Back to the trend of the day: Many Singles' Day women's clothes pages show computer-made models, not real people (in Chinese)
Poly Rating · from the trend of the day

High Return Rate Products 2026: Women's Clothes vs AI Models

China has required labels on AI-made images since September 2025, yet in October 2026 reporters still found headless and three-armed AI models selling women's clothes, a category where 50–60% of online orders come back.

  • No. 1: Silk and satin slip dresses
  • 10 places

01Final list

The leading AIs each built their own list of ten independently. Here is one result that is easy to check: every place has a reason, a source, and the list of AIs that included it.

Order: from most to least AI agreement.

A bright badge means this AI put the entry on its list. A dim badge means it did not.

  1. 1

    Silk and satin slip dresses

    Rendered sheen and liquid drape are exactly what cheap satin fails to deliver. No slip-dress return rate is published; women's clothing overall runs at 50–60%.

    Included it: Claude, ChatGPT, Meta, Gemini, Grok, DeepSeek, Wenxin, Kimi. Did not include: Nemotron, StepFun.

    In their lists: Claude: 2 · ChatGPT: 1 · Meta: 3 · Gemini: 2 · Grok: 1 · DeepSeek: 1 · Wenxin: 2 · Kimi: 1

    What each AI noted 8
    • Claudeplace 2

      A render shows a flawless liquid drape, but the real fabric clings, creases and shows its sheen differently.

      Source ↗
    • ChatGPTplace 1

      The same dress can look liquid silk on-screen and turn out stiff, shiny polyester in the parcel.

      Source ↗
    • Metaplace 3

      AI sees silk as glossy pixel not 12-15% elongation filament; bias-cut slips fail AI try-on consistently

      Source ↗
    • Geminiplace 2

      Generative rendering simulates an unbroken liquid sheen, while buyers open packages to find static cling, raw seams, and polyester sheen that prompt near-70% return waves.

      Source ↗
    • Grokplace 1

      AI can invent perfect drape and sheen that real fabric rarely delivers on arrival

      Source ↗
    • DeepSeekplace 1

      The fabric that AI flatters most is the one that lies: liquid drape and sheen are easy to render, almost impossible to guarantee in a parcel.

      Source ↗
    • Wenxinplace 2

      AI renders fake silk sheen 100% inaccurately, driving 28-32% return rate for budget-tier listings

    • Kimiplace 1

      AI gives polyester the luster of top-grade silk - the sheen is free, the drape is not.

      Source ↗
    8 AIs put it here
  2. 2

    Tailored blazers

    Shoulder line, canvas and lapel roll decide how a blazer fits, and an AI image smooths over all three. Reporters even found an AI model whose back was still its front.

    Included it: Claude, ChatGPT, Meta, Gemini, Grok, DeepSeek, Wenxin, Kimi. Did not include: Nemotron, StepFun.

    In their lists: Claude: 3 · ChatGPT: 2 · Meta: 2 · Gemini: 3 · Grok: 2 · DeepSeek: 2 · Wenxin: 3 · Kimi: 4

    What each AI noted 8
    • Claudeplace 3

      The fit depends on shoulder line, lapel roll and stiffness, and an AI model's perfect fit hides all three.

      Source ↗
    • ChatGPTplace 2

      A convincing digital shoulder is easy to draw; a shoulder that actually fits is harder to deliver.

      Source ↗
    • Metaplace 2

      Drape coefficient under 0.3 required for jacket; AI adds impossible folds with no anchor at shoulder

      Source ↗
    • Geminiplace 3

      Diffusion models hide boxy shoulder pads and misaligned lapel roll-lines behind mathematically smoothed virtual postures, driving fit-mismatch refunds above 58%.

      Source ↗
    • Grokplace 2

      Shoulder line and stiffness that AI renders crisp often collapse in real cloth

    • DeepSeekplace 2

      A blazer is a building: shoulder line, canvas and weight decide whether it stands or collapses — and an AI model never has to carry the weight.

      Source ↗
    • Wenxinplace 3

      AI erases 100% of bad tailoring flaws, 62% of buyers report delivered items are baggier than shown

    • Kimiplace 4

      A blazer is shoulder padding, canvas and drape - and a computer-made model has perfect shoulders every single time.

      Source ↗
    8 AIs put it here
  3. 3

    Wool and cashmere coats

    A coat's look comes from its weight and structure. One fur-and-cashmere coat shop hit 80% returns, but that is one shop, not a rate for the category.

    Included it: Claude, ChatGPT, Meta, Gemini, Grok, StepFun, DeepSeek, Kimi. Did not include: Nemotron, Wenxin.

    In their lists: Claude: 1 · ChatGPT: 3 · Meta: 1 · Gemini: 9 · Grok: 5 · StepFun: 1 · DeepSeek: see list · Kimi: 2

    What each AI noted 8
    • Claudeplace 1

      A coat's look comes from fabric weight and shoulder structure, which a render can fake and a parcel cannot.

      Source ↗
    • ChatGPTplace 3

      A virtual model can wear a perfect winter coat without ever having to carry its weight.

      Source ↗
    • Metaplace 1

      Needs 400-800 GSM and body to hold crisp silhouette — AI renders flat sheen; 75-80% of Double 11 wool coats sent back vs 60% last year

      Source ↗
    • Geminiplace 9

      A coat that lives or dies on fabric drape is flattened by AI generation into unwrinkled perfection, concealing the stiff cardboard hand-feel that repels buyers.

      Source ↗
    • Grokplace 5

      Drape and structure AI invents often feel stiff or thin when the parcel opens

    • StepFunplace 1

      Women's clothes came back at 50-60% last year while close to 80% of sellers showed AI-made models - and a coat is mostly shoulder line, cloth weight and how it falls.

    • DeepSeek

      Named differently in its own list: open the full list.

      see list
    • Kimiplace 2

      The single highest shop-level return rate reported by any supplier: 80%, from a fur-and-cashmere coat specialist.

      Source ↗
    8 AIs put it here
  4. 4

    Cashmere and wool sweaters

    50,000 wool and cashmere sweater complaints in 2024, up 12.3%. Knit gauge, fibre content and pilling cannot be seen in any image.

    Included it: Claude, Meta, Gemini, StepFun, DeepSeek, Wenxin, Kimi. Did not include: ChatGPT, Nemotron, Grok.

    In their lists: Claude: 6 · Meta: 8 · Gemini: 1 · StepFun: 6 · DeepSeek: see list · Wenxin: see list · Kimi: 7

    What each AI noted 7
    • Claudeplace 6

      Knit thickness, pilling and stretch are invisible in a render, yet they decide whether it is a sweater or a thin top.

      Source ↗
    • Metaplace 8

      Pilling and weight invisible to AI – shoppers cite fabric quality difference as top complaint

      Source ↗
    • Geminiplace 1

      AI algorithms render seamless, weightless perfection on digital torsos, but real overcoats arrive stiff, creased, and see return rates climb beyond 60% due to unmatchable fabric density.

      Source ↗
    • StepFunplace 6

      Reviewers' standing gripe is that model photos hide fit flaws and colour difference - gauge, weight and pilling are what a smooth render erases.

      Source ↗
    • DeepSeek

      Named differently in its own list: open the full list.

      see list
    • Wenxin

      Named differently in its own list: open the full list.

      see list
    • Kimiplace 7

      50,000 complaints a year about sweaters whose 'cashmere' exists only in the photo's caption.

      Source ↗
    7 AIs put it here
  5. 5

    Down jackets and puffer coats

    133,000 complaints in 2024, up 40.3%, the fastest-rising clothing complaint. Most are about hidden fill, which no photo, real or AI-made, can show.

    Included it: Claude, ChatGPT, Meta, Gemini, StepFun, DeepSeek, Kimi. Did not include: Nemotron, Grok, Wenxin.

    In their lists: Claude: 9 · ChatGPT: 9 · Meta: 6 · Gemini: 4 · StepFun: 3 · DeepSeek: 9 · Kimi: 3

    What each AI noted 7
    • Claudeplace 9

      Puffiness and fill look the same in a render, so the gap shows up in warmth and bulk, not in the picture.

      Source ↗
    • ChatGPTplace 9

      A digital puffer can promise cloud-like volume without revealing the actual filling or warmth.

      Source ↗
    • Metaplace 6

      Down jacket + black silk = this winter's highest return rate outfit per Laitimes – loft vs AI smoothness

      Source ↗
    • Geminiplace 4

      Virtual renderings mask uneven baffle fill and crinkly shell noises, leading to high-friction returns as parcels arrive either paper-thin or overwhelmingly bulky.

      Source ↗
    • StepFunplace 3

      The CCA made down jackets a named complaint hotspot in 2024, the year clothing complaints jumped 49.96% - fill power and warmth are claims, not pixels.

      Source ↗
    • DeepSeekplace 9

      Volume is easy to exaggerate and hard to verify: a puffer that looks cloud-like on an AI model can arrive flat, and the return rate for coats is already 28–34%.

      Source ↗
    • Kimiplace 3

      133,000 complaints in one year, up 40.3% - the fastest-rising clothing complaint in China, all about fill you cannot see in a picture.

      Source ↗
    7 AIs put it here
  6. 6

    Pleated skirts

    A rendered pleat never falls flat. Reporters found AI sale pages where even the folds of the clothes were identical on every model. No skirt return rate is published.

    Included it: Claude, ChatGPT, Gemini, Grok, StepFun, DeepSeek. Did not include: Meta, Nemotron, Wenxin, Kimi.

    In their lists: Claude: 7 · ChatGPT: 4 · Gemini: 10 · Grok: 7 · StepFun: 8 · DeepSeek: 6

    What each AI noted 6
    • Claudeplace 7

      A pleat holds its line in a render and often collapses in real fabric.

      Source ↗
    • ChatGPTplace 4

      A perfect digital pleat can stay perfect forever; a real pleat has to survive gravity, movement and packing.

      Source ↗
    • Geminiplace 10

      Algorithmic image rendering fixes symmetrical accordion pleats in permanent static perfection, while cheap polyester heat-sets flatten into uneven ruffles upon opening.

      Source ↗
    • Grokplace 7

      Pleat sharpness and hang that look permanent on screen flatten in transit

      Source ↗
    • StepFunplace 8

      A pleat is a physical fact - set, sharpness, swing - and reporters found the AI pages coming out with identical folds on every model.

    • DeepSeekplace 6

      A pleat is a promise about memory and weight: AI can draw the fold, but not the fabric's willingness to hold it.

      Source ↗
    6 AIs put it here
  7. 7

    Wide-leg and tailored trousers

    Crease and hang make the garment, and price wars push sellers to cheaper fabric that hangs unlike any photo. The 20–24% trouser return benchmark is US data, not Chinese.

    Included it: Claude, ChatGPT, Meta, Grok, DeepSeek, Kimi. Did not include: Gemini, Nemotron, StepFun, Wenxin.

    In their lists: Claude: 8 · ChatGPT: 6 · Meta: 5 · Grok: 9 · DeepSeek: 8 · Kimi: 8

    What each AI noted 6
    • Claudeplace 8

      The break, rise and fall of the leg depend on fabric weight, which a computer-made image cannot test.

      Source ↗
    • ChatGPTplace 6

      One extra crease or two centimeters of rise can change the entire look of a pair of trousers.

      Source ↗
    • Metaplace 5

      Fit depends on drape and tailoring at waist and hip; 70% of returns cite fit as reason

      Source ↗
    • Grokplace 9

      Leg line and fabric fall AI idealises rarely match real drape or length

      Source ↗
    • DeepSeekplace 8

      The difference between a fluid wide leg and a stiff tube is invisible in a still image — and return data for bottoms stays stubbornly high.

      Source ↗
    • Kimiplace 8

      The crease and the hang are the whole garment - and both are exactly what a render invents.

      Source ↗
    6 AIs put it here
  8. 8

    Rigid and skinny jeans

    Stiffness, wash and shop-by-shop size labels drive returns. Peak-season US denim returns topped 51%, but only one source gives that figure.

    Included it: Meta, Gemini, StepFun, DeepSeek, Kimi. Did not include: Claude, ChatGPT, Nemotron, Grok, Wenxin.

    In their lists: Meta: 9 · Gemini: 7 · StepFun: 7 · DeepSeek: 4 · Kimi: 9

    What each AI noted 5
    • Metaplace 9

      Sizing chaos – can fit children's size but not women's same size, drives 70% returns for fit

    • Geminiplace 7

      AI mannequins never show crotch whiskering or stiff hip bunching, inducing shoppers to multi-size order only to send back 55% of non-stretch denim.

      Source ↗
    • StepFunplace 7

      Jeans are bought for stiffness, wash and colour - and denim suppliers themselves say returns track how far the photo matches the real cloth.

    • DeepSeekplace 4

      Denim is where the picture lies twice: once about stiffness, once about fit — and return rates above 50% in some peak windows show it.

    • Kimiplace 9

      Denim returns less than dresses, but 'L码 means whatever this shop says it means' makes fit a lottery.

      Source ↗
    5 AIs put it here
  9. 9

    Fitted and A-line dresses

    In 2024 one Guangzhou dress factory said it sold 1,000 dresses a day and got 800 back. That is one factory's figure; waist and length are what a render flatters most.

    Included it: Claude, Grok, StepFun, Kimi. Did not include: ChatGPT, Meta, Gemini, Nemotron, DeepSeek, Wenxin.

    In their lists: Claude: 4 · Grok: 3 · StepFun: 5 · Kimi: 6

    What each AI noted 4
    • Claudeplace 4

      Lightweight fabric floats in a render but hangs flat or clings on a real body.

      Source ↗
    • Grokplace 3

      Proportion and hem hang that look ideal on AI models fail real body ratios

      Source ↗
    • StepFunplace 5

      Sizing explains about 60% of returns - the other 40% is waist, length and drape, exactly what a render is kindest to.

      Source ↗
    • Kimiplace 6

      A Guangzhou dress factory's rule of thumb in 2024: sell 1,000, get 800 back.

      Source ↗
    4 AIs put it here
  10. 10

    Hanfu and 'new Chinese style' sets

    The only type with a measured picture-mismatch signal: complaints rose 70% in 2024, mostly about goods not matching their pictures. One top hanfu brand's returns reached 80%.

    Included it: Meta, Gemini, Wenxin, Kimi. Did not include: Claude, ChatGPT, Nemotron, Grok, StepFun, DeepSeek.

    In their lists: Meta: 7 · Gemini: see list · Wenxin: 6 · Kimi: 5

    What each AI noted 4
    • Metaplace 7

      Maillard and new Chinese style flagged as two of three no-touch categories for high returns

      Source ↗
    • Gemini

      Named differently in its own list: open the full list.

      see list
    • Wenxinplace 6

      Over 60% of AI product pages alter set proportions, 57% of consumers report top and bottom sizing mismatch

    • Kimiplace 5

      The style whose complaints jumped 70% in a year - concentrated on goods that did not match the picture.

      Source ↗
    4 AIs put it here

More about the top three ↓

02More about the top three

The ranking is above. Here is more about the three places at the top: what they are, why they are here, and what is worth knowing.

The figures are not on the same base: 50–60% is all women's clothing online, while the other numbers come from single shops or wholesale and are not category rates.

1Silk and satin slip dresses

No. 1 — Silk and satin slip dresses. On the lists of 8 of 10 AIs; median place there: 1.5. Fake silk sheen and liquid drape are easy to render, but the real fabric arrives clingy, creased and shiny polyester.

Did you know

  • The slip dress came back in a big way in 2016. Yves Saint Laurent and Burberry both showed it in their spring 2016 collections. Versace added slip dresses with a 'beachy theme' in spring 2021.

What it is
Light, glossy dresses made of silk or satin that hang loose on the body.
Its edge
Its sheen and drape are the easiest things for AI to fake.
Who should know
Useful for anyone buying dresses online who wonders why the photo and the parcel differ.
Why it is here
  • No slip-dress return rate is published. For women's clothing overall, a supplier to Douyin livestream sellers said wholesale returns rose to 75–80% this year from 60% last year. Cheap satin that misses its glossy photo adds to that risk.

  • Fake sheen is cheap to make. A human model costs 1,400–1,600 yuan an hour, and an eight-hour shoot usually runs about 36,000 yuan. Alibaba's AI tool Taji is free. This is clothing in general, not dresses alone.

Did you know
  • In the 1990s, slip dresses were often worn over a white tee or bodysuit, with opaque tights, leggings or even bike shorts. The shorter length was the most popular, so the 'underwear' look was often layered.

History
  • In 1914 the First World War forced Madeleine Vionnet to close her fashion house. She reopened in 1923 on Avenue Montaigne, and the building became known as the "Temple of Fashion."

Where to go further
  • Read this momme guide to learn that momme measures how dense the silk is, not how good it is. Check the fiber label first, then compare weights.

Compare the top three ↓

2Tailored blazers

No. 2 — Tailored blazers. On the lists of 8 of 10 AIs; median place there: 2.5. An AI image smooths over shoulder line, canvas and lapel roll, which decide how a blazer fits, so the real one disappoints.

Did you know

  • The rowing club that claims the word says 'blazer' came from its bright red jackets. The Lady Margaret Boat Club of Cambridge says the name came from that vivid red cloth. The earliest known printed use is 22 August 1889.

What it is
A structured jacket with padded shoulders, stiff inner cloth and shaped lapels.
Its edge
Its fit depends on hidden structure that a perfect AI model never shows.
Who should know
Useful for anyone who buys jackets online and wants to know why fit goes wrong.
Why it is here
  • No blazer-only rate exists. Chinese women's clothing returns went from about 20% before 2022 to about 70% at one Taobao shop. Insiders blame ill-fitting clothes and items that don't match the photos, two things a blazer is judged on.

  • Fixing a blazer's look in software is cheap. A clothing e-commerce company claimed AI seller photos cut its costs by more than 90%, per a headline listed by OECD.AI. Buyers report goods that differ greatly from the AI visuals.

History
  • In 1966 Yves Saint Laurent showed his women's tuxedo jacket in his couture show, and his couture clients rejected it. Only one jacket sold. His younger ready-to-wear buyers then made the jacket a hit.

Where to go further
  • Read this guide to Savile Row, then walk the street if you can. Look first at the shoulders and lapels. Row suits often use more shoulder padding and canvas, with straighter lines.

Compare the top three ↓

3Wool and cashmere coats

No. 3 — Wool and cashmere coats. On the lists of 8 of 10 AIs; median place there: 2. A render can fake a coat's weight and structure, but the real cloth feels thin or stiff when the parcel opens.

Did you know

  • A cashmere goat gives only about 150 grams (0.33 lb) of fibre a year on average. That is a big reason cashmere coats cost so much.

What it is
Heavy winter coats made of wool or cashmere, where weight and cloth quality matter.
Its edge
Its weight can't be shown in a flat picture, though it ranks below the other two.
Who should know
Useful for anyone shopping for a winter coat online who wants to avoid a letdown.
Why it is here
  • Single shops show how bad it gets on Singles' Day. One Xiaohongshu seller reported 10,000 of 20,000 parcels sent back, and shoppers said items didn't match the photos. These are anecdotes, not a coat-category rate.

History
  • In 1990 production at Crombie's Grandholm Mill in Aberdeen stopped, and the work moved to other mills in Scotland and England. In March 2022 the Crombie name passed to new owners, Crombie 1805.

Where to go further
  • Read this guide to overcoating cloths. Check the weight in ounces first; it suggests 20–25oz for a first coat. Then feel the cloth, because how dense it is matters as much as the weight.

Claude, Meta, StepFun put it first. What they saw in it
  • Claude: A coat's look comes from fabric weight and shoulder structure, which a render can fake and a parcel cannot.

  • Meta: Needs 400-800 GSM and body to hold crisp silhouette — AI renders flat sheen; 75-80% of Double 11 wool coats sent back vs 60% last year

  • StepFun: Women's clothes came back at 50-60% last year while close to 80% of sellers showed AI-made models - and a coat is mostly shoulder line, cloth weight and how it falls.

Compare the top three ↓

The top three side by side

Silk and satin slip dressesTailored blazersWool and cashmere coats
In the AIs' lists8 / 108 / 108 / 10
Put first byChatGPT, Grok, DeepSeek, Kimi—Claude, Meta, StepFun

Places with the same number of AIs are ordered by their average place in the AIs' own lists.

03Worth knowing

What the AIs noticed while building the list.

Plain T-shirts, which look the same in any photo, still come back a lot: one Guangzhou seller reported about 60%. So size and quality drive many returns, not only the picture.

Noticed by:
Included it: Claude, Kimi. Did not include: ChatGPT, Meta, Gemini, Nemotron, Grok, StepFun, DeepSeek, Wenxin.
Source ↗

The 'nearly 80% of sellers use AI' headline comes from 76.95% of surveyed merchants using any AI-made material. About 35% of clothing sellers, and roughly 40% on Taobao, use AI models or AI try-on.

Noticed by:
Included it: StepFun, DeepSeek. Did not include: Claude, ChatGPT, Meta, Gemini, Nemotron, Grok, Wenxin, Kimi.
Source ↗

One report attributes 57.32% of livestream clothing returns to items looking different from the screen. It is a single source and has not been confirmed elsewhere.

Noticed by:
Included it: DeepSeek. Did not include: Claude, ChatGPT, Meta, Gemini, Nemotron, Grok, StepFun, Wenxin, Kimi.
Source ↗

Women's clothing return figures disagree: 50–60% (CCTV and 36Kr), 65–80% (100EC database) and 80–90% in livestream sales. They measure different channels and periods, so they should not be averaged.

Noticed by:
Included it: ChatGPT, Meta, Grok, StepFun, Kimi. Did not include: Claude, Gemini, Nemotron, DeepSeek, Wenxin.
Source ↗

China's AI-content labelling rules have been in force since 1 September 2025, with duties on platforms, yet unlabelled AI models were still on Singles' Day pages in October 2026.

Noticed by:
Included it: ChatGPT, StepFun. Did not include: Claude, Meta, Gemini, Nemotron, Grok, DeepSeek, Wenxin, Kimi.
Source ↗

04How we counted

In plain words, no formulas.

1The question

Top 10 Women's Clothing Types Most Likely to Look Different From a Computer-Made Model

2Why this topic

The reader learns which pieces of clothing are most worth doubting before buying, which comes from putting complaint data, return rates and fabric behaviour side by side. No single return-rate table shows this.

3Where it applies

Women's clothing sold online in China around Singles' Day (November 11)

4What we looked at

  • Return and refund rate for each clothing type during big sale periods, as reported by Chinese e-commerce researchers
  • Share of consumer complaints for each type that say the item did not match its picture
  • How much the look depends on fabric weight, stiffness, drape and tailoring, which a computer-made image can fake (an editorial judgment backed by tailoring and textile sources)
  • How often each type is shown on computer-made models on sale pages
  • 100EC (网经社) and 36Kr Research Institute, via People's Daily — Return rates for women's clothing: 50–60% online, above 80% in livestream sales (Q4 2024)
  • CCTV, via Beijing Daily — Women's online clothing returns of 50–60%, and above 80% for livestream clothing sales on Double 11 2025
  • State Administration for Market Regulation (SAMR) complaint data, via CCTV — Complaint counts and growth by clothing type, including complaints that goods did not match their pictures
  • China Consumers Association (CCA) 2024 complaint analysis — Clothing complaints up 49.96%; false advertising made up 6.6% of all complaints

5Who ranked

The leading AIs, each building its own list of ten independently and naming a source for every place: Claude, ChatGPT, Meta, Gemini, Nemotron, Grok, StepFun, DeepSeek, Wenxin, Kimi.

6How the places were set

The same entry written differently by different AIs was counted as one. The more AIs put an entry on their list, and the higher they put it, the higher its place. "AI agreement" shows how many of the AIs had it on their own list. Sources are only those the AIs found themselves; nothing was added by hand.

05Each AI's own list

🇺🇸 Claude10 places · 3 notes

The same question, answered by Claude alone. The final list above combines all the AIs.

  1. 1
    Structured wool and tailored coats (羊毛大衣)

    A coat's look comes from fabric weight and shoulder structure, which a render can fake and a parcel cannot.

    Source ↗
  2. 2
    Silk and satin slip dresses (真丝吊带裙)

    A render shows a flawless liquid drape, but the real fabric clings, creases and shows its sheen differently.

    Source ↗
  3. 3
    Tailored blazers and suit jackets (西装外套)

    The fit depends on shoulder line, lapel roll and stiffness, and an AI model's perfect fit hides all three.

    Source ↗
  4. 4
    Chiffon and A-line dresses (雪纺/A字连衣裙)

    Lightweight fabric floats in a render but hangs flat or clings on a real body.

    Source ↗
  5. 5
    Faux-leather and leather jackets

    Stiffness and surface sheen are the first things a render gets wrong, and they decide whether the jacket looks premium or plastic.

    Source ↗
  6. 6
    Knitwear and sweaters (针织衫/毛衣)

    Knit thickness, pilling and stretch are invisible in a render, yet they decide whether it is a sweater or a thin top.

    Source ↗
  7. 7
    Pleated and flowing skirts (百褶/半身裙)

    A pleat holds its line in a render and often collapses in real fabric.

    Source ↗
  8. 8
    Wide-leg tailored trousers

    The break, rise and fall of the leg depend on fabric weight, which a computer-made image cannot test.

    Source ↗
  9. 9
    Down and puffer jackets (羽绒服)

    Puffiness and fill look the same in a render, so the gap shows up in warmth and bulk, not in the picture.

    Source ↗
  10. 10
    Basic T-shirts and loose hoodies

    They look much the same in any photo, yet they return a lot: one Guangzhou seller reported about 60%, which shows the cause is size and quality, not the picture.

    Source ↗

What Claude noticed

Women's clothing is the highest-return category in China: 50–60% (36Kr) and above 80% for livestream sales (网经社, Q4 2024).

The simplest garments can still return a lot: a Guangzhou T-shirt seller reported about 60%, so the picture is not the only cause of returns.

China's AI-content labelling measures are already cited in commentary calling for AI-made images to be marked.

🇺🇸 ChatGPT10 places · 4 notes

The same question, answered by ChatGPT alone. The final list above combines all the AIs.

  1. 1
    Women's silk slip dresses and satin slip dresses

    The same dress can look liquid silk on-screen and turn out stiff, shiny polyester in the parcel.

    Source ↗
  2. 2
    Women's tailored blazers

    A convincing digital shoulder is easy to draw; a shoulder that actually fits is harder to deliver.

    Source ↗
  3. 3
    Women's wool coats and tailored trench coats

    A virtual model can wear a perfect winter coat without ever having to carry its weight.

    Source ↗
  4. 4
    Women's pleated skirts and structured midi skirts

    A perfect digital pleat can stay perfect forever; a real pleat has to survive gravity, movement and packing.

    Source ↗
  5. 5
    Women's fitted knit dresses and bodycon dresses

    The digital waist never stretches out, and the digital fabric never has to recover after sitting down.

    Source ↗
  6. 6
    Women's wide-leg trousers and tailored pants

    One extra crease or two centimeters of rise can change the entire look of a pair of trousers.

    Source ↗
  7. 7
    Women's satin blouses and sheer chiffon tops

    A screen can make sheer fabric look luxurious without showing what is visible underneath.

    Source ↗
  8. 8
    Women's faux-leather jackets and leather-look skirts

    The screen can imitate a leather shine; it cannot tell you whether the real material feels like plastic.

    Source ↗
  9. 9
    Women's puffer jackets and quilted outerwear

    A digital puffer can promise cloud-like volume without revealing the actual filling or warmth.

    Source ↗
  10. 10
    Women's basic T-shirts and loose cotton hoodies

    The simplest garment is the hardest to distort dramatically—but the wrong cotton weight can still disappoint.

    Source ↗

What ChatGPT noticed

The return-rate headline is real, but its denominator matters. CCTV reporting puts women's online apparel returns at 50–60% generally and livestream apparel above 80% during Double 11 2025. Another 100EC-cited report gives 65–80% for women's apparel and 80–90% for livestreaming. These estimates should not be averaged as if they measured the same merchants, orders and period.

The most visually vulnerable clothes are not necessarily the most frequently returned. There is no reliable public dataset linking AI-model use, image-mismatch complaints and return rates for individual garment types. The ranking therefore combines industry-wide return evidence with construction-based judgments instead of inventing garment-specific percentages.

AI-generated images are appearing beyond product pages. Chinese reporting has described generated pictures in customer reviews as well as product listings, weakening a familiar way shoppers check what a garment really looks like.

China's AI-content labeling regime took effect on September 1, 2025, requiring visible and technical identification in relevant circumstances. The existence of that rule does not establish that all apparel listings visibly disclose AI use or that enforcement has eliminated misleading imagery.

🇺🇸 Meta10 places · 4 notes

The same question, answered by Meta alone. The final list above combines all the AIs.

  1. 1
    Women's Wool Overcoat / Long Wool Coat

    Needs 400-800 GSM and body to hold crisp silhouette — AI renders flat sheen; 75-80% of Double 11 wool coats sent back vs 60% last year

    Source ↗
  2. 2
    Tailored Blazer

    Drape coefficient under 0.3 required for jacket; AI adds impossible folds with no anchor at shoulder

    Source ↗
  3. 3
    Silk Slip Dress / Satin Dress

    AI sees silk as glossy pixel not 12-15% elongation filament; bias-cut slips fail AI try-on consistently

    Source ↗
  4. 4
    Bodycon Hot Girl Dress / Spicy Girl Outfit

    Industry insiders say don't touch hot girl outfits – verified highest return of three risky categories

    Source ↗
  5. 5
    Women's Suit Pants / Tailored Trousers

    Fit depends on drape and tailoring at waist and hip; 70% of returns cite fit as reason

    Source ↗
  6. 6
    Down Jacket / Puffer Jacket

    Down jacket + black silk = this winter's highest return rate outfit per Laitimes – loft vs AI smoothness

    Source ↗
  7. 7
    New Chinese Style Qipao / Cheongsam Dress

    Maillard and new Chinese style flagged as two of three no-touch categories for high returns

    Source ↗
  8. 8
    Knitted Cashmere Sweater / Wool Sweater

    Pilling and weight invisible to AI – shoppers cite fabric quality difference as top complaint

    Source ↗
  9. 9
    Skinny Jeans / Women's Jeans

    Sizing chaos – can fit children's size but not women's same size, drives 70% returns for fit

  10. 10
    Faux Leather Jacket / Leather Jacket

    AI renders leather as plastic sheen without stiffness recovery – look different after first wear

    Source ↗

What Meta noticed

50-60% is the new normal for women's online returns, 80% for live-stream, up to 90% extreme – China Garment Association says each return costs ∼16 yuan in logistics

AI sees silk as a glossy pixel not a 12-15% elongation filament – Alibaba notes bias-cut silk slips consistently fail AI try-on because drape is choreography not geometry

Oversized A4 hangtags stamped 'no refund once removed' went viral with 40M+ Weibo views during Double 11 2025 to block wearing-then-returning

Hot girl outfits, Maillard brown and new Chinese style are the three categories insiders say not to touch – verified highest returns

🇺🇸 Gemini10 places · 2 notes

The same question, answered by Gemini alone. The final list above combines all the AIs.

  1. 1
    Tailored Wool and Cashmere Overcoats (China, Double 11)

    AI algorithms render seamless, weightless perfection on digital torsos, but real overcoats arrive stiff, creased, and see return rates climb beyond 60% due to unmatchable fabric density.

    Source ↗
  2. 2
    Silk Slip and Bias-Cut Satin Dresses (China, Double 11)

    Generative rendering simulates an unbroken liquid sheen, while buyers open packages to find static cling, raw seams, and polyester sheen that prompt near-70% return waves.

    Source ↗
  3. 3
    Structured Shoulder-Padded Blazers (China, Double 11)

    Diffusion models hide boxy shoulder pads and misaligned lapel roll-lines behind mathematically smoothed virtual postures, driving fit-mismatch refunds above 58%.

    Source ↗
  4. 4
    Voluminous Down Parkas and Puffers (China, Double 11)

    Virtual renderings mask uneven baffle fill and crinkly shell noises, leading to high-friction returns as parcels arrive either paper-thin or overwhelmingly bulky.

    Source ↗
  5. 5
    Modernized Cheongsam and Qipao (China, Double 11)

    A cut that demands millimeter-precise torso contours meets synthetic images that smooth away every tension crease, sparking 65% returns for puckering zippers.

    Source ↗
  6. 6
    Banquet Gowns and Cocktail Dresses (China, Double 11)

    Heavily generated AI lookbooks fake multi-layered bodice boning and tulle spring, creating an expectation gulf that fuels single-use wear returns exceeding 70%.

    Source ↗
  7. 7
    Rigid High-Waisted Denim Jeans (China, Double 11)

    AI mannequins never show crotch whiskering or stiff hip bunching, inducing shoppers to multi-size order only to send back 55% of non-stretch denim.

    Source ↗
  8. 8
    Mohair and Chunky Cable-Knit Cardigans (China, Double 11)

    Computer-generated pixels paint cloud-soft yarn halos, leaving shoppers with prickly synthetic blends and stretched knit openings that yield a 52% bounce-back.

    Source ↗
  9. 9
    Double-Breasted Cotton Trench Coats (China, Double 11)

    A coat that lives or dies on fabric drape is flattened by AI generation into unwrinkled perfection, concealing the stiff cardboard hand-feel that repels buyers.

    Source ↗
  10. 10
    Pleated Accordion Midi Skirts (China, Double 11)

    Algorithmic image rendering fixes symmetrical accordion pleats in permanent static perfection, while cheap polyester heat-sets flatten into uneven ruffles upon opening.

    Source ↗

What Gemini noticed

Producing a live photoshoot with human models costs Chinese e-commerce merchants 2,000 to 10,000 RMB a day, whereas generating unlimited AI model variations costs under 0.50 RMB per image, accelerating widespread adoption on Singles' Day.

While women's clothing returns hover around 50% to 60% during normal months, Double 11 cart promotions and automated shipping insurance push return rates past 80% on live-streamed synthetic product showcases.

🇺🇸 Grok10 places · 3 notes

The same question, answered by Grok alone. The final list above combines all the AIs.

  1. 1
    Silk or satin slip dresses

    AI can invent perfect drape and sheen that real fabric rarely delivers on arrival

    Source ↗
  2. 2
    Structured blazers and tailored jackets

    Shoulder line and stiffness that AI renders crisp often collapse in real cloth

  3. 3
    A-line and midi dresses (non-stretch)

    Proportion and hem hang that look ideal on AI models fail real body ratios

    Source ↗
  4. 4
    Chiffon or lightweight blouses

    Transparency and float that AI perfects rarely survive real fabric weight

    Source ↗
  5. 5
    Wool or heavy coats

    Drape and structure AI invents often feel stiff or thin when the parcel opens

  6. 6
    Bodycon or fitted knits

    AI smooths stretch and cling that real yarn fails to hold

    Source ↗
  7. 7
    Pleated or structured skirts

    Pleat sharpness and hang that look permanent on screen flatten in transit

    Source ↗
  8. 8
    Lace or detailed overlay tops

    Intricate texture AI renders crisply often looks sparse or cheap in hand

  9. 9
    Wide-leg or tailored trousers

    Leg line and fabric fall AI idealises rarely match real drape or length

    Source ↗
  10. 10
    Basic T-shirts and loose hoodies

    Simplest shapes still suffer colour and fabric feel gaps, but least dependent on AI flattery

    Source ↗

What Grok noticed

Women's apparel return rates on Chinese platforms climbed from ~30% in 2019 to 65-80% by 2025, with live-stream peaks at 80-90%.

Computer-made models are now common on Singles' Day women's pages; shoppers report the gap between picture and parcel is widest for fabric-dependent styles.

CCA data places clothing at the top of specific-goods complaints, driven by quality, fit and description mismatches rather than pure price issues.

🇨🇳 StepFun10 places · 4 notes

The same question, answered by StepFun alone. The final list above combines all the AIs.

  1. 1
    Structured wool overcoat (women's, Double 11)

    Women's clothes came back at 50-60% last year while close to 80% of sellers showed AI-made models - and a coat is mostly shoulder line, cloth weight and how it falls.

  2. 2
    Silk and satin slip dress (women's)

    Global benchmarks put dresses and swimwear at roughly three times fine jewellery's return rate - pure drape, nothing to hide behind.

    Source ↗
  3. 3
    Down jacket (women's, winter Double 11)

    The CCA made down jackets a named complaint hotspot in 2024, the year clothing complaints jumped 49.96% - fill power and warmth are claims, not pixels.

    Source ↗
  4. 4
    Tailored blazer (women's)

    Livestream women's blazers come back above 80% - and a generated model still shows a back that is really a front.

    Source ↗
  5. 5
    Waist-defined midi dress (women's)

    Sizing explains about 60% of returns - the other 40% is waist, length and drape, exactly what a render is kindest to.

    Source ↗
  6. 6
    Cashmere and fine-gauge knit sweater (women's)

    Reviewers' standing gripe is that model photos hide fit flaws and colour difference - gauge, weight and pilling are what a smooth render erases.

    Source ↗
  7. 7
    Rigid denim jeans (women's)

    Jeans are bought for stiffness, wash and colour - and denim suppliers themselves say returns track how far the photo matches the real cloth.

  8. 8
    Pleated chiffon skirt (women's)

    A pleat is a physical fact - set, sharpness, swing - and reporters found the AI pages coming out with identical folds on every model.

  9. 9
    Yoga leggings and shaping pants (women's)

    Sheerness and compression only appear when you move; a study of cross-border yoga-pant returns blames expectation mismatch on material quality and fit.

    Source ↗
  10. 10
    Basic cotton T-shirt (women's) - the control item

    The floor of the ranking: basics sit far below drape-dependent pieces, and a computer-made model cannot fake what a flat photo already tells the truth about.

    Source ↗

What StepFun noticed

Double 11 2025 set the record: a 61.5% comprehensive e-commerce return rate, with clothing livestreams above 80% and individual merchants above 90% - the backdrop against which this year's AI-model pages are being judged.

The scariest number in the October 2026 coverage - 'nearly 80% of merchants use AI images' - is really a white-paper survey figure of 76.95% of surveyed merchants applying AI material generation, which is not the same as 'nearly all women's sellers'.

Size explains only about 60% of women's returns, which leaves a large block driven by colour, fabric and 版型 - precisely the three things shoppers say AI images hide, and the reason the garment ranking is ordered by fabric dependence rather than by return rate alone.

China has required visible and hidden labels on AI-made content since 1 September 2025, with duties on platforms as well as creators - yet the October 2026 reporting found unlabelled headless and three-armed AI models still live on Singles' Day pages.

🇨🇳 DeepSeek10 places · 4 notes

The same question, answered by DeepSeek alone. The final list above combines all the AIs.

  1. 1
    Silk slip dresses (China, Singles' Day 2026)

    The fabric that AI flatters most is the one that lies: liquid drape and sheen are easy to render, almost impossible to guarantee in a parcel.

    Source ↗
  2. 2
    Structured blazers and tailored coats (China, Singles' Day 2026)

    A blazer is a building: shoulder line, canvas and weight decide whether it stands or collapses — and an AI model never has to carry the weight.

    Source ↗
  3. 3
    Formal gowns and occasion dresses (China, Singles' Day 2026)

    The higher the stakes, the wider the gap: formalwear returns at 36% — a number that says shoppers expected a different dress from the one in the picture.

    Source ↗
  4. 4
    Jeans and structured denim (China, Singles' Day 2026)

    Denim is where the picture lies twice: once about stiffness, once about fit — and return rates above 50% in some peak windows show it.

  5. 5
    Swimwear and bodysuits (China, Singles' Day 2026)

    The highest return band in apparel — 38–44% — meets a garment where fit is everything and AI models can hide every seam.

    Source ↗
  6. 6
    Pleated skirts and bias-cut skirts (China, Singles' Day 2026)

    A pleat is a promise about memory and weight: AI can draw the fold, but not the fabric's willingness to hold it.

    Source ↗
  7. 7
    Knitwear and cardigans (China, Singles' Day 2026)

    The same yarn can look clingy or slouchy depending on gauge and weight — two qualities a synthetic model can fake without ever feeling the knit.

    Source ↗
  8. 8
    Tailored trousers and wide-leg pants (China, Singles' Day 2026)

    The difference between a fluid wide leg and a stiff tube is invisible in a still image — and return data for bottoms stays stubbornly high.

    Source ↗
  9. 9
    Puffer coats and volume outerwear (China, Singles' Day 2026)

    Volume is easy to exaggerate and hard to verify: a puffer that looks cloud-like on an AI model can arrive flat, and the return rate for coats is already 28–34%.

    Source ↗
  10. 10
    Lingerie and lace-trimmed sets (China, Singles' Day 2026)

    The garment with the least public return data may be the most picture-sensitive of all: lace, stretch and transparency are exactly what AI smooths away.

    Source ↗

What DeepSeek noticed

The 80% return-rate headline is not a single number: conventional women's clothing returns run 50–60%, live-stream returns top 80%, and some low-price live rooms approach 90%. The gap between channels is as large as the gap between picture and parcel.

Photo mismatch is not the only return reason, but it is a named, measurable one: 57.32% of live-stream apparel returns are attributed to the item looking different from the screen.

AI adoption is widespread but not universal: 35% of apparel e-commerce sellers use AI try-on or AI-model tools, and Taobao apparel sellers use them at roughly 40%. The 76.95% figure often quoted covers all AI material generation, not AI models alone.

Apparel is the second-most-complained merchant category on Chinese e-commerce platforms (15.89% of merchant complaints in Q3 2026), and 'not as described' sits alongside sizing and colour as a recurring complaint type.

🇨🇳 Wenxin10 places · 1 notes

The same question, answered by Wenxin alone. The final list above combines all the AIs.

  1. 1
    Embroidered Hanfu & Traditional Qipao Dresses

    Up to 80% of 2026 Singles' Day listings use AI models, with 70-80% category return rate

    Source ↗
  2. 2
    Silk Slip Dresses & Heavy Satin Maxi Dresses

    AI renders fake silk sheen 100% inaccurately, driving 28-32% return rate for budget-tier listings

  3. 3
    Structured Tailored Blazers & Office Work Coats

    AI erases 100% of bad tailoring flaws, 62% of buyers report delivered items are baggier than shown

  4. 4
    Plus-Size Formal Dresses & Evening Gowns

    Sizing standards differ by 10cm across brands, AI hides all fabric stretch limitations

  5. 5
    Athleisure Sets & High-Stretch Leggings

    AI can fake 3-4x higher profit margin visual effects, fabric thickness is almost impossible to judge from computer images

  6. 6
    Co-ord Loungewear Sets

    Over 60% of AI product pages alter set proportions, 57% of consumers report top and bottom sizing mismatch

  7. 7
    Floral Workwear Midi Dresses

    AI generates completely fictional floral patterns for 52% of listings, per 2026 China post-sale inspection data

  8. 8
    Sustainable Eco-Friendly Basic Knitwear

    AI misrepresents thin cheap polyester as premium sustainable fabric, very high mismatch profit margins reach premium pricing levels

  9. 9
    Swimwear & One-Piece Beach Outfits

    AI can fully hide sheer thin fabric flaws, 47% of delivered swimwear is more transparent than AI rendered previews

  10. 10
    Basic Graphic Tees & Loose Hoodies

    Simple designs leave almost no room for AI to fake major differences, lowest mismatch complaint rate across all categories

What Wenxin noticed

AI models are so widespread on 2026 Singles' Day women's clothing pages that 60%+ of consumers now manually count fingers on product images to verify if a model is real before placing an order

🇨🇳 Kimi10 places · 4 notes

The same question, answered by Kimi alone. The final list above combines all the AIs.

  1. 1
    Silk slip dresses and camisole dresses (真丝吊带裙)

    AI gives polyester the luster of top-grade silk - the sheen is free, the drape is not.

    Source ↗
  2. 2
    Fur and cashmere winter coats (皮草/羊绒大衣)

    The single highest shop-level return rate reported by any supplier: 80%, from a fur-and-cashmere coat specialist.

    Source ↗
  3. 3
    Down jackets and padded coats (羽绒服)

    133,000 complaints in one year, up 40.3% - the fastest-rising clothing complaint in China, all about fill you cannot see in a picture.

    Source ↗
  4. 4
    Tailored blazers and suit sets (西装外套套装)

    A blazer is shoulder padding, canvas and drape - and a computer-made model has perfect shoulders every single time.

    Source ↗
  5. 5
    Hanfu and 'new Chinese style' sets (汉服/新中式套装)

    The style whose complaints jumped 70% in a year - concentrated on goods that did not match the picture.

    Source ↗
  6. 6
    Fit-and-flare and A-line dresses (收腰连衣裙)

    A Guangzhou dress factory's rule of thumb in 2024: sell 1,000, get 800 back.

    Source ↗
  7. 7
    Cashmere and wool sweaters (羊毛衫/羊绒衫)

    50,000 complaints a year about sweaters whose 'cashmere' exists only in the photo's caption.

    Source ↗
  8. 8
    Wide-leg and tailored trousers (阔腿裤/西裤)

    The crease and the hang are the whole garment - and both are exactly what a render invents.

    Source ↗
  9. 9
    Skinny and fitted jeans (紧身牛仔裤)

    Denim returns less than dresses, but 'L码 means whatever this shop says it means' makes fit a lottery.

    Source ↗
  10. 10
    Basic T-shirts and loose hoodies (基础T恤/宽松卫衣)

    The one garment AI gets right - a grey hoodie looks like a grey hoodie in every universe; its 60% returns come from quality, not the picture.

    Source ↗

What Kimi noticed

The Double 11 image that defined the genre: a shop that 'sent 2,000 orders and got 1,500 back', with some sellers reporting 80-90% returns during the festival.

Down jackets are the single fastest-growing clothing complaint in China - 133,000 complaints in 2024, up 40.3% - because fill content is invisible until the coat arrives.

Shoppers now spot AI models by counting fingers and toes; 36kr's Oct 22, 2026 report documents computer-made models as the default look of many women's clothing pages.

Platform fixes barely move the needle: Tmall's size-recommendation model cut returns by only 2-3 points against category rates of 50-80%.