Do Virtual Try-On and Size Tools Cut Your Returns?
I ordered the same pair of chinos three times last year. The first pair came back because the waist gaped, the second because the leg was cut for someone eleven centimetres taller, and the third fitted.

I ordered the same pair of chinos three times last year. The first pair came back because the waist gaped, the second because the leg was cut for someone eleven centimetres taller, and the third fitted. On every one of those product pages a widget had told me, with a confident little tick, that I was a 33 regular. The widget was not lying. It was answering a different question from the one I was asking.
Clothing and footwear carry the worst return rates in online retail. Industry figures usually land somewhere between 20 and 30 percent for apparel, against roughly 10 percent across all categories, and almost all of that gap is fit. So the fitting tools matter commercially, which is exactly why they are built the way they are. I spent a few years writing browser extensions before I started reviewing shopping tools, and the thing I now look at first is not the accuracy claim. It is what the tool is being paid to optimise.
Three different technologies wearing one label
Retailers call all of these virtual try-on, and shoppers reasonably assume they do the same job. They do not. There are three families, and knowing which one you are looking at tells you how much weight it deserves.
| Family | What it actually uses | Where you see it | How much I trust it on fit |
|---|---|---|---|
| Size recommendation | Your height, weight, age and sometimes brands you already own, matched against what other shoppers kept | Jeans, shirts, dresses, running shoes | Moderate for the brands with a lot of data, poor for new or small labels |
| Body measurement | Two phone photos or a short video, turned into estimated chest, waist and inseam numbers | Tailored shirting, suiting, some denim | Reasonable on torso width, weaker on length and shoulder slope |
| AR overlay | Camera plus face or foot tracking, draws the product on the live image | Glasses, makeup, watches, sneakers | Zero. It shows appearance, not fit, and it says so in the small print |
The third family is the one people misread most often. An AR overlay that puts a sneaker on your foot through the camera knows nothing about the width of the toe box or the volume of the instep. It is a mirror with a product in it. Useful for deciding whether you like navy, useless for deciding between a 42 and a 42.5.
Who the recommendation is really for
Here is the part that changed how I use these widgets. A size recommendation engine is sold to retailers on a single number: reduction in returns. Vendors advertise figures in the region of 20 to 30 percent fewer size related returns, and retailers buy on that basis.
Think about what that target rewards. The safest prediction for the engine is not the size that fits your body best. It is the size that shoppers with a profile like yours were least likely to send back. Those two overlap most of the time, which is why the tools are not useless. But they come apart in a specific and predictable way: when a garment sits between sizes for you, the engine will nudge you towards the larger one, because people keep clothes that are slightly loose far more often than clothes that are slightly tight. I have watched this happen across enough orders that I now mentally subtract half a size from any recommendation that comes with a message about a relaxed or comfortable fit.
The common advice is to trust the tool because it has seen millions of orders. I would put it the other way round. The tool has seen millions of orders and almost nothing about you, usually three or four numbers you typed in thirty seconds. Your own best fitting shirt, measured flat with a tape, contains more information about your body than the entire model does.
What the tool collects while it helps you
Permissions are my old habit and I have not shaken it. Three things are worth checking before you hand over anything.
Body measurement tools that work from photographs need camera access, and the image usually goes to a server rather than staying on the handset. Some vendors say they discard the photo after extracting measurements. Some keep an anonymised body model. The difference is in the privacy notice, not in the interface, and it takes about two minutes to find. If the notice does not mention retention at all, that absence is itself the answer.
Cross retailer profiles are the second thing. The larger size recommendation vendors work with hundreds of brands and keep one profile per shopper, which is how they can say a shopper who takes a medium in one label takes a small in another. That is genuinely the source of their accuracy, and it also means your sizing history is a shared record held by a company you never chose to deal with. Most of them support a data deletion request, and the link is normally buried in the widget itself under a small question mark.
Then there are the extension versions, which install a shopping helper that promises fit advice on any store. Those request permission to read and change data on every site you visit, because they have to inject their widget into pages the vendor does not control. That is a heavy permission for a size chart. I go into why that request pattern keeps appearing in our piece on shopping apps that ask for more permissions than they need.
The method that beat the widgets for me
None of this means skipping the tools. It means demoting them to a tiebreaker. The routine below takes about fifteen minutes once, then thirty seconds per order, and it cut my own clothing returns from six in a year to one.
Start by measuring garments rather than your body. Take the best fitting shirt you own, the best fitting trousers and the shoes you wear most, lay each one flat, and write down chest width pit to pit, waist across, sleeve or inseam length, and the insole length of the shoes in centimetres. Those five or six numbers live in a note on my phone and have outlived four shopping accounts.
Then look for flat measurements on the product page, usually in a tab labelled garment measurements. They beat body measurements for one blunt reason: they describe the object that will arrive in the box rather than a model's guess about you. If the figures land within about a centimetre of your note, order that size no matter what the widget says. Save the recommendation for pages that offer nothing but S, M and L.
Two smaller checks catch most of what is left. Read the fabric composition before the size, because a non stretch cotton twill in the wrong size is a return while the same size with four percent elastane usually is not. And price the exit before you buy, since who pays for the return decides whether ordering two sizes is clever or expensive. Our guide to who pays return shipping covers how to avoid the fee.
Ordering two sizes deliberately is still the most reliable method available, and I do it for anything over about 80 dollars from a brand I have not bought before. It is also the behaviour every one of these tools was built to stamp out, which is worth remembering when a site makes it awkward. For more on judging a tool before you install it, the rest of shopping tools covers the same ground for price trackers and coupon helpers.
FAQ
Are virtual try-on tools accurate for shoes?
The AR versions are not, since they only draw the shoe over your foot in the camera view. Foot scanning tools that measure length and width from photographs do better, often landing within about half a size, but they cannot judge instep volume, which is what usually makes a shoe uncomfortable rather than too short.
Do I have to give my height and weight?
Most size recommendation widgets will not produce an answer without them, because weight is the strongest single predictor in their model. If you would rather not, use the flat garment measurements on the page instead. You lose very little, and in my experience you often gain.
Why does the same tool suggest different sizes on two sites?
Because it is reading each brand's own pattern data, and brands cut differently even when the label says the same number. That variation is a real feature rather than a fault. It also means a recommendation carries almost no weight on a brand the vendor has little data for, such as a new label or a small independent maker.
Can the camera based tools work without sending my photo anywhere?
A few newer ones run the model on the device, and they will usually say so prominently because it is a selling point. Assume everything else uploads the image. If the app works with the phone in airplane mode, the processing is local.
More in Shopping Tools
Shopping Tools
Checkout Bots and Why You Keep Losing Limited Drops
The sneaker release opened at 10:00:00. You had the page loaded at 9:59, your card saved, your size picked.
Shopping Tools
Can You Trust an AI Shopping Assistant's Product Picks
You ask an AI shopping assistant for "the best budget air purifier," it names three models with confident little summaries, and...
Shopping Tools
Is It Safe to Save Your Card in Browser Autofill
You click into a checkout field, your browser offers to fill in your card, and two taps later the order is done.