Google, OpenAI, Amazon, Meta: the fight over how shoppers find fashion

In November 2025 Walmart put roughly 200,000 products into ChatGPT, letting shoppers buy without ever leaving the chat. By March 2026 it had pulled out. According to The Paypers, the in-chat pilot produced conversion rates roughly three times lower than when users clicked through to Walmart's own website. Within weeks OpenAI itself was retreating from the checkout model it had launched with fanfare six months earlier.

That reversal is the most telling data point in the contest between Google, OpenAI, Amazon and Meta for the moment a shopper decides what to buy. The short answer to who is winning: nobody owns fashion discovery yet, but the platforms that already sit on payment, catalogue and advertising data are pulling ahead. Amazon reports nearly $12bn in incremental annualised sales from its assistant, Meta says $60bn a year now flows through its fully automated ad tools, and Google is turning its 50 billion product listings into an agent that can buy on a shopper's behalf.

For fashion brands the stakes are not abstract. Whoever controls the conversational layer decides which products a shopper sees, which retailer gets the order, who keeps the customer data and who pays a fee for the privilege. The past 18 months have shown four different answers to that question.

What has each platform actually launched for shopping?

Google moved first in scale. At its I/O conference on 20 May 2025 it announced a shopping experience inside AI Mode that combines its Gemini models with the Shopping Graph, which Google said held more than 50 billion product listings, with more than 2 billion refreshed every hour. The same day it launched a virtual try-on in the US through Search Labs that lets users see shirts, trousers, skirts and dresses on a full-length photo of themselves, and previewed an agentic checkout: a shopper tracks a price, sets size, colour and budget, and taps "buy for me" when the price drops. On 13 November 2025 Google rolled that checkout out in the US with launch merchants including Wayfair, Chewy, the apparel brand Quince and selected Shopify merchants, and made shopping features available to all Gemini app users in the US.

OpenAI launched Instant Checkout on 29 September 2025. US users on Free, Plus and Pro plans could buy single items from US Etsy sellers inside ChatGPT, and OpenAI said more than a million Shopify merchants, naming Glossier, Skims, Spanx and Vuori, would follow. Merchants pay a fee on purchases; shoppers pay nothing extra. The feature ran on the Agentic Commerce Protocol, which OpenAI built with Stripe and open-sourced. On 24 November 2025 it added "shopping research", a mode powered by a version of GPT-5 mini that asks clarifying questions and produces a buyer's guide.

Amazon has kept the shopper inside its own walls. Its Rufus assistant gained memory, price tracking and automatic purchasing during 2025: Prime members can set a target price and let Rufus buy when it is reached, with a 24-hour cancellation window. Lens Live, launched in September 2025, puts Rufus into the camera for real-time product scanning, and Amazon reported Lens usage up 45% year on year. Its "Buy for Me" button goes further, letting Amazon buy on a shopper's behalf from a brand's own website when the item is not sold on Amazon.

Meta's approach is the least visible to shoppers and arguably the most consequential for brands' budgets. Rather than a shopping chatbot, it has automated advertising itself. Its Advantage+ suite builds, targets and places ads end to end, and its Meta AI assistant has more than a billion monthly users, according to Mark Zuckerberg.

Pinterest, the fifth contender, built its assistant on what it calls a Taste Graph. On its second-quarter 2026 call the company said the Pinterest Assistant had been made available to the vast majority of US users by the end of July, that it handles more than 80 billion searches a month, most of them visual and over half commercial, and that more than 96% of text searches are unbranded. Gen Z makes up more than half of its 640 million monthly users.

Why did OpenAI's Instant Checkout stall?

OpenAI had the audience but not the shopping habit. According to reporting summarised by Shopifreaks in March 2026, the original buy button converted at under 1%, against an industry average of 3% to 4%, and merchant adoption was limited. OpenAI's answer was to move to an app-based model in which retailers build their own storefronts inside ChatGPT and handle the transaction themselves. That shifts the development work, and the checkout, back to the merchant. Shopify said it would not build a ChatGPT app with checkout; Etsy gave no timeline.

Walmart's experience shows why retailers welcomed the change. Its assistant Sparky reached paid ChatGPT subscribers in March 2026, but this time shoppers sign in with their Walmart account, sync their basket and complete the purchase in Walmart's own system. A senior official called the earlier in-chat experience unsatisfying; Shopifreaks reported that Walmart described it as "rudimentary".

OpenAI's own product documentation reflects the caution. It says shopping research results are organic, that chats are never shared with retailers, and that the tool "might make mistakes about product details like price and availability". It also states that Instant Checkout items are not preferred in rankings, although enablement is one of the factors considered when several merchants sell the same product.

How is Google turning search into a checkout?

Google's bet is that the protocol, not the chatbot, is the prize. On 11 January 2026, timed to the National Retail Federation's annual show, it launched the Universal Commerce Protocol (UCP), an open standard covering discovery, purchase and post-purchase support. Constellation Research counted 20 supporters at launch, including Shopify, Etsy, Wayfair, Target, Best Buy, PayPal, Visa, Stripe and American Express. UCP powers a checkout inside AI Mode through Google Pay.

Google layered two commercial products on top. A Business Agent answers product questions in a brand's own voice inside search, live at launch with Lowe's, Poshmark and Reebok. Vidhya Srinivasan, Google's vice president for ads and commerce, called it "a virtual sales associate that can just answer product questions in the brand's voice." A Direct Offers pilot lets advertisers push free shipping, bundles or discounts to shoppers inside AI Mode, tested with Shopify merchants, e.l.f. Cosmetics, Petco and Samsonite among others.

The first large apparel group to go live was Gap Inc. On 24 March 2026 it began offering checkout inside Google's Gemini for Gap, Old Navy, Banana Republic and Athleta, so a shopper asking for a wedding or interview outfit can buy the suggestions in the same conversation, RetailDetail reported.

What is Amazon doing differently, and what does it earn?

Amazon is the only contender that has put a number on the sales its assistant generates. In its fourth-quarter 2025 results it said more than 300 million customers used Rufus during the year and that the assistant drove nearly $12bn in incremental annualised sales, above the $10bn pace it had given after the third quarter. Customers who used Rufus while shopping converted at rates more than 60% higher than those who did not, according to November 2025 data. These are company estimates built on Amazon's own attribution assumptions, not audited figures.

The advantage is structural. Rufus answers questions about a catalogue Amazon controls, using purchase histories Amazon owns, and the sale closes in Amazon's checkout. Buy for Me extends that logic beyond its store: Amazon said inventory for the feature grew from 65,000 products at launch to more than 500,000 by November 2025.

Where an item is not available through Buy for Me, Rufus can show a "Shop Direct" button that links to the merchant instead. Amazon also added a personalised deal finder for events such as Prime Day and Black Friday, built on browsing, basket and wish-list activity, and said Rufus received more than 50 technical upgrades in 2025. For a fashion brand that sells through Amazon, the implication is that the assistant, not the search results page, increasingly decides which variant, colour or rival product a shopper is shown. For a brand that does not sell there, Buy for Me means Amazon may still be the place where the shopper finds it, and the shopper's relationship stays with Amazon rather than with the label.

Do brands lose control when an AI agent shops for them?

The clearest test case came from small fashion and gift brands. In January 2026 Modern Retail reported that independent sellers had found their products listed on Amazon through Buy for Me without having agreed to it. Mochi Kids, a children's apparel brand in Salt Lake City, saw a catalogue of about 4,000 products appear and received around 16 Buy for Me orders. A self-reported survey by one affected owner collected 145 responses from brands that believed they had been listed without consent. Amazon said the programmes were tests, that it had received positive feedback, and that businesses could opt out at any time by email. Amanda Stewart of Mochi Kids put the brand view simply: "It builds distrust."

The episode illustrates three risks that apply to every platform. First, pricing and channel conflict: one artist told Modern Retail that wholesale-only products had been listed, raising sales tax and pricing concerns. Second, data: when an agent buys on the shopper's behalf, the brand may receive an order from a platform address rather than a customer relationship. Third, accuracy: an agent working from scraped data can show outdated sizes, prices or stock, a problem OpenAI acknowledges in its own documentation.

How does Meta's AI advertising change the economics for fashion brands?

For most fashion brands the largest AI shopping platform is not a chatbot at all but an ad system. Meta's chief financial officer Susan Li told investors in October 2025 that the annual run rate of revenue through its end-to-end automated solutions had reached $60bn. She reported that a new ranking model drove a more than 2% lift in conversions on Instagram in testing, and that rolling out its Lattice model to app ads drove a nearly 3% gain in conversions. Zuckerberg said Meta's business AIs would let tens of millions of businesses scale conversations with customers.

The trade-off is control. Automated campaigns choose creative, audiences and placements inside Meta's systems, with the brand supplying budget, product feeds and assets. Gains in conversion flow back to advertisers, but so does dependence on a model whose reasoning they cannot inspect. Pinterest is making a similar pitch from the other end: its chief executive Bill Ready told analysts that the assistant was not a "build it now and monetize it later" product, and that "AI by itself does not have style or taste. People have style and taste."

What are the limits and counter-arguments?

Three caveats temper the narrative of an AI land grab. The first is scale. Amazon's and Meta's figures are large but sit inside businesses that were already enormous; neither disclosed how much of the effect comes from apparel. The numbers are company estimates, and Amazon's rests on attribution assumptions about what customers would otherwise have bought.

The second is behaviour. The OpenAI and Walmart experience suggests shoppers use chat to research and still prefer to buy where they trust the returns policy and the size chart. That is a particular issue in fashion, where fit, fabric and returns matter more than in electronics or household goods. OpenAI itself listed electronics, beauty, home and garden, kitchen and appliances, and sports and outdoor, not clothing, as the categories where shopping research performs especially well.

The third is the counter-argument from the platforms themselves: that AI agents need retailers more than retailers need them. Google pitches its Business Agent as a way for brands to answer questions in their own voice, and Carrie Tharp, a Google Cloud vice president, told retailers at NRF in January 2026: "AI isn't retailers' competition. It should be their superpower." Gap's choice to sell inside Gemini, rather than wait, reflects a calculation that being present in the conversation is less risky than being absent from it. Brands that are well described, well stocked and well reviewed may simply be recommended more often, wherever the conversation happens.

What should fashion brands do now?

1. Treat product data as storefront Agents read feeds, not campaigns. Complete attributes for fit, fabric, size range and stock are what get a product recommended and correctly described.
2. Audit where you are already listed Check whether Buy for Me, Google's Shopping Graph or ChatGPT shopping research show your products, and whether prices, wholesale-only lines and availability are accurate.
3. Keep the final checkout where possible The Walmart pilot and OpenAI's pivot suggest merchant-controlled checkout converts better and protects the customer relationship. Favour integrations that hand the shopper to your account and returns policy.
4. Decide on protocols deliberately UCP and the Agentic Commerce Protocol are open standards with different backers. Assess fees, data sharing and who owns the customer record before enabling either.
5. Separate media and commerce budgets for AI Meta's and Google's automated ad tools and their conversational shopping features reward different inputs. Measure incrementality for each rather than relying on platform-reported attribution.
6. Test try-on and fit tools with returns data Virtual try-on is free to appear in on Google; whether it reduces returns for your categories is an empirical question your own data can answer.
7. Set rules for unauthorised listings Prepare an opt-out and escalation process, so that an agent selling your products without consent is caught quickly, not by a confused customer.