption was maximal at 20%, and magnesium ascorbyl phosphate, ascorbyl-6-palmitate and dehydroascorbic acid did not raise skin levels of L-ascorbic acid at all.¹⁰

That is why packaging and pH matter more than the number on the front: a 20% serum that has gone amber delivers less than a well-formulated 10% one, and the percentage is the only part of that a buyer can see. It is also why the derivative story deserves scepticism. Derivatives are used because they are stable, and stability is a formulation convenience, not evidence of activity.

The photoprotection claim rests on thinner ground than its confident repetition suggests. The founding demonstration, in the British Journal of Dermatology in 1992, showed that topical vitamin C protected porcine skin against UVB erythema and sunburn-cell formation and against UVA phototoxicity.¹¹ That is a real result in a pig. Vitamin C is not a sunscreen, is not tested under any sunscreen standard, and carries no SPF.

The human trials are small. A 1999 Archives of Otolaryngology — Head & Neck Surgery study ran a commercial ascorbic acid serum against its own vehicle on opposite sides of nineteen faces for three months.¹² A 2003 Experimental Dermatology trial ran 5% vitamin C cream against its excipient for six months on the low neck and arms and found improvement in a global score and in skin microrelief — though the published report never states how many volunteers were enrolled.¹³ A 2002 Dermatologic Surgery half-face study of a 10% ascorbic acid plus 7% tetrahexyldecyl ascorbate gel ran twelve weeks in ten patients.¹⁴ That is the shape of the literature: the 2023 review that collected it found vitamin C smooths and lightens skin, and that the ideal concentration is not established.¹⁵

That 73.7% is worth an aside, because it circulates as though the wrinkles got 73.7% better. Do the arithmetic on the four percentages the paper reports — 73.7, 68.4, 84.2 and 57.9 — and each is an exact multiple of one nineteenth: 14, 13, 16 and 11 of the nineteen participants.¹² The economical reading is that these count subjects who did better on the active side, not the size of any change. The paper never says which, and a figure whose unit cannot be established from the publication should not be repeated as if it could.

One detail in that review is worth stopping on: hydration improved equally on the vitamin C side and the placebo side. That is the vehicle doing its job, and the shortest illustration of why an uncontrolled before-and-after photograph proves nothing.

Niacinamide is the least glamorous of the three and the best behaved

Niacinamide — nicotinamide, vitamin B3 — has the broadest set of controlled skin trials of any ingredient that is unambiguously a cosmetic, and the clearest conflict of interest: two of its defining studies were run by Procter & Gamble scientists and one by Kanebo, stated openly in the author affiliations. That does not make a result false. It does mean much of it has never been independently replicated.

The barrier work is the most mechanistically satisfying. A 2000 British Journal of Dermatology paper from Kanebo showed that nicotinamide raised ceramide synthesis in cultured human keratinocytes 4.1 to 5.5-fold over six days, by upregulating serine palmitoyltransferase — and that topical application raised stratum corneum ceramide levels and reduced water loss in dry skin.¹⁶ The fold-changes are cell culture; the human half of the paper never says how many people were tested. The confirmation is a 2005 International Journal of Dermatology study in twenty-eight atopic dermatitis patients, 2% nicotinamide cream on one forearm against white petrolatum on the other: nicotinamide significantly reduced transepidermal water loss and petrolatum did not.²⁰ Beating white petrolatum on a barrier measure is a meaningful comparator, not a straw man.

On pigmentation, a 2002 British Journal of Dermatology paper reported both a mechanism and two clinical arms. Niacinamide did not inhibit tyrosinase or melanin production in melanocytes; it inhibited melanosome transfer to keratinocytes by 35–68% in co-culture. Clinically, eighteen Japanese women with facial hyperpigmentation used 5% niacinamide against a vehicle moisturiser and showed less pigmentation and greater skin lightness at four weeks; a separate 120 women were each given two of vehicle, sunscreen, or 2% niacinamide plus sunscreen.¹⁸ Read that second arm carefully: its niacinamide arrives on top of a sunscreen, so it cannot separate the vitamin from the sun protection. Eighteen people carry the unconfounded claim.

The appearance trial most often quoted is a 2005 Dermatologic Surgery split-face study of 5% niacinamide against its own vehicle in fifty women over twelve weeks, reporting less fine wrinkling, pigmentation, redness and yellowing.¹⁷ Its authors worked for Procter & Gamble, its endpoints were graded from facial images, and it has not been independently repeated. Reasonable evidence, and one trial.

The sebum evidence is where an honest reading has to contradict the marketing. A 2006 Journal of Cosmetic and Laser Therapy paper reported two trials of 2% niacinamide. In Japan, 100 subjects against a placebo moisturiser for four weeks showed a significantly lowered sebum excretion rate; in the United States, 30 subjects over six weeks showed a fall in casual sebum level but no significant fall in the excretion rate.¹⁹ "Niacinamide reduces oil" is a stronger sentence than that data supports.

For acne, the trial usually cited is a 1995 International Journal of Dermatology study of 4% nicotinamide gel against 1% clindamycin gel in 76 patients over eight weeks, in which neither arm significantly beat the other.²¹ There was no vehicle arm, so the finding is that nicotinamide performed comparably to a topical antibiotic — not that either outperformed doing nothing. Acne is a medical condition; how to treat it is a decision for a doctor.

Niacinamide is not niacin, and it does not cause flushing

The worry that niacinamide will make skin flush comes from confusing two molecules. Nicotinic acid causes flushing from about 30–50 mg. Nicotinamide, in the words of the United States National Institutes of Health Office of Dietary Supplements, "does not produce skin flushing because of its slightly different chemical structure".²³ Redness after a niacinamide serum is far more likely a response to something else in the formula — see fragrance and essential oils — than a vitamin B3 flush.

One result does not transfer to topical use: the 2015 New England Journal of Medicine ONTRAC trial, in which 386 Australian patients with at least two prior non-melanoma skin cancers took 500 mg of nicotinamide twice daily or placebo for twelve months and had 23% fewer new such cancers (95% CI 4–38, P=0.02), with no benefit once the tablets stopped.²² That is an oral, dosed, secondary-prevention result in a high-risk population. It says nothing about a serum.

Dose, formulation and time are what separate a studied ingredient from a studied product

Every trial above tested a concentration, in a base, for a stated number of weeks, and none of those three variables is visible on a shelf. An ingredient list gives order by weight only down to one percent, so a product can carry niacinamide below the preservative and still print the name on the front — the practice called fairy dusting, against which reading an INCI list is the only tool a buyer has, and a blunt one. "Clinically proven" and "dermatologist-tested" do not close the gap; several such phrases have no legal definition at all. Some of the vagueness is legal rather than evasive: the FDA counts altering melanin production among the claims that turn a cosmetic into a drug.⁹

These three ingredients are the exception, not proof that the category works. They are the small set of cases where someone ran a vehicle-controlled trial and got back an answer that was positive and modest — and, for cosmetic retinol, a set where four such trials came back with nothing. What survives is real and small: fine wrinkling scores, profilometry roughness, transepidermal water loss, pigment measurements. Not the effects of a procedure, and they cannot be, for reasons that have nothing to do with formulation and everything to do with what a topical can reach.