Last week, OpenAI rolled out new features on ChatGPT, enabling the conversational artificial intelligence tool to make shopping recommendations. Interested to find out how it would handle the notoriously complex topic of sustainable fashion, I took it for a spin.
I started simple, sharing my size and location, and asking ChatGPT to recommend a white T-shirt. It quickly came back with follow-up questions: did I have any preferences in terms of fabric, fit, budget, purpose or neckline? I told ChatGPT I wanted the T-shirt to be as versatile as possible, but my main concern was sustainability. What would it recommend?
A pop-up informed me that I was using an updated version of ChatGPT, so there were two possible responses for me to choose between. Both made a number of recommendations, highlighting different facets of sustainability, so I could select a product based on my personal preferences and priorities. In the mix, there was an organic cotton T-shirt for £12 for the budget-conscious, one made in a renewable energy-powered factory for £18 for those with ‘carbon tunnel vision’, and a premium choice at £58 that ChatGPT told me “combines style with sustainability, using eco-friendly materials and transparent production processes” (although it didn’t provide any further information to back up these claims).
OpenAI has announced updates to the shopping experience on its generative AI platform ChatGPT, recommending products and where to buy them. Is this the future of online shopping?
Hoping to understand more about the selection criteria, I looked into the sources. In some cases, it drew information from brand websites. But in most cases, it leant on SEO-informed shopping listicles from consumer-facing magazines, which often earn affiliate commission or get products gifted in return for their recommendations. One of the referenced sources was a New York Post article about Swedish brand Asket, which prioritises transparency, but the resulting recommendation was for H&M Group-owned Arket.
I amended my search: could ChatGPT recommend the most sustainable white T-shirt available, based solely on certified sources, including brand sustainability reports, and could it use product-level data, not generic brand claims. I asked it to exclude sources using affiliate marketing and stated my preference for a regenerative organic cotton white T-shirt, which has no toxic synthetic dyes, from a brand that supports farmers directly and pays its garment workers a living wage.
The next product list was much shorter, but it was still leaning on me to make the final judgment on which product was most sustainable. When pushed, ChatGPT finally gave me a definitive answer. Patagonia’s regenerative organic cotton tee came out on top, on the basis that it “supports farming practices that rehabilitate soil and improve farmer livelihoods”, “is produced in a Fairtrade-certified factory, ensuring workers receive a premium for their labour”, “prioritises Bluesign-approved chemicals” and “publishes detailed impact reports, factory lists and third-party audits”.
It wasn’t a bad recommendation, but it took some pretty sophisticated prompts and a deep knowledge of sustainable fashion to get to this point. It also skipped over several smaller, independent brands that I know fit the criteria, and I wouldn’t feel confident clicking ‘buy’ without a fair bit of manual fact-checking and research to verify the claims. Not to mention, it neglected to offer secondhand options.
OpenAI says the sustainability element of ChatGPT’s recommendations are in the early stages, and recommendations are based on the user’s perceived intent — so secondhand options wouldn’t show up unless I specifically asked for them. Still, my experiment threw up some concerns about how AI tools will likely process and judge sustainability data moving forward. It showed the lack of consensus on what makes a product sustainable or not, the lack of transparent and verified data to base green claims on, and the high risk of AI simply regurgitating greenwashing instead of helping consumers wade through it. Where do we go from here?
Bad data in, bad recommendations out
Diana Sarai Hernández Manzo is currently studying for a PhD in virtual reality at Scotland’s Robert Gordon University, exploring whether AI-based conversational agents can be used to promote sustainable fashion. She has built a virtual reality escape room game, which educates players on sustainable fashion through the Scottish heritage fabric Harris Tweed. In one scenario, players interact with a text-based agent, similar to ChatGPT. In another, users will interact with an embodied avatar resembling Hernández Manzo herself, which uses voice commands, similar to Apple’s Siri or Amazon’s Alexa. The idea is to figure out which types of AI have the biggest potential as sustainability changemakers.
In her research, Hernández Manzo stumbled upon early attempts to use either text-based or voice-based chatbots to make sustainable product recommendations, as well as the myriad challenges associated with these ventures. Namely, recommendations are only as good as the data they are based on. “AI works in real time, and most AI chatbots save your preferences, so they have the potential to make personalised recommendations, which are more up to date than books,” she explains. “That could be incredibly helpful to consumers, but only with the right data access. Many brands are not transparent about their supply chains and the real environmental impact of their products, so AI cannot make recommendations without reinforcing those gaps and biases.”
Last year, the European Parliament passed the AI Act, the world’s first legal framework for the responsible development and use of AI. Its primary motive is making AI “trustworthy” and minimising both social and environmental risks. How this pertains to preventing greenwashing remains to be seen.
Hoping to address greenwashing before it even reaches conversational AI tools is UK startup Compare Ethics, an AI-enabled platform that verifies sustainability data and then recommends whether or not a green claim is valid based on the data provided. It’s designed to support brands in minimising greenwashing, says co-founder and CEO Abbie Morris, noting that regulatory action and financial penalties for greenwashing are on the rise. “We’re in the middle of a very compelling AI revolution that will reshape retail — but at the same time, there is a green claims correction going on, which has rewritten the rules and expectations for businesses. It’s no longer good enough to just talk about sustainability, you need to think about the systems that gather and verify sustainability data. Brands need to get their own houses in order before they even think about how consumers engage.”
Morris points to a recent report by the European Commission, which found that 53 per cent of green claims give vague, misleading or unfounded information while 40 per cent have no supporting evidence. Certifications aren’t necessarily helpful either. The auditing system underpinning sustainability certifications has significant flaws and the breadth of potential certifications are almost as confusing as the green claims they attempt to simplify — the European Commission estimates that there are over 230 different sustainability labels active in the EU, each with vastly different metrics, checks and levels of transparency.
“We are reckoning with a real trust deficit,” says Morris. “If we continue to erode this trust, we are going to make it even harder for ourselves to build a society that trusts the role of businesses in sustainability.”
What does ‘sustainable’ really mean?
Beyond the data deficit, the biggest challenge for AI chatbots making sustainability recommendations is working out what ‘sustainable’ even means in the context of fashion.
Amsterdam-based software company Dayrize uses AI to help companies measure and manage their environmental and social impacts. Some of the data it uses is proprietary, shared by the brands Dayrize works with, but a lot of it is publicly available, tapping into some of the same challenges conversational AI tools will face. Co-founder Bart Nollen has learnt the hard way that making a value judgement on the sustainability of one product over another is no easy feat.
Dayrize assesses companies across five key dimensions: circularity, climate impact, ecosystem impact, livelihoods and well-being, and purpose. Each is weighted equally, to give a final score that is then comparable across products and businesses. “You need to look at all the elements of sustainability. You can’t just take carbon emissions, biodiversity or labour in isolation,” says Nollen. “That’s where the complexity lies.”
It’s a question that fashion has been trying to pin down for decades now, with little progress. Some say fashion should riff on the nutrition label promoted on food packaging, while others push for more singular metrics — shoe company Allbirds, for example, rose to popularity for publishing the carbon footprint of each product, while Asket has been sharing some variation of an “impact receipt” since 2020, detailing the carbon emissions, water and energy consumption of producing each garment, along with the emissions associated with different packaging and shipping choices.
In the EU, the closest thing to a consensus is the hotly contested Product Environmental Footprint (PEF), first introduced in 2013. However, critics argue that the PEF framework ranks synthetic fibres as more sustainable than natural ones, discounting the impact of fracking and habitat destruction while magnifying the impact of extracting raw materials, without acknowledging farmers’ livelihoods.
In the meantime, the debate around sustainable fashion — and the role of AI within it — rages on. While there are potential benefits to using AI, there are also major consequences, says Hernández Manzo, pointing to the significant environmental impacts associated with generative AI — namely the emissions and water usage racked up by data centres. “We are taking baby steps right now, but we don’t really know the impact that all of this will have,” she says. “Everything in excess is bad, so we need to be very careful about how we are using generative AI.”
Could generative AI support a circular fashion future?
While AI chatbots work on mastering sustainable fashion recommendations, other AI-driven solutions are popping up to challenge the need for new products entirely.
At the end of April, former Stanford roommates Phoebe Gates (daughter of Microsoft co-founder Bill Gates) and Sophia Kianni (a climate activist and long-time advisor to the United Nations) launched Phia in New York, a search engine extension and app, which prompts users to consider secondhand options when shopping online and scours resale platforms for the best price on their desired product.
Phia is intentionally positioned as a money-saving play rather than a sustainability one, explains Kianni. “I’ve always shopped secondhand out of concern for the planet, but we surveyed hundreds of women about their shopping behaviour and around 82 per cent said the thing they care about most is getting the best deal. The reality is that price is a bigger driver right now. Phia is a win-win because the cheaper options tend to be secondhand,” she says. The company earns affiliate commission on any purchases, she adds, but the only thing that can push certain options up the ranking is their similarity to the user’s original search.
In future, AI tools like ChatGPT and Phia might plug into other sustainability solutions like digital wardrobe apps, allowing them to make personalised solutions based on the user’s personal style, what they actually wear day to day, and undercutting new purchases as the end goal. Instead of asking “what is the most sustainable version of this product I can recommend?” conversational AI tools might ask “does the user need a new product, or can I recommend circular services like repair, upcycling, or renting instead?”
In such a fragmented industry, there could be a real benefit to AI chatbots that accompany the user through the whole lifecycle of their garment, eventually recommending more sustainable end-of-life options as well as new purchases. With a healthy dose of caution, I decided to ask ChatGPT one more question: when I am eventually ready to dispose of my white T-shirt — when it is slightly stained, the shape has loosened and it’s unsuitable for resale — what should I do with it?
The best option, according to ChatGPT, would be to repurpose or upcycle the T-shirt at home, either for cleaning rags, gardening or craft projects. This way, I would avoid incurring extra transport emissions and passing the burden onto someone else, somewhere else. Other options, it said, include donating it to verified textile recycling schemes — something the UK is pushing to expand — or finding local reuse initiatives. Maybe there’s hope for AI after all.
Sign up to receive the Vogue Business newsletter for the latest luxury news and insights, plus exclusive membership discounts.
Comments, questions or feedback? Email us at feedback@voguebusiness.com.
.jpg)
