Open data studies on how people actually dress.
Plain-language, evidence-backed reports on AI fashion analysis, scored outfit datasets, and the behavioral patterns behind everyday dressing.
OutfitScore Research is the publishing arm of the OutfitScore platform. Every report draws on aggregated, de-identified data from real outfit submissions and is released under CC BY 4.0 for free reuse by journalists, researchers, and fashion writers. Methodology is documented in full. Datasets are available on request for academic and editorial use — write to [email protected].
The Anatomy of a Score
What 492 AI-analyzed outfits reveal about how people actually dress.
A data-driven study of 492 outfit submissions scored by AI between April and May 2026. Identifies the five dominant deficiency clusters — fit, accessories, texture, layering, and footwear — that account for nearly 90% of all improvement recommendations. Quantifies the 25-point gap between occasion-specific and undirected dressing, and proposes a ranked hierarchy of interventions by impact-per-effort.
The Psychology of Fashion Perfection Seekers
What 492 self-submitted outfits reveal about the pursuit of sartorial approval.
A behavioral analysis of the same 492-outfit corpus. Identifies six interlocking findings — the Pursuit Paradox, the Dunning-Kruger Wardrobe, the Comfort Zone Plateau, the Variance Paradox, the Asymmetric Self-Critic, and the Mirror Confessional — and proposes a perfectionism-vs-satisficing framework for escaping the score plateau most dressers find themselves stuck in.
The Anatomy of a Makeup Score
What 917 real faces reveal about modern beauty — and the one thing the industry refuses to talk about.
A data-heavy audit of 917 makeup analyses identifying the five deficiencies that explain almost all score loss. The headline finding: application technique, not product choice, is implicated in 28.6% of all recommendations — more than any single product category. Top-tier looks share three traits the beauty industry rarely foregrounds: even skin tone, cohesive palette, well-groomed brows. Includes a recommendation hierarchy and references to Etcoff, Mulhern, Tolentino, and Bobbi Brown.
The Clean Girl Verdict
Why 60% of real makeup submissions now embrace the no-makeup aesthetic — and what the data says about whether they're right.
A behavioral audit of the same 917-look makeup corpus. Roughly six in ten submissions describe themselves in some variant of natural / clean girl / no-makeup makeup terms; only eleven describe themselves as full glam. The AI rubric agrees: natural-leaning looks outscore glam-leaning ones by ~12 points. But the data also exposes the price of the aesthetic — the underlying skin, time, money, and racial archetype it quietly assumes. Includes a cultural cycle diagram and references to Bourdieu, Tolentino, Etcoff, and Vogue Business trend reporting.
The Neglected Hem
What 20,814 AI-scored outfits reveal about pooling, break, and unfinished trousers.
A frozen snapshot of 20,814 scored fashion and complete analyses. After accessories, texture, and the French tuck — already published — the most common unpublished finish flag is hem pooling and a dirty break: 1,016 rows (4.9%), peaking at 6.5% in the 60–74 band. Keyword method, n-floor 100, no causal lift. Pitch this URL, not the rater.
The One-Notch Lift
What 10,050 men's outfit scores reveal about Casual vs Smart Casual.
A frozen snapshot of 10,050 scored men's fashion and complete analyses. The gender gap is 0.6 points and is not the finding. On exact labels, Smart Casual means 76.4 (n=433) against Casual 64.5 (n=1,104) and Casual/Loungewear 52.6 (n=112). Each step is 11.9 points. Exact strings, n-floor 100, no causal lift. Pitch this URL, not the rater.
The Context Gap
What 20,883 outfit scores reveal about worn looks vs individual pieces.
A frozen snapshot of 20,883 scored fashion and complete analyses. On exact labels, complete_outfit means 68.6 (n=13,469) against individual_pieces 57.4 (n=957) and flat_lay 64.1 (n=218). Worn minus pieces is 11.2 points. Pieces are not a 55–72 band. Exact strings, n-floor 100, no causal lift. Pitch this URL, not the rater.
The Colour Myth
What 972 fully-scored outfits reveal about which dimension actually decides the score.
A frozen snapshot of 972 fully-scored fashion and complete analyses — the rows where all five 0–20 rubric pillars reconcile with the headline score. Ranked by how far each pillar separates a top-quartile outfit from a bottom-quartile one, colour harmony is third of five (6.04). It is the weakest pillar on just 9.6% of outfits; details & quality is on 56.4%. Colour is the trend, not the lever. Frozen snapshot, no causal lift.
The Potential Gap
What 10,200 outfits reveal about the distance between the score and the ceiling.
A frozen snapshot of 10,200 fashion analyses carrying both a current score and a model-assigned potential. Mean current 63.2, mean potential 78.0, mean gap 14.9 points (median 15, SD 6.6). The gap is unimodal and shrinks as the score rises — 19.2 points below 60, 5.8 points at 85-plus, corr −0.58. Potential is not a fixed ceiling: it tracks the outfit (69.6 below 60, 94.4 at 85-plus). Only 0.3% of rows carry a negative gap. Frozen snapshot, no causal lift.
The Weakness Taxonomy
What 21,286 outfits reveal about what the model actually tells you to fix.
A frozen snapshot of 21,286 scored analyses carrying improvement advice. Rolled into twelve disclosed keyword families, the most common theme is accessories (45.2%), then styling (44.2%) and footwear (39.3%); colour is eighth (15.0%). The mode is three tips per outfit (59.6%), and the advice escalates with the score — below-60 rows led by styling and occasion, 85-plus rows by accessories and footwear. Frozen snapshot, no causal lift.
His & Hers
What 19,995 outfits reveal about the difference between male and female scores.
A frozen snapshot of 19,995 scored analyses the model labelled male (10,346) or female (9,649). The scores sit 0.4 points apart (67.2 vs 66.8) and every percentile is within two points. The divergence is in the language: male-labelled rows clear n ≥ 100 on five archetypes (all shared), female-labelled rows on eleven (romantic 74.0, classic 79.1), and the advice diverges on fit (45.7% male vs 22.4% female) and accessories (49.8% female vs 41.5% male). Frozen snapshot, no causal lift.
The Occasion Effect
What 21,778 outfits reveal about naming where you are going.
A frozen snapshot of 21,778 scored analyses split by whether the wearer named an occasion. The 18,843 named rows average 65.7; the 2,935 unnamed rows average 71.4 — a 5.7-point gap. The exact string you pick matters more than the family it belongs to: wedding guest climbs to 73.3 (32.6% at 85+), while daily office wear sits at 61.3 with 43.8% below 60. Frozen snapshot, no causal lift.
The Archetype Ladder
What 6,036 style-labelled outfits reveal about the model's own vocabulary.
A frozen snapshot of the 7,618 outfits scored in the window where the model writes an outfit_archetype (1 May – 14 October 2026); 6,036 carry a label. Ranked by the score the same model gave, twelve labels clear n ≥ 100 and span 22.7 points: classic leads (77.7, 32.8% at 85+), then romantic (73.7); loungewear trails (55.0, 63.5% below 60) and the model's failure label undefined is lowest of all (41.4, 90.5% below 60). The unlabelled remainder averages 60.1. Because label and score share an author, this maps the model's priors, not a style ranking. Frozen snapshot, no causal lift.
About OutfitScore Research
Reports are authored by Saad, founder of OutfitScore, drawing on the aggregated analysis corpus of the platform. New issues are published approximately quarterly. All charts, figures, and findings may be reproduced with attribution. The underlying scoring methodology — a five-dimension rubric applied via multimodal large language models — is documented in Report No. 01, Section 1.
For media inquiries, citation guidance, or to request the underlying dataset for academic or editorial purposes, contact [email protected].
