AI in fashion: what Zalando, Inditex, H&M and Google have actually said, graded

AI in fashion today means four things brands have put on the record: generated marketing images, virtual try-on, shopping assistants, and digital twins of human models. The best-documented numbers come from Zalando, which says AI-generated product content reached 90% in 2025, and Inditex, which reports more than 7 million Zara Try-on sessions. Here is who is using what, and how much of it holds up.

What the brands themselves have said

Most coverage of AI in fashion leans on market-size forecasts and vendor blogs. We went to the companies instead. Zalando's full-year results release is dated March 12, 2026. Inditex's FY2025 results page is dated March 11, 2026. H&M Group's news post on its digital twins is dated July 2, 2025, and Google's try-on announcement is dated July 24, 2025. Everything below traces to those pages or to named trade coverage of them, and each claim gets a grade.

Zalando: the biggest number, and a company claim

Zalando's March 12, 2026 release says the company "scaled its AI-generated product content from near zero to 90% in just one year." Retail Gazette, reporting the same day, put the effect in operational terms: campaign production time fell from six weeks to a few days.

The same release says the Zalando Assistant reached six million users, "a fourfold increase from last year," and that its AI matchmaking "has already driven a 13% increase in items added to bags." It also says AI is "avoiding size-related returns by more than 8%." Zalando named itself one of only two European launch partners for Google's Universal Commerce Protocol, which is meant to let shoppers buy through AI chatbots such as Gemini.

Grade: real scale, unaudited. These are company-reported figures in a results release. The release does not say what counts as "AI-generated product content," whether a retouched photo counts, or how the 8% was measured. Items added to a bag are not sales.

Inditex: Zara Try-on, counted in sessions

Inditex's FY2025 release, dated March 11, 2026, describes Zara Try-on as "an AI-based virtual fitting system that allows customers to create a synthetic avatar from their own photos." It reports the tool in 43 markets with over 7 million sessions, running only on Zara.com for now, with a rollout to Inditex's other brands ongoing. The same page reports online sales of 10.7 billion euros and net profit of 6.2 billion euros for the year.

Grade: genuine deployment, thin evidence of effect. A session is a visit, not a purchase and not a person. The release gives no data on conversion or returns. For what independent data exists on try-on tools, see our piece on virtual try-on tools and what the return data shows.

H&M: digital twins of 30 models, one campaign so far

H&M's news post of July 2, 2025 says the company released "its first set of images with digital twins," set against fashion capitals and focused on seasonal denim, with more planned for the fall. In March 2025, FashionUnited reported that 30 models had given permission for AI twins, that they would be paid for their digital replicas in the same way as for photographs, and that the first AI images on social media would carry watermarks. That pay and watermark detail comes from H&M's interview with Business of Fashion as relayed by FashionUnited. H&M's own July post does not mention labeling or rights.

Chief creative officer Jörgen Andersson wrote of the plan, per FashionUnited, that "The possibilities are almost endless." Model Mathilda Gvarliani, quoted by the same outlet, said of her twin: "She's just like me, but without the jet lag."

Grade: a real but small experiment. H&M called it a test of emerging technology. We found no published figure on how many images, how much spend, or what the twins earned the models.

Google: try-on as a shopping layer

Google's July 24, 2025 announcement said "Today, try on is launching in the U.S.," moving it from a Search Labs experiment into apparel results across Search, Shopping and Images. Shoppers tap a try-on icon and "Upload a full-length photo of yourself." Google said the tool covers "billions of apparel items." Digital Commerce 360 reported that Google's chief business officer Philipp Schindler told investors on Alphabet's second-quarter call: "Now people can try billions of clothing products on themselves virtually."

Grade: widely available, unmeasured. Google's post gives availability and catalog scale, not accuracy, fit or return rates. It also does not say how uploaded photos are stored.

Guess in Vogue: when the AI model is the story

The one case where AI in fashion became a public fight was not a retailer's results release. Guess ran an ad in the August 2025 issue of American Vogue featuring AI-generated models made by the London agency Seraphinne Vallora. ABC News reported on August 1, 2025 that the agency created a blonde named Vivienne and a brunette named Anastasia, and that critics online objected. TikTok creator Payton Wickizer called it "The fact that they are using fake women in their magazines -- speechless." The agency's co-founder Andreea Petrescu told Good Morning America: "This is meant to just supplement and to add."

ABC News said it had contacted Vogue and Guess and had not received a response when it published. We did not find a statement from either company, so we do not characterize their position.

Grade: a real controversy over disclosure and labor, with no company data on results.

What the pattern shows

Three things stand out. First, the largest numbers sit in content production, which is the least visible to shoppers. Second, try-on is everywhere and measured nowhere: Zara counts sessions, Google counts catalog size. Third, the people-facing uses, twins and AI models, are the ones where consent, pay and labeling decide the reception, and the brands' public detail on those points is thinnest.

Our take

Treat Zalando's 90% as a credible direction and an unverified number. Treat Zara Try-on as a real product with no proof yet that it changes what people keep. Watch H&M's next drop for whether labeling and model pay are stated in the company's own words. Skip any claim that AI is transforming fashion that cannot name a brand, a date and a figure. For background on the high-volume end of the industry these tools serve, see what fast fashion is.