Ask the internet whether virtual try-on reduces returns and it answers with a confident number — a different confident number every time. A single search on 2026-09-01 returned pages stating returns fall by 36%, 40%, 48%, 64%, and 65%, each presented as the fact, none naming the study behind it. This is the fact-check: which try-on statistics survive scrutiny, which are zombies, and how to tell them apart.
Because most were never measured — they were copied. A vendor publishes a round number, the next blog cites the vendor and re-rounds, and after a few hops the figure looks like consensus while its source has evaporated. The tell is the scatter itself: one vendor's own site surfaced two live pages this week — one headlined "up to 40%," the other "up to 48%" — for the identical claim. A number that changes depending on which page you land on was never a measurement.
| "Returns reduced by…" claim | Where it circulates (2026) | Traceable primary source? |
|---|---|---|
| 36% | Fashion-VTO vendor blogs | No study named |
| 40% | Vendor blogs; "Shopify saw 40%" | No — attributed to "Shopify," no report exists |
| 48% | Vendor blog (same site as the 40% one) | No |
| 64% ("brands with VTO average 64% fewer") | Listicles, repeated widely | None — untraceable to any primary |
| 65% ($2.3M saved) | AR "success story" pages | Single unnamed case study, no methodology |
| "up to 40%" (Zalando, one brand) | Trade press | Yes — but a single-brand anecdote [B], not a category rate |
Only the last row has a real source, and even it is one brand's early test — honest as an anecdote, useless as "the" number for your store. The rest sit on our permanent do-not-cite list ("64% fewer returns," "94% higher conversion," "Macy's under 2% returns") — repeated everywhere, tracing to no readable methodology.
The survivors share three traits: a named source, a disclosed sample size, and a claim that matches what was measured. Notice what they are about — confidence and purchase intent, not a headline returns percentage.
| What it measures | Figure | Source | Tier |
|---|---|---|---|
| Online fashion vs in-store conversion | ~1.5–3% vs 15–30% (up to ~10×) | IRP Commerce 2026 + TruRating | A/B |
| Try-on users less likely to return (self-reported) | 67% | Snap + Publicis 2022, N=4,028 | A |
| Try-on users more confident in the purchase | 80% | Snap + Publicis 2022, N=4,028 | A |
| Try-on raises purchase intent | across 505,416 shoppers | Vieira et al., J. of Business Research 2022 | A |
| Expect AR shopping (Gen Z / Millennials) | 81% | Klarna 2023, n=5,055 | A |
| Say a try-on helps them picture the item | 59% | Nosto / Censuswide 2022, n=2,019 | A |
| Size/fit = #1 cited apparel return reason | ~50–70% of returns | Coresight 53% / EEA ~70% | B/A |
| Google Shopping try-on images | +60% high-quality views | Google, Sept 2024 | A |
Look at the sample size. A real study tells you how many people it measured; a laundered stat cannot, because there was no study. Sample sizes behind each figure:
A real effect converges as more people measure it. When five "the number" return stats fan out from 36% to 65% with no study behind any of them, the scatter isn't disagreement — it's the fingerprint of a number that was never measured.
The well-evidenced effect is upstream of returns: try-on raises confidence and purchase intent, and confident shoppers return less — but the clean sourced returns figure is Snap's self-reported 67% less likely to return from a 4,028-person study, not a store-wide "X% fewer returns" you can bank. Don't borrow anyone's percentage; compute your own. A 200-orders/week store at a 25% return rate, ~60% of it size/fit, at ~
Ello holds itself to the same bar. At one live Ello store, measured in its own dashboards with an always-on 10% holdout: 18.3% of try-on sessions purchased (21 of 115), ,150 attributed revenue over 30 days, ~$0.067 compute per try-on — labeled (our data), denominators shown, never extrapolated into a category claim. One store, results vary, try-on users are higher-intent. That discipline is the pitch for Ello: 2D AI try-on on your existing product photo, covering clothing and accessories, with numbers we source or label rather than inflate. Verify for yourself — compare the Shopify try-on apps and read the real client numbers.
The well-evidenced effect is on confidence and purchase intent, which reduce returns indirectly. The cleanest sourced returns figure is Snap + Publicis's self-reported 67% less likely to return, from a 4,028-person study (2022). The store-wide "cuts returns 36/40/48/64%" figures trace to no primary study — treat them as marketing and measure your own fit-return rate instead.
Because most are copied rather than measured. A vendor publishes a round number, other blogs cite the blog and re-round, and it looks like consensus while its source disappears. A genuinely measured effect converges; when the same claim appears as 36%, 40%, 48%, and 64% with no study named, the scatter is the sign it was never measured.
The ones with a named source and disclosed sample: 80% more confident and 67% less likely to return (Snap + Publicis, N=4,028); purchase intent up across 505,416 shoppers (Vieira et al., 2022); 81% expect AR shopping (Klarna, n=5,055); 59% say try-on helps them picture the item (Nosto, n=2,019); size/fit as the #1 cited apparel return reason at ~50–70% (Coresight/EEA).
Click through to the origin until you reach a named study with a sample size, not another blog. If the same claim shows up as several different percentages, treat the whole cluster as unsourced. And confirm the wording matches the measurement — "67% less likely to return" is self-reported survey intent, not a measured store-wide return drop.