For most of the internet era, retailers have had a familiar bargain with digital intermediaries: accept the traffic, pay the toll and try not to lose control of the customer relationship in the process.
Artificial intelligence is creating a new version of that bargain — but this time retailers are entering it with more experience and more leverage.
Consumers are increasingly using ChatGPT, Google Gemini and other AI systems to research products, compare options and narrow purchase decisions. The traffic flowing out of those systems is no longer experimental or low-intent. Adobe Digital Insights reported that AI-referred traffic to U.S. retail sites rose 393% year over year in the first quarter of 2026. In March, those visits converted 42% better than non-AI traffic, spent 48% longer on retail sites and viewed 13% more pages per visit.
The trend held through the summer. During the June Prime Day period, Adobe found AI traffic to retail sites was up 89% from a year earlier and converted 40% better than channels such as paid search, email and social media. Reuters reported Friday that 41% of U.S. consumers used generative AI for online shopping in June and that AI-referred visitors generated 41% more revenue per visit than shoppers arriving through traditional channels.
That is the point at which AI shopping stops looking like a novelty and starts looking like a distribution channel.
Discovery is moving before the transaction does
The most important shift may be where the customer makes the first meaningful decision.
A shopper who once began with a Google query, marketplace search or retailer homepage can now start with a much more specific prompt: the best carry-on for an international trip under $300, a moisturizer for sensitive skin without fragrance, or a television that fits a particular room and works well in bright light.
The AI system does more than retrieve products. It can interpret constraints, compare features, synthesize reviews and reduce a large catalog into a short list. By the time the shopper reaches a retailer, much of the consideration process may already have happened somewhere else.
That helps explain why AI referrals are converting so well. They are not necessarily replacing the store. They are increasingly doing the work that happens before the store visit.
For retailers, that makes visibility inside AI recommendations strategically important. Walmart, Ulta Beauty, Wayfair and other companies are adapting product content and digital infrastructure so their assortments can surface accurately inside conversational systems. Reuters reported that Ulta is already seeing roughly twice the conversion and intent from shoppers arriving through Gemini and ChatGPT.
But retailers are drawing a line between being discoverable and becoming invisible.
The fight is over the handoff
Retailers have good reasons to want the transaction to remain on their own rails.
A completed purchase is not only revenue. It adds to a retailer’s understanding of browsing behavior, basket composition, price sensitivity, loyalty, repeat purchase patterns and lifetime value. It can connect a transaction to a rewards program, customer-service history, app behavior or store visit. Those signals improve future merchandising and personalization — and they become more valuable as retailers build their own AI systems.
Ulta’s position captures the tension. The beauty retailer is working with Google on AI-powered shopping experiences that can incorporate its rewards program, according to Reuters, but still prefers customers to complete purchases through Ulta. The objective is not to keep AI out of the journey. It is to keep the relationship intact after AI helps create the demand.
The major AI platforms are beginning to acknowledge that boundary as well.
In March, OpenAI expanded its Agentic Commerce Protocol around product discovery and said the initial version of its Instant Checkout model had not provided merchants with enough flexibility. The company emphasized merchant-controlled checkout options while expanding direct product feeds and discovery integrations with retailers including Target, Sephora, Nordstrom, Lowe’s, Best Buy, The Home Depot and Wayfair.
Walmart went a step further, introducing a tailored environment inside ChatGPT that supports account linking, loyalty and Walmart payments. The retailer can participate inside the conversational interface without reducing itself to an anonymous SKU provider.
Google is moving in a similar direction. Its Universal Cart, announced in May, is designed to work across Search, Gemini, YouTube and Gmail. Consumers can check out with participating merchants through Google Pay or transfer items to the merchant’s site. Google explicitly says the brand remains the merchant of record.
That architecture matters. The emerging model is not simply an AI company swallowing ecommerce whole. It is a negotiation over how much of the journey each layer controls.
Retailers have seen this movie before
Retailers spent the past two decades learning that a distribution channel can become a dependency.
Search engines could send enormous traffic while changing the rules of visibility. Marketplaces could create incremental sales while training customers to begin with the marketplace instead of the brand. Social platforms could create demand while keeping much of the audience relationship inside their own systems.
AI has the potential to become more powerful than any of those intermediaries because it can compress discovery, comparison and recommendation into one interface — and eventually take action on the shopper’s behalf.
That makes the early retailer response notable. Rather than choosing between participation and resistance, many are trying to design a third option: make product data available to the agent, let the agent create qualified demand, then preserve the merchant’s identity, transaction systems and customer relationship at the point of conversion.
In other words, retailers want AI to function more like a high-intent distribution layer than a replacement storefront.
Product data is becoming distribution infrastructure
There is another consequence for brands that may be easier to miss.
In conventional search, a retailer could optimize a category page around keywords, backlinks and technical SEO. Conversational shopping asks different questions. An AI assistant needs enough structured and descriptive information to understand who a product is for, what distinguishes it, what constraints it satisfies and whether price, inventory and specifications are current.
Adobe found that large portions of U.S. retail sites remain difficult for machines to interpret completely, even as AI-driven traffic grows. That creates a new kind of merchandising problem: a product can exist online, rank reasonably well in conventional search and still be poorly represented when an AI system is asked to compare it with alternatives.
OpenAI and Google are both responding with commerce protocols designed to carry product feeds, pricing, availability and transaction information between merchants and AI systems. Those standards may look like technical plumbing, but they are becoming part of distribution strategy.
The product catalog is no longer only the database behind the website. It is increasingly the source material from which machines decide what to recommend.
The customer relationship is the scarce asset
The near-term winners in AI commerce may not be the companies that automate the most steps of checkout. They may be the companies that make themselves easiest for agents to understand while preserving enough direct interaction to continue learning from the customer.
That balance will differ by category. A commodity purchase may move easily through an agent with little brand interaction. Beauty, apparel, furniture, luxury and other considered purchases may give retailers more reason to pull shoppers into their own environments for richer recommendations, loyalty benefits and service.
But the strategic question is becoming consistent across categories: Who owns the next decision after the recommendation?
If an AI system controls discovery, comparison and ultimately the transaction, the retailer risks becoming fulfillment infrastructure behind someone else’s interface. If retailers make AI useful only after a shopper arrives on their own site, they risk missing a rapidly growing source of high-intent demand.
The emerging compromise is more interesting than either extreme. AI can own more of discovery without necessarily owning the merchant relationship. Retailers can expose more of their catalogs and capabilities without surrendering every customer signal.
That is why the current fight over checkout, loyalty and data is consequential. AI shopping is not waiting for some future moment when autonomous agents purchase everything on our behalf. It is already reshaping where buying decisions begin.
The companies that understand that distinction early will not simply optimize for AI traffic. They will decide which parts of the customer journey they are willing to let the new middlemen own — and which parts are too valuable to give away.