Try it on. Just don't ask if it fits.

On 30 April 2026, Google switched off Doppl.

Doppl was a Google Labs experiment launched in June 2025: upload a full-length photo, see clothes rendered onto your body, generate a short clip of the outfit in motion. Ten months later the app stopped working and was pulled from both app stores.

It would be easy to file this next to the body-scanning retreat we covered a fortnight ago. It isn't the same story. Doppl didn't fail — it graduated. Google folded the technology into Search and Shopping product listings, where a "Try it on" button now sits on non-sponsored apparel results. The experiment closed because it no longer needed to be an experiment.

That is a distribution change of a completely different order. A standalone app requires a download and an intention. A button on a search result requires neither. Merchants with a shopping feed are enrolled automatically; the opt-out is a support ticket. The feature is live in the UK, alongside the US, Canada, Australia, Japan and a dozen other markets. Supported categories are shoes, tops, bottoms and dresses. Lingerie, swimwear and accessories are excluded.

So: the most widely deployed piece of fit-adjacent technology in fashion retail, sitting on ordinary search results, in this country, today. Which makes the obvious question worth asking properly.

Does it reduce returns?

The number that isn't there

Search for the answer and you will find a lot of numbers. Twenty per cent. Thirty. Thirty-six. Forty. Forty-eight.

Follow each one back and the trail ends in the same place: a company that sells virtual try-on software, or a blog citing a company that sells virtual try-on software. Not one of the figures in wide circulation traces to a published study, a retailer's audited accounts, or any source with something to lose by being wrong.

That is not an accusation of dishonesty. Vendors quoting their own pilot results is ordinary commercial life. But a figure repeated across a hundred pages is not thereby verified, and a claim that only ever appears downstream of the people selling the thing is a claim that hasn't been tested.

We hold ourselves to a rule here: no statistic goes on this site unless we can reach a primary source. Applying that rule to virtual try-on returns data leaves the page blank.

What the evidence actually says

There is one piece of serious independent work. Gallino and Moreno, publishing in Manufacturing & Service Operations Management in 2018, ran randomised field experiments at an online retailer — customers assigned at random to see virtual fit information or not. They found it lifted conversion and order value and cut the fulfilment costs caused by returns and by customers ordering several sizes at once.

Real evidence, properly done. But read the mechanism the authors identify. Among the reasons the tool helped: it reduced uncertainty by providing a size recommendation.

The tool in that study bundled two things — a picture of the garment on you, and a prediction of which size to order. The study can't separate them. And the half that carries a number, a specific instruction to buy the 32 rather than the 34, is the prediction.

Ask the retailer who went first

ASOS ran Europe's first large-scale virtual try-on trial. See My Fit launched in 2020 on 800 dresses, rendering each onto sixteen models across a wide size range. It was genuinely ahead of the field.

Six years on, ASOS published its FY25 results. Returns get a substantial section — the company describes them as both a cost and a source of friction for customers. It sets out what it did about them: improved size guides, more customer reviews, and a refined fair-use policy for the small group of persistently high-returning accounts. Those actions, it reports, cut the underlying returns rate by around 150 basis points year on year, and the benefit shows up in distribution costs.

Virtual try-on is not mentioned.

That isn't a criticism of ASOS — quite the opposite. ASOS looks like a company that measured honestly and then invested in what moved the number. Better size guides. More reviews. Clearer information about the garment. The unglamorous work.

Walmart tells a similar story from the other direction. It bought Zeekit in 2021 and by September 2022 had Be Your Own Model live across more than 270,000 items — probably the largest deployment of photo-based try-on anywhere. Its launch announcement described the experience in terms of discovery, confidence and how a garment drapes. Four years later, no published returns figure has followed.

The sentence in the small print

Which brings us to the documentation.

Google maintains a Merchant Center help page explaining how try-on works. It's written for retailers, not shoppers, which is precisely why it's worth reading — it's where a company describes its product to people who need to know what it actually does.

The page frames the feature as enhancing discovery and encouraging exploration. It notes the result is not a perfect representation. And then, under limitations, it says the generated image <cite index="75-1">doesn't indicate fit, suggest a size or indicate size availability from the merchant</cite>.

The FAQ puts it even more plainly: the tool is there to give a sense of style and colour, and does not show exact fit or size.

That is Google, in its own words, about its own product, on a page it maintains for its own merchants.

There is no gotcha here. Nothing is being hidden — the page is public and the wording is unambiguous. The gap isn't between what Google built and what Google says. It's between what Google says and what the category has been sold as ever since.

Two questions, not one

Standing in a fitting room, you ask two things.

Will I like this on me? Colour against your skin. Length against your leg. Whether the shape does what you hoped. This is a question about appearance, and it is genuinely hard to answer from a flat product photo on a model who isn't you.

Virtual try-on answers it. Well, quickly, and now at the scale of ordinary search results. That is a real achievement and it deserves to be described accurately rather than dismissed.

Will this fit me? Whether the chest measurement clears yours with room to move. Whether the inseam lands where you want it. Whether this brand's 32 is the same 32 as the last brand's.

This is a question about numbers. It is answered by comparing the garment's measurements to your body's measurements. A generated image cannot do it, which is why Google says it doesn't, and no amount of rendering fidelity changes that — a photorealistic picture of a garment that is four centimetres too tight is still a photorealistic picture.

The two questions get conflated constantly, because in a physical fitting room you resolve both in the same thirty seconds. Online they come apart, and they need different tools.

Where this leaves us

Virtual try-on is not a failed technology. It's a solved problem in the wrong column of the ledger — filed under returns, when it belongs under discovery.

The returns problem is a measurement problem. It gets solved by knowing the body and knowing the garment, and comparing them. That's slower, less demoable, and considerably harder to put in a launch video.

It's also the thing we're building.