The Clean Girl Verdict

There is a tendency in beauty writing to treat every micro-aesthetic that crosses TikTok as a discrete cultural event, separable from the one before it and the one after. Strawberry girl makeup. Espresso makeup. Latte makeup. Tomato girl summer. Mob wife era. Office siren. The pace of the cycle, in the years since lockdown produced the conditions for at-home beauty content to displace magazine beauty content, has accelerated to the point where the trends are now visibly out of phase with the seasonal cycle of clothing — a new face every six weeks, sometimes a new face every three.

What the data in this corpus shows, if you stand far enough back from the micro-cycles to see the macro-trend, is that almost all of these phases are variations on a single dominant aesthetic that has been building since roughly 2018 and that consolidated its grip on the cultural mainstream during the pandemic. That aesthetic is variously called clean girl, no-makeup makeup, natural enhancement, quiet luxury beauty, skinification, and a dozen other names depending on which Substack or magazine you happen to be reading on any given week. The names matter less than the shared underlying principle, which is: the goal of a good makeup look is, increasingly, to look as though you are not wearing makeup at all.

I have spent the last week reading 917 real user submissions to an AI makeup evaluator. The corpus is, in its dominant tendency, overwhelmingly devoted to this aesthetic. The looks that score highest in the corpus are the ones that look least like makeup. The looks that score lowest, with the exception of the 130 submissions where no makeup is detected at all, are the ones that read as obviously and effortfully pigmented. The data is unambiguous on this point. What is more interesting — and what occupies the second half of this report — is what the data does not say. It does not say that the aesthetic is universally accessible. It does not say that the aesthetic is racially neutral. It does not say that the aesthetic is, on its own terms, achievable for everyone who tries to achieve it. The clean girl wins the score. The clean girl also, on closer inspection, costs more than the score reveals.

Section 1What "Clean Girl" Actually Means, And When It Got Here

It is worth, before walking through the numbers, being specific about what the aesthetic is and when it arrived, because the conversation tends to assume both things. The conversation is wrong about both in instructive ways.

The phrase clean girl in its current usage was popularized on TikTok in roughly the spring and summer of 2022, with creators including Tinx, Vivien Conca, and a much larger group of subsequently anonymous originators producing short-form videos under the tag. The visual grammar of the look, as it stabilized over the following six months, is by now familiar: slicked-back hair in a low bun or claw-clipped, dewy skin with no visible base or with a tinted moisturizer rather than a foundation, a brushed-up brow with no visible product, a single cream blush blended high on the cheekbone, no eyeshadow or a single neutral wash, a clear or tinted lip balm. Pairs with gold hoops and a small gold necklace and a Hailey Bieber-style nail. The whole look reads as I just rolled out of bed and happen to have inherited expensive bone structure, which is, of course, exactly what it is meant to communicate.

The aesthetic did not, however, originate on TikTok. The TikTok phrase was the cultural condensation of a much longer trajectory in beauty marketing that traces back at least to the founding of Glossier by Emily Weiss in 2014, which itself drew on the editorial sensibility of Weiss's earlier blog Into the Gloss. Glossier's defining advertising image, the one I remember from a New York City subway car in 2016, was a face that the brand's copywriting described in terms of "skin first, makeup second": a Boy Brow, a Cloud Paint cheek, a Balm Dotcom lip. The look was, in its essence, the clean girl aesthetic, six years before the phrase existed.

You can trace the lineage back further. Bobbi Brown, the makeup artist who founded her eponymous line in 1991, has been writing about "natural" makeup as a positive aesthetic value, rather than as a euphemism for laziness, since the late 1980s. Her thesis, expressed across multiple books — Bobbi Brown Beauty (1997), Bobbi Brown Makeup Manual (2008), and many subsequent titles — is that makeup should enhance the wearer's underlying face rather than impose a constructed face onto it. That formulation, written before "no-makeup makeup" had a TikTok hashtag, anticipates the clean girl aesthetic almost line for line. You can go further back still, to Calvin Klein's 1990s campaigns with Kate Moss — the heroin chic moment, distorted in cultural memory but in its commercial execution a deliberate rejection of the Joan Collins–era full-face glamour that preceded it. Or back to Coco Chanel's well-circulated maxim about taking one accessory off before leaving the house, which is the same principle, applied to a different surface.

The point is that the clean girl aesthetic is not new. What is new is its universality. For most of the late twentieth century, "natural" makeup was a stylistic option that existed alongside other stylistic options — Hollywood glamour, punk, goth, the 1980s color block, the 1990s grunge, the early-2000s baby pink. What the data in this corpus suggests is that, in mid-2020s mass-market beauty culture, the natural aesthetic has stopped being one option among several and has started being the default. Roughly sixty percent of the looks in the corpus describe themselves in some variant of natural or minimalist terms. Approximately one percent describe themselves as full glam. The competition is over.

Section 2The Dominance, Charted

The OutfitScore makeup pipeline tags each analysis with a "primary look" field that captures the model's best guess at the aesthetic category of the submitted image. The vocabulary the model produces is open-ended; the model is not constrained to a fixed list, and there is significant variation in the specific phrases it uses. Across the 917 analyses, I counted just over four hundred unique primary-look strings. Most of them are clearly variations on a much smaller number of underlying categories.

I built a hand-curated taxonomy to collapse the variation. The categories are mine; reasonable readers might draw the lines slightly differently, but the dominant tendency is robust to taxonomy choices.

The hierarchy is brutal. The natural-leaning category absorbs nearly three-fifths of the corpus. Add the "soft glam" bucket — which is essentially natural-leaning with a touch more product, the kind of look Hailey Bieber wore to most of her 2023 public appearances — and you are at roughly three quarters of all submissions. The combined glam-leaning categories (theatrical, editorial, full glam) account for under ten percent. The aesthetic competition that beauty magazines like to dramatize as natural versus glam is, in actual user submissions to an AI makeup evaluator in 2026, not a competition at all. It is a rout.

Two specific findings inside this distribution are worth pausing on. The first is the size of the "no makeup detected" category. One hundred thirty submissions, or 14.2% of the total, were sent to a makeup analyzer with no detectable makeup on the user's face. That is its own remarkable phenomenon and I will treat it as a separate report. The second is the smallness of the full-glam category. Eleven submissions. Across roughly four months. In a corpus where the dominant user behavior is to submit a face to an algorithm for evaluation, almost nobody is showing up with a full cut crease and a winged lip. The constituency for editorial glamour, as a self-applied everyday aesthetic, is, in this corpus, essentially gone.

Section 3What the AI Thinks of the Clean Girl

The aesthetic verdict of the corpus matches what the broader cultural conversation has been saying for three years. The empirical verdict — that is, the verdict the AI scoring engine returns when actually evaluating these looks against a fixed rubric — is, by my reading, more interesting, because it can be falsified. It says specifically which versions of the clean girl aesthetic work, and which versions fail, and by how much.

The mean score by category, restricted to categories with at least twenty observations, is as follows.

Several things are happening in this table. The headline finding is that soft glam — that is, the clean girl aesthetic with a small amount of additional product, the lipstick and lash configuration that reads as polished rather than as natural — outscores pure natural-leaning looks by a couple of points. The peak of the curve, in score terms, is not at the absolute minimum of cosmetic intervention. It is slightly to the right of the absolute minimum. The model is not telling you to wear nothing. It is telling you to wear less than the full-glam end of the spectrum, but more than the bare-face end. The Goldilocks zone is real, and it is centered around what you might call polished natural.

The second thing happening in the table is the steepness of the drop on the theatrical and full-glam end. Theatrical submissions average a 62; full-glam submissions average a 59.5. These are not poor scores in absolute terms, but they are about twelve points below the soft-glam peak. Twelve points, on a hundred-point scale, is the difference between a "Good" and a "Fair" verdict. It is, in score terms, the cost of going all the way in the dramatic direction.

The third thing happening, and the one that is most directly anchored to the existing experimental literature, is that the entire shape of this distribution matches almost exactly the finding from Nancy Etcoff and colleagues' 2011 study in PLoS ONE. Etcoff's team showed 149 raters a series of faces under four conditions — bare-faced, natural makeup, professional makeup, and glamorous makeup — and measured perceptions of attractiveness, competence, likeability, and trustworthiness. The headline finding was that all makeup conditions outscored bare-faced on attractiveness, but that the natural and professional conditions outscored the glamorous condition on competence and trustworthiness. The glamorous condition, in particular, was the one that produced reliably lower ratings on competence, even though it produced the highest ratings on attractiveness on certain dimensions. The corpus in this report reproduces that pattern with depressing fidelity. The AI evaluator is not, in any sense I can detect, special. It is encoding an aesthetic preference that has been documented in experimental psychology for fourteen years.

Section 4What Makes the Top Tier of Clean Girl Looks Win

I extracted the 144 submissions scoring 80 or above and looked specifically at the subset that fell into the natural-leaning categories. That subset — roughly 115 of the 144 — gives the cleanest possible answer to the question which specific clean-girl looks actually work, and which versions are merely TikTok pantomime?

Three traits appear with overwhelming regularity in the top-tier clean girl looks, each of which reflects a discipline the looser TikTok renditions of the aesthetic tend to skip.

1. Even skin tone, not just bare skin

The most frequent strength in the top-tier subset is "even skin tone." This is not the same as bare skin. It is skin that has been treated to read as evenly pigmented across the entire visible face — no visible redness at the cheeks, no shadow under the eyes, no patchiness at the forehead — without the use of a base heavy enough to read as foundation. The technique here is fairly specific: a sheer tint, often a Glossier Skin Tint or a Westman Atelier Vital Skin Foundation Stick or a Charlotte Tilbury Hollywood Flawless Filter mixed with moisturizer, applied not as full coverage but as targeted spot correction. The model is rewarding the output of this technique. It is not rewarding the bare-skinned input.

This is, I think, the most consequential single fact about the clean girl aesthetic that doesn't get said honestly enough in beauty media. The look's defining visual quality — skin that reads as healthy and uniform without reading as covered — is almost never achieved by going truly without product. It is achieved through a more skillful and lighter-touch product regime than the full-coverage approach it replaced. The work that used to be done by Charlotte Tilbury Magic Cream and a heavy concealer is now being done by a lighter cream and a different concealer. The thing that has changed is not the amount of effort involved. It is the visibility of the effort.

2. Cohesive color palette

The second most frequent strength, appearing in 28 of the top-tier strengths, is "cohesive color palette." In the natural-leaning subset, this almost always means that the cream blush, the lip color, and the eye color (if any) are drawn from a single tonal family. A peachy cheek pairs with a peachy lip and a peachy lid wash. A rosy cheek pairs with a rosy lip and a rosy lid wash. The Pinterest moodboard for this is fairly explicit: the colors of a single fruit, or a single flower, or a single sunset, applied at low saturation across all three pigmented surfaces.

This is consistent with how the makeup artist Pat McGrath talks about color theory in her own work, and how Lisa Eldridge (a former Lancôme global creative director) writes about it in her 2015 book Face Paint: The Story of Makeup. The principle is that the eye reads discord faster than it reads specific colors. A face with a coral lip and a magenta cheek will read as discordant before the viewer can articulate why, even if both colors are individually flattering. The clean girl aesthetic, by collapsing the pigmented surfaces toward a single tonal family, eliminates the discord.

3. Brow grooming that does not read as brow makeup

Brows appeared as the third most common top-tier strength in the makeup corpus and also, in Report 03, as the third most common deficiency. The asymmetry — well-groomed brows are barely noticed, badly-groomed brows are the first thing seen — is the same. What distinguishes the top-tier clean girl brow from the median brow is that the product, when present, is invisible. The brow has been brushed up, set with a clear gel, and (where filled) filled with a pencil shade lighter than the natural hair color, applied in feathered upward strokes that mimic the direction of growth. Anastasia Beverly Hills, Benefit, and Glossier all sell versions of the appropriate product. The brand matters less than the application.

This is the same finding as the application thesis from Report 03, applied to a specific feature. The product is not the bottleneck. The brush stroke is. A pencil two shades too dark, applied in horizontal sweeps, is unfixable as long as the stroke remains horizontal. The same pencil one shade lighter, applied in upward feathers, becomes invisible. The brand is, again, downstream.

Section 5The Price of the Aesthetic

I want to spend a section on what the clean girl aesthetic costs. I do not mean only in dollars, although the dollars are real. I mean in the broader sense — what it requires of the person attempting it, what it assumes about that person's existing baseline, who it includes, and who it leaves out. The data in this corpus does not, on its own, answer all of these questions. But it gestures at them clearly enough that a careful reader can see the shape of the problem.

The skin problem

The clean girl aesthetic, at the top tier, requires skin that already reads as healthy. The corpus is unambiguous on this point: the top-tier natural-leaning looks belong to faces where the underlying skin is in good condition before any product is applied. The technique I described above — sheer tint, targeted spot correction, no full base — works because there is relatively little to correct. On a face with active acne, with persistent redness, with melasma, with significant hyperpigmentation, the same technique fails. The light-touch products do not provide enough coverage; the corrected zones become visible against the uncovered zones; the overall look reads as patchy rather than as fresh.

This is not a matter of skill. It is a matter of substrate. The clean girl aesthetic has, as a structural feature, an implicit prerequisite that the user's skin be in a state where minimal coverage is sufficient. For users whose skin is not in that state — and this is a meaningful fraction of the human population at any given time — the aesthetic is, in its purest form, unachievable. The fallback for these users, which I saw repeatedly in the lower-tier submissions in the corpus, is to attempt the aesthetic anyway with insufficient coverage and produce a look that the model penalizes for unevenness. The aesthetic is, in this sense, a trap: it punishes faces that try to participate without having the necessary baseline.

The fix, as the beauty industry has been quietly selling for the last five years, is better skin. The shift in spending patterns, documented in tracking by NPD Group (now Circana) and Mintel since roughly 2018, has been away from cosmetics and toward skincare. The phrase skinification of makeup is industry jargon for this shift. The implicit promise is that, with sufficient investment in serums and treatments and dermatologist visits, the user can move their baseline skin into the state where the clean girl aesthetic becomes achievable. This is sometimes true. It is also, in dollar terms, a much larger investment than the cosmetics it replaces.

The class problem

Pierre Bourdieu, in his 1979 work Distinction, made the argument that taste is a class signal even when it presents itself as natural or universal. The taste preferences of any given social class, Bourdieu argued, are not random or autonomous; they are systematically aligned with the material conditions of that class, and they are taught and reinforced through the institutions and habits that the class controls.

The clean girl aesthetic, considered through this lens, is a fairly clean illustration of his point. It signals — by demanding good skin, by demanding the time and money for skincare, by demanding access to the specific products that produce the invisible-product look, by demanding the leisure to apply them carefully — a particular class position. The aesthetic of looking like you woke up like this requires having access to the conditions under which one might plausibly have woken up like that. This is true of every aesthetic to some degree; what is specific to the clean girl is the way it disavows its own infrastructure. A full-glam look at least has the honesty of being visibly costly. A clean girl look hides its costs.

The journalist Sophie Gilbert has made related arguments in The Atlantic, particularly in pieces on the "quiet luxury" trend that overlapped with the clean girl moment in fashion. The cultural critic Ana Andjelic, who writes about the marketing of aesthetic categories with unusual analytical clarity in her newsletter The Sociology of Business, has made the parallel argument in commercial terms: aesthetic trends that signal class without visibly costing money are, marketing-wise, more lucrative than aesthetic trends that signal class through visible cost, because the audience is broader. Clean girl outsells full glam because more people aspire to it; more people aspire to it because the cost is hidden.

The race problem

The most visible exemplars of the clean girl aesthetic, in the form the mass market has adopted it, are mostly white, mostly thin, mostly straight-haired, and mostly possessed of the kind of facial structure that has historically been over-represented in Western beauty media. Hailey Bieber, Sofia Richie Grainge, Pamela Anderson in her 2023 incarnation, Bella Hadid's 2024 reset. The pattern is, on inspection, fairly uniform.

This has been noted, accurately and repeatedly, by Black and Brown writers in beauty media. Vox's Aja Romano and The Cut's contributing writers have made the case at length that the clean girl aesthetic, in its dominant commercial form, codes as a specific kind of bodily privilege — that the "natural" face the aesthetic celebrates is, in fact, a fairly narrow racialized facial archetype, and that the aesthetic's spread on TikTok has, intentionally or not, reinforced the visual norms of that archetype against alternatives.

The data in this corpus cannot adjudicate this argument. I do not have reliable ethnicity tagging for the submissions, and I would not trust the model's inferred ethnicity even where it provides one. What I can say is that the recommendation patterns in the corpus do encode certain assumptions about skin behavior that, in the cosmetic-science literature, are documented to vary by skin type — the way oils sit on the face, the way concealers oxidize, the way certain tints read on melanin-rich versus melanin-poor skin. The model has been trained on a distribution of faces that almost certainly skews toward the lighter end of the spectrum. The advice it generates will, on average, be better-calibrated for lighter-skinned users than for darker-skinned users. This is a real limitation, and one I am working to address through a more diverse training distribution. I will report on progress in a future update.

Section 6The Cycle

Cultural aesthetics, the journalist Anne Helen Petersen has argued in her newsletter Culture Study, do not last forever. They peak, they saturate, they decay, and they are replaced by something that often turns out, on inspection, to be a partial rebellion against them. The dominance of the clean girl aesthetic in the mid-2020s is, by the standard length of these cycles, approaching saturation, and the partial rebellions are already visible.

The "tomato girl summer" of 2023, popularized in the spring of that year by a handful of TikTok creators (notably Aimee Song and assorted Vogue-trained imitators), was the first visible partial rebellion. The aesthetic added one specific element back into the clean girl framework: a saturated, red-orange lip and cheek combination, drawn from a Mediterranean-coded color story. The look retained the clean girl's skin, brow, and hair architecture, but reintroduced color where the clean girl had subtracted it. It was, in score-data terms, modest in its claims: a soft-glam variant with a tomato-shifted palette. The aesthetic peaked in July 2023 and faded by September.

The "mob wife era" of early 2024, popularized by TikTok creator Kayla Trivieri in a single viral video on January 17, 2024, was a much more aggressive rebellion. The aesthetic reintroduced deliberate visual effort: smoky eyes, dark lips, fur coats, gold jewelry, the entire Italian-American 1990s diaspora image. It was the first aesthetic in three years to explicitly reject the principle that a good face should look effortless. The score data from our corpus suggests that, at the application level, mob wife is much closer in scoring profile to the theatrical/dramatic bucket than to the natural-leaning bucket — that is, on the AI rubric, it scored worse on average than the clean girl looks it was rebelling against. The cultural verdict was, however, more enthusiastic than the algorithmic verdict, and the trend had real cultural reach through the winter of 2024 before subsiding into a more diffuse "maximalist" residue.

What comes next is, at the time of writing, unstable. Vogue Business's State of Fashion: Beauty 2026 report, published in February, identifies "post-clean girl" as an organizing trend without specifying what its content will be. Bof's 2026 forecasts use similar hedged language. The honest answer is that no single new aesthetic has yet stabilized at the scale that the clean girl achieved. The corpus in this report covers the late period of clean girl dominance, with early signals of its successors. I expect to revisit this question in a follow-up report in late 2026, with a year more of data.

Section 7What I Actually Think

I have tried, in the first six sections of this report, to keep my own opinions out of it. Where I have offered them, I have flagged them as opinions. The data has been allowed to do most of the talking. In this section I want to drop the pose and say what I actually think, because it is what readers of these reports have, in their emails to me, asked for.

I think the clean girl aesthetic is, on its own narrow aesthetic terms, basically correct. The data in this corpus confirms what experimental psychology has been documenting for fourteen years: subtle makeup reads better than dramatic makeup on most of the dimensions humans evaluate faces on, and the aesthetic gain from going dramatic is, on average, smaller than the aesthetic loss. A user who is choosing between full glam and clean girl as their everyday default is, on the data, choosing correctly when they choose clean girl. The aesthetic wins the score because it is, in fact, the visually superior choice for most faces in most contexts.

I think the clean girl aesthetic is, on its broader social terms, more problematic than its proponents like to admit. It hides its own infrastructure. It pretends not to require what it requires. It frames an aesthetic that costs hundreds of dollars a month in skincare as an aesthetic of effortlessness. It systematically rewards users whose underlying faces and skin already match the aesthetic's narrow archetype, and it systematically penalizes users whose faces do not. It is, on inspection, a much more demanding aesthetic than its marketing pretends, and the demand falls disproportionately on people who are not already its winners.

I think the right response to this, for an individual reader, is neither to embrace the aesthetic uncritically nor to reject it on principle. It is to use it tactically. The principles that make the clean girl look win — even skin, cohesive palette, restraint, well-groomed brows — are durable principles, supported by both the data in this corpus and by the experimental literature stretching back to the early 2000s. The specific commercial expression of those principles in the form of clean girl is contingent. You can take the principles without taking the specific commercial wrapper. You can use a peachy cream blush and a sheer tinted moisturizer and a brushed-up brow and a clear lip balm without identifying yourself as a clean girl. The aesthetic move underlying the look is older than the TikTok hashtag and will outlast it. The tribal identification is optional.

I also think that the next phase, when it arrives — and it will arrive, on the cycle Petersen and others have documented — will, by design, feel transgressive against the current aesthetic. The cultural pendulum is built to swing. Whatever follows clean girl will be marked by its visible differences from clean girl, in the same way that clean girl was marked by its visible differences from the Instagram face that preceded it. The data in this corpus will, by the time of the next major report in this series, look like a snapshot of a moment that has already started to feel dated. That is fine. The principles will not change. Only the wrapper around them will.

Section 8Limitations

The standard limitations from earlier reports in this series apply here. The corpus is voluntary and self-selecting. The scoring engine is a multimodal model with the biases of its training distribution. The corpus is overwhelmingly female (96.5%). The time window is roughly four months in early 2026.

Three limitations specific to this report deserve mention. First, the taxonomy I used to collapse the four hundred unique primary-look strings into six categories is my own and is debatable. Reasonable readers might draw the lines slightly differently — moving some "natural glow" looks into the soft glam bucket, or splitting the "clean girl" subset from the broader natural category — and the percentages would shift by a few points. The dominant tendency is robust to those alternatives; the specific numbers are not.

Second, the corpus contains no information about user intent. I do not know whether users who submitted natural-leaning looks did so because they preferred the aesthetic or because they did not have the time or skill or budget for a more elaborate one. The 60% number is therefore a measure of what users uploaded, not of what users would have uploaded if all variables had been free. There is almost certainly a significant subset of the corpus where the natural look is a default rather than a preference. Disentangling default-driven submissions from preference-driven submissions would require additional instrumentation in the platform, which I am considering for a future iteration.

Third, the racial and class arguments in Section 5 are arguments. They are not derived from the corpus in the way the score and theme findings are. They are derived from the existing cultural critique of the clean girl aesthetic by writers and academics with more expertise in the relevant questions than I have. I have tried to summarize their arguments fairly and to flag where my own contribution stops and theirs begins. Readers who want to engage with those arguments at their primary source should follow the references in Section 9 and read the linked work directly.

Section 9Conclusion

The clean girl aesthetic, in the form it has assumed in mid-2020s mass-market beauty culture, dominates the corpus of real makeup submissions to a degree I had not previously seen quantified. Roughly six out of every ten users who submit a face to an AI makeup evaluator are presenting a face that, in its dominant features, conforms to the principles of the aesthetic. Only roughly one in a hundred is presenting full glam. The competition the beauty industry likes to dramatize between natural and glamorous as coexisting aesthetic options is, in the data, not really a competition at all.

The AI scoring engine concurs with the cultural verdict: natural-leaning looks score higher than glamorous looks, with the peak of the curve at a "polished natural" point that combines minimal base with deliberate but restrained use of cream blush, brow, and lip. This finding is, on the experimental psychology literature, well-anchored: it reproduces the Etcoff and colleagues 2011 finding almost exactly, fourteen years later, in a corpus that is two orders of magnitude larger than the original study sample.

The aesthetic, however, costs more than its marketing admits. It requires good underlying skin. It requires time and money for the skincare that produces good underlying skin. It systematically rewards users whose faces already conform to a fairly narrow racial and structural archetype. It hides its infrastructure beneath a presentation of effortlessness. These are not, on the data alone, fatal objections to the aesthetic; they are honest accounts of what the aesthetic asks. The reader who chooses to participate should know what is being asked.

And finally, the aesthetic will not last. No aesthetic does. The current dominant phase began consolidating in roughly 2018 with the founding of Glossier and reached its mass-market peak in 2022–2024 with the TikTok diffusion of the clean girl hashtag. The partial rebellions of 2023 and 2024 — tomato girl, mob wife — were small enough to leave the dominant aesthetic intact. The larger replacement, when it arrives, will look meaningfully different. The principles underneath it, on the data in this corpus, will probably not.