What 1308 Real AI-Scored Outfits Tell Us About How People Actually Dress

Most people overestimate their outfit by about ten points. That's the cleanest way I know to summarize what we've seen since launch. Tell someone the AI is going to score their look and they'll guess 85, maybe 88 if they're feeling generous about the shoes. The actual number usually lands somewhere in the low seventies. Not bad. Not great. The dressed equivalent of "I tried."

I run OutfitScore. I'm the only developer, and I look at our analytics most mornings before I look at my email. Over the last 30 days, the platform has scored 1308 real outfits across 5 analysis types — full looks, outfit-only, makeup, and accessories. The average overall score across everything is 63 out of 100. That number doesn't move much from month to month, which is itself interesting: people get dressed about as well in March as they do in October.

This page exists because I keep getting asked the same question — "what's a good score?" — and the answer depends entirely on what you're comparing yourself to. So instead of guessing, here's what the data actually says. I'll update these numbers as they shift. The window is always the last thirty days, recalculated overnight.

The score most people imagine isn't the one most people get

Here's the score distribution from the last 30 days. I think of it as the "what does dressed look like" curve, because every dot on it is a real outfit that someone wore today or last week or yesterday and decided was worth uploading.

The bell shape is real but it leans left. The middle of the distribution sits in the seventies, which the rubric calls "Good." Almost nobody scores below 40 — those outfits exist, but they don't usually get photographed and uploaded. People self-select. The interesting question isn't why so few outfits score low; it's why so few score high. Crossing 85 isn't a tiny improvement on 75. It's a different category of effort.

The 90+ tier is rare for a reason I'll come back to: it requires every one of five dimensions to score 18 or above, and most outfits don't even know which dimension is hurting them. That's the gap this page is trying to close.

Where outfits lose their points, ranked by the data

The model scores outfits across five dimensions, each out of 20: fit and proportion, color harmony, occasion appropriateness, styling and cohesion, and details and quality. Every outfit gets all five. The interesting part isn't which scores high in the abstract — it's which is dragging the average down right now.

Styling & Cohesion is the floor right now, averaging 13/20. Occasion Appropriateness is the ceiling at 16/20. That's a 3-point gap, which is significant. When one dimension is meaningfully behind the rest, fixing it is the highest-leverage move available. It's worth two or three points of overall score for almost no extra effort.

Styling and cohesion measures whether the outfit reads as one decision or five separate ones. When this scores low, the typical pattern is too many statement pieces competing for attention. Subtract a thing.

And Occasion Appropriateness leading the pack at 16/20 isn't a coincidence — it's usually the dimension people are already thinking about when they get dressed. The lesson worth taking away: the dimension you worry about least is usually the one that's hurting you most.

Occasion changes the game more than people expect

The same outfit doesn't score the same for every context. We tell the model what the occasion is before it scores, and the rubric weights things differently depending on whether you said "wedding" or "weekend brunch." So aggregating by occasion gives you a real read on which contexts are well-dressed and which aren't.

A pattern shows up in the data that I didn't expect when we launched: high-stakes occasions score higher than everyday ones. Job interviews and weddings consistently come in five to seven points above brunch and casual Friday. People take effort seriously when the context demands it. The interesting takeaway isn't that they dress well for big events. It's that the same person dressing for "Tuesday" produces an outfit that would have scored 78 on Saturday and produces a 71 on Tuesday. The effort cap is real, and most people are working under it on most days.

Date night is somewhere in the middle. It draws strong submissions but inconsistent execution. The high scorers nail one specific lever — usually a defined silhouette with a single statement piece — and the low scorers reach for too many things at once. A great date-night outfit, by the numbers, looks under-styled rather than over-styled. There's a lesson in that.

What 85+ outfits keep doing right

I went through several hundred of the top-scoring outfits looking for patterns. Some of what I found was obvious. Some wasn't.

Obvious: top scorers wear clothes that fit. That isn't tailoring as an aesthetic. It's tailoring as a baseline. The cuff hits the wrist bone. The trouser breaks cleanly. The shoulder seam meets the shoulder. None of this is hard. Most of it is one trip to a tailor and forty dollars.

Less obvious: 85+ outfits are almost always built around three or four pieces, not five or six. Restraint is doing more of the work than people realize. The score rewards intentionality, and intentionality reads as fewer choices made better. An expensive watch and a cheap belt scores worse than a decent watch and no belt. Empty surfaces give the eye somewhere to rest.

Also less obvious: footwear is the most disproportionate scorer of all five dimensions. A great pair of shoes can move an outfit four points on its own. The reverse is also true — flip-flops will tank the rubric on a look that's otherwise an 82. If you only fix one thing about your outfit before uploading it, fix the shoes.

And one I keep noticing: the very top of the distribution — the 90+ tier — almost always has one element that the model interprets as deliberate. A pocket square. A turned-up cuff. A blazer with the sleeves pushed slightly above the wrist. Stylists call this "an intentional choice." The AI does too, though it phrases it differently. Either way, you can't accidentally score 92. Someone made a decision.

The color season variable hiding in every score

Color harmony isn't a universal scale. A burgundy that scores beautifully on one person reads as flat on another, and the AI is calibrated to notice the difference. The reason isn't aesthetic — it's contrast math. Skin undertone, hair, and eye color set the baseline; the clothes either harmonize with that baseline or fight it.

Pattern-wise, the looks that lose the most color points usually aren't wearing "bad" colors. They're wearing colors that don't belong to the person wearing them. Warm earth tones on someone with cool undertones. Pure black on someone whose face needs softer contrast. Pure white on someone whose features would lift better with off-white. None of that shows up in a mirror. All of it shows up in the score.

This is the entire reason the Style DNA feature exists on the platform. Knowing your color season — the four-quadrant system that sorts people into Winter, Summer, Autumn, or Spring based on contrast and undertone — turns color harmony from a guessing game into a checklist. There's a separate color season guide for each one if you want to dig in.

How the score works, and why I trust it

Every outfit gets analyzed by Google's Gemini model, prompted with a fixed scoring rubric. The rubric has five dimensions, each scored out of 20 against a five-tier scale: 0–6 for severe issues, 7–10 below average, 11–14 average, 15–17 very good, 18–20 flawless. The five dimension scores sum to the 0–100 overall score. There's no bonus, no curve, no penalty for repeat users.

Two design choices keep the scores honest. First, the rubric is embedded directly in the prompt, not the system message, which means the model can't drift between scoring sessions. Second, we run with the temperature deliberately set to a moderate value — high enough that two different outfits get genuinely different scores, low enough that the same outfit scored twice lands within two points of itself. Score diversity matters. A scoring system that gives everyone 78 is useless.

The aggregate data on this page is pulled from the live analyses table with personally identifying information stripped out. No images. No faces. No usernames, emails, or device IDs. Just the score, the occasion the user selected, and the timestamp. The rolling window resets every twenty-four hours so the numbers stay current, and the page recompiles at most once an hour.

What to do with all of this before you next get dressed

If you've read this far you already care more about how your outfit lands than most people do. The data says you only need three things to lift a 72 into the low 80s. One: get the fit right. That's a tailor, not a wardrobe. Two: pick the occasion deliberately and dress for that one, not for "Tuesday." Three: subtract a piece. Almost any outfit gets better when you take something off.

And if you want a baseline before you start making changes — upload what you're wearing right now and see where you sit. The score takes about ten seconds. It'll tell you which of the five dimensions is hurting you most, which is more useful than every "tips" article on the internet because it's about you, not in general.

How scores break down by what you're analyzing

The platform handles four kinds of submissions: outfit-only, makeup-only, accessories-only, and the full "complete look" which gets all three at once. Each scores differently on average, and the gap between them tells you something about where the platform's standards bite hardest.

The complete-look submissions sit lower than outfit-only on average, and that's the expected pattern. A complete look has three things going on instead of one, so the chance that all three are clicking simultaneously is mathematically smaller. The makeup-only category behaves differently — different dimensions, different ceiling, and worth a separate page entirely. If you came here for that, the makeup trends report is where to go next.

Accessories-only is the smallest sample and the most volatile. The dimensions are stricter (you can't hide a bad watch behind a great shirt), so the curve runs wider than the others. People who lean heavily on accessories either nail it or don't, and the data shows that almost clinically.

Common questions