Not by the amount the category advertises. The 25 to 40 percent figure that appears on almost every try-on vendor page has no published retailer source behind it. The retailers that did publish numbers reported smaller, narrower results: Zalando avoided 8 percent of size-related returns in 2025, and ASOS’s 160 basis point improvement was not credited to try-on alone.
That gap matters because you are being asked to fund a project on the strength of the bigger number. A brand with 25 percent returns hears “40 percent fewer” and models a drop to 15 percent. The published evidence supports nothing close to that, and the difference is the entire business case.
This guide traces where the number came from, lists what each retailer actually disclosed with the scope attached, and covers the thing that changed in 2026: Google now runs try-on inside Search, enrolled most merchants automatically, and renders it from images you already supply.
Key Takeaways
- No retailer has published a return reduction attributable to virtual try-on alone. That is CNBC’s finding after surveying the category, not an inference (cnbc.com, April 2026).
- Zalando’s published figure is 8 percent of size-related returns in 2025, not 8 percent of all returns. The scope is doing most of the work (corporate.zalando.com, accessed August 2026).
- ASOS improved its returns rate by 160 basis points, which is 1.6 percentage points, and attributed it to cost management and a transparent returns policy alongside try-on trials (pymnts.com, April 2026).
- Google put try-on in Search and auto-enrolled merchants with a qualifying shopping feed. You may already be in it (support.google.com/merchants, accessed August 2026).
- Google renders from your on-model image, and publishes exactly what that image must look like. The source photo is the variable you still control.
- Shopper-facing try-on and brand-side try-on are different projects. Only one of them touches biometric consent law.
Does virtual try-on reduce returns?
There is directional evidence that fit and visualisation tools reduce size-related returns, and no published evidence that virtual try-on on its own produces the 25 to 40 percent reduction the category quotes. The retailers running it at the largest scale have either disclosed narrower figures or disclosed none. Treat try-on as one input to a returns programme, not as the programme.
Where the 25 to 40 percent number comes from
Search the primary keyword and you get vendor guides, agency explainers and listicles. They agree on the numbers with unusual precision: 25 to 40 percent fewer returns, roughly 30 percent higher conversion, and a returns market sized somewhere between 550 and 800 billion dollars.
None of them cite a retailer. Follow the citations and they end at another vendor page, a survey run by a company selling the software, or nothing at all. One of the better-ranked pages attributes its most specific consumer statistic to its own unpublished internal research.
The returns market figures are similarly loose. The National Retail Federation put total US merchandise returns at 849.9 billion dollars in 2025, or 15.8 percent of annual sales, with 19.3 percent of online sales returned (nrf.com, October 2025). That is all retail across every category, not fashion, and not online only. A fashion-specific number carved out of it without stating the method is not a statistic, it is a rounding.
This is not a claim that try-on does nothing. It is a claim that the number you are budgeting against was not measured.
What each retailer actually published
Four disclosures are worth having in front of you, with their scope intact.
| Source | Published figure | What it actually covers |
|---|---|---|
| Zalando | 8 percent of size-related returns prevented in 2025 | All fit solutions combined, from algorithmic size advice to body measurements. Not try-on alone, and size-related returns only. |
| Zalando | Up to 40 percent return reduction | Virtual Fitting Room pilots, not general rollout. The one place the 40 percent figure is real, and it is a pilot. |
| ASOS | 160 basis point returns rate improvement | Credited to more rigorous cost management and a transparent returns policy, alongside try-on experimentation with AIUTA. |
| ASOS | No returns or conversion metric | Its February 2026 try-on launch announcement discloses none. |
Zalando’s 8 percent is the most instructive line in the table. If size-related returns are, say, 60 percent of your returns, then preventing 8 percent of them moves your total return rate by under 5 percent relative. On a 25 percent return rate that is roughly 1.2 percentage points. Real money at Zalando’s volume. Not a rebuilt P&L at yours.
The 40 percent figure does exist, in Zalando’s Virtual Fitting Room pilots. A pilot is a controlled subset, often a single brand’s garments, often with self-selecting users. It is a reason to run your own test. It is not a planning assumption.
ASOS launched a hybrid try-on with AI platform AIUTA on 17 February 2026, covering around 10,000 products on its iOS app for selected UK and US customers, with each render loading in four to seven seconds (asosplc.com, February 2026). The announcement quotes a product lead on shopper choice. It quotes no performance number at all. A company that publishes campaign-level asset counts in its interim results would publish a returns figure if it had a clean one.
Google put try-on in Search and enrolled you automatically
While brands were evaluating try-on vendors, Google shipped try-on into Search and Shopping itself, rendering from the Shopping Graph.
The enrolment detail is the one most brands miss. Merchants are automatically enrolled if they have a shopping feed and qualify for free listings. Opting out means contacting Google support, and it only removes the button from your own listings. It does not remove it from other merchants selling the same product (support.google.com/merchants, accessed August 2026).
Read that twice. A competitor stocking your SKU can carry a try-on experience for your garment whether or not you chose one.
The supported categories are shoes, tops, bottoms and dresses. Lingerie, swimwear and accessories are outside it. The feature operates across 19 regions including the US, UK, Canada and Australia.
What this does to the buying decision is straightforward. The shopper-facing try-on widget, the thing the category has been selling for three years, is now partly a commodity Google supplies alongside your listing. Paying to build a second one is a legitimate choice, but it is no longer the default move, and the business case has to clear a higher bar than it did in 2024.
What Google requires from your product images
Here is the part that turns a platform story into a production task. Google’s try-on does not invent your garment. It renders from the product image in your feed, and Google publishes the specification.
The requirements, verbatim from Google’s merchant documentation (accessed August 2026):
- The image should show the entire garment and should be a full body picture
- The garment is on a model facing directly forward with arms down to the side
- Images at least 512 by 512 pixels, ideally 1024 pixels or higher
- Avoid excessive folds, and do not cover garment details
- For tops, sleeves should be rolled down
That is a shot list. Full body, front-facing, arms down, no occlusion, high resolution, one per garment. If your catalogue is flat lays and three-quarter editorial crops, your listings render worse than a competitor whose images match the spec, on an experience neither of you controls.
This is the argument for spending on the source image rather than the widget. The on-model photograph is the input to Google’s try-on, to any vendor try-on you later add, and to the product page itself. It is the only asset in the chain you own outright.
Producing one front-facing, full-body, arms-down on-model image per SKU used to mean a shoot day per drop. Virtual try-on in DesignerBox generates it from a flat garment photo you already have, and Outfit to Image runs the same job for a full outfit. Try-on sits on the Premium plan at 75 dollars a month. The practical guide to the shot itself is in on-model photography with AI, which covers matching new images to an existing catalogue.
The three questions to ask any try-on vendor
Every claim in a try-on pitch resolves under three questions. Ask them in this order.
1. Percent of what? “40 percent fewer returns” is meaningless without the denominator. Size-related returns, or all returns? Of the products with try-on enabled, or of the whole catalogue? Of shoppers who used the feature, or of all shoppers? Zalando’s honest 8 percent and a vendor’s 40 percent can describe the same underlying effect measured two different ways.
2. Measured against what? A return rate that falls after launch is not evidence. Returns move with season, mix, promotion depth and returns-policy changes. ASOS’s 160 basis points arrived alongside a policy change, which is precisely why nobody credits it to try-on. Ask for a holdout group. If there was no holdout, there was no measurement.
3. Whose customers? Shoppers who choose to use try-on are already higher-intent than shoppers who do not. Compare try-on users to non-users and you measure self-selection, not the feature. This is the single most common flaw in published try-on results and it inflates every number in the category.
A vendor with real data answers all three in a sentence each. A vendor without one changes the subject to case studies.
What to spend on instead this quarter
If the goal is fewer returns, the evidence points at sizing accuracy and image accuracy, in that order, and at neither being a try-on widget.
Sizing accuracy is Zalando’s actual finding. Their 8 percent came from size advice and body measurement, not from rendering garments on people. Size charts per garment rather than per brand, plus fit feedback surfaced on the product page, are cheaper than any try-on integration and are what the published number describes.
Image accuracy is the half you can move this quarter. A return caused by “looks different in person” is caused by the photograph, and no try-on render fixes a source image that misrepresents colour, drape or scale. That is also where the compliance line now sits. The European Commission’s guidelines on the AI Act’s transparency rules give an AI-generated product image that makes a product “appear not identical to the real product, more appealing or with improved quality than in real life” as an example of content requiring disclosure, while a real product shown against an AI-generated background is given as an example that does not (Commission Guidelines C(2026) 5054 final, 20 July 2026). Article 50 has applied since 2 August 2026.
The practical read: generate the scene freely, and keep the garment honest. A flattered garment is now both a returns driver and a regulatory question.
Two adjacent decisions are worth settling at the same time. Whether you need shopper-facing try-on at all, which is a biometric consent project rather than an imaging one, is covered in how to add virtual try-on to your store. What try-on can and cannot tell a shopper about fit is covered in virtual try-on fit accuracy, including why current models render a good fit by default whatever size is involved.
For the catalogue work itself, the Model Studio turns one product photo into the full set of angles and on-model shots a product page needs, and the fashion OOTD workflow reruns the same treatment for every new drop.
FAQ
Does virtual try-on reduce returns?
There is directional evidence that fit and visualisation tools reduce size-related returns, and no published retailer figure attributing a reduction to virtual try-on by itself. CNBC reported in April 2026 that no platform has published a definitive return rate reduction tied solely to virtual try-on. Zalando’s published 8 percent covers all its fit solutions combined and only size-related returns.
Where does the 25 to 40 percent figure come from?
It circulates across vendor marketing pages without a traceable retailer source. The one place a 40 percent figure appears in a company’s own disclosure is Zalando’s Virtual Fitting Room pilots, which are a controlled subset rather than a general rollout. Treat the range as a marketing claim until a vendor supplies the denominator and the holdout group.
Is my store already in Google’s virtual try-on?
Possibly. Google automatically enrols merchants with a shopping feed that qualifies for free listings. Supported categories are shoes, tops, bottoms and dresses, across 19 regions. Opting out requires contacting Google support and only removes the button from your listings, not from other merchants selling the same product.
What product images does Google’s try-on need?
A full body image showing the entire garment, on a model facing directly forward with arms down at the side, at least 512 by 512 pixels and ideally 1024 or higher. Avoid excessive folds, do not cover garment details, and roll sleeves down on tops. Flat lays and cropped editorial shots do not meet the specification.
Is shopper-facing try-on the same as on-model AI imagery?
No. Shopper-facing try-on takes a photo of the customer and is a biometric consent project in several jurisdictions. Brand-side try-on puts your garment on a generated model to produce the images on your product page. Most brands asking for try-on want the second one.
Do AI-generated on-model images need a disclosure label?
The AI Act’s transparency rules have applied since 2 August 2026. The Commission’s guidelines treat a product image that makes the product look better than the real thing as requiring disclosure, and treat a real product against an AI-generated background as not requiring it. Whether an anonymous synthetic model in an ordinary clothing advert triggers the obligation is still an open question with no enforcement decision yet.
What should a fashion brand fund first?
Sizing accuracy, then image accuracy. Per-garment size charts and fit feedback are what Zalando’s published number actually describes. Front-facing, full-body on-model images are the input to every try-on experience you or Google run, and they are the only asset in the chain you own.
Sources
- CNBC, “Silent killers: How AI is trying to solve retail’s returns problem”, April 2026
- PYMNTS, “Retailers Bet on AI Fitting Rooms to Slash Costly Returns”, April 2026
- Zalando Corporate, how Zalando helps customers find the right size, accessed August 2026
- ASOS plc, hybrid virtual try-on launch announcement, February 2026
- Google Merchant Center Help, how Google’s Try-on tool works, accessed August 2026
- National Retail Federation, 2025 consumer returns report, October 2025
- European Commission, guidelines on transparency obligations for AI-generated content, C(2026) 5054 final, 20 July 2026
Retailer figures verified from company disclosures and reporting as of August 2026. Return-rate outcomes vary by category, price point and returns policy. Individual results vary.
