The interface is changing
The way people shop online is being redesigned in real time, and the people doing the redesigning are not retailers. They are AI laboratories building shopping agents that sit between the buyer and every seller on the internet.
Fifty-two percent of US consumers now use AI when shopping, according to the IAB. ChatGPT passes nine hundred million weekly users. Amazon's Rufus has surpassed two hundred and fifty million monthly users. By 2028, McKinsey projects that seven hundred and fifty billion dollars in US revenue will funnel through AI-powered search. AI already mediates commerce.
What is less understood is how comprehensively the shift rewires every layer of the ecommerce system: discovery, advertising, product data, brand equity, validation, retail media, measurement and the economics of categories. The search box is becoming an answer. The product page is becoming a data feed. The checkout is becoming an API.
This whitepaper examines the evidence already available and draws the implications for brands, retailers and platforms. It does not predict. It reports what is happening, what the data shows, and what that means for anyone selling goods through digital commerce.
AI already mediates commerce.
Inflexion
The shopper has changed. The data is unequivocal
The most striking single data point is from Adobe: AI-referred traffic to US retail sites grew 1,324% since October 2024. In March 2026 alone, AI traffic to US retail sites was up 269% year on year. By May 2026, it was still growing at 138% YoY. This is not a niche behaviour.
Shopify reports that AI traffic to its merchants' stores is up seven times and AI-attributed orders up eleven times since January 2025. Orders from AI searches rose nearly thirteen times in early 2026. The early narrative that AI traffic converts poorly has reversed: Adobe's 2026 data shows AI traffic now converts comparably to non-AI traffic.
The shopper who arrives via AI is not a lower-quality lead. The shopper who arrives via AI is simply a shopper. The technology that brought them has become a primary discovery mechanism.
During Cyber Week 2025, Salesforce found that 20% of global orders were influenced by AI agents. One in five purchases already involved an AI somewhere in the decision chain. That number will not shrink.
The early narrative that AI traffic converts poorly has reversed. AI traffic now converts comparably to non-AI traffic.
Adobe, 2026
The consideration set becomes the battleground
Search advertising worked because a shopper who typed a query was actively considering a set of options and the advertiser could insert into that set. AI answers change this. When a user asks ChatGPT or Google's AI Mode for a product recommendation, the answer is a curated list of three to five options. The consideration set is determined by the model, not by the shopper's own browsing.
This has a measurable effect. Similarweb data from 2026 shows that Google AI Mode generates referral traffic on only 1.6% to 2.5% of queries, compared with the 17% to 19% referral rate from traditional Google search. The zero-click rate has risen to 64.82% in 2026, up from roughly 50% in 2019. More shoppers get the answer without ever visiting a site.
Being left out of the AI answer is not the same as ranking low on a search results page. It is invisibility. McKinsey estimates brands may see 20% to 50% traffic decline from traditional search as AI-mediated discovery expands. The consideration set is not being rearranged. It is being replaced.
Being left out of the AI answer is not the same as ranking low on a search results page. It is invisibility.
Inflexion
The consideration set is not being rearranged. It is being replaced.
Inflexion
From search advertising to AI-mediated advertising
For twenty years, search advertising was simple: the shopper typed a keyword and the advertiser bought the keyword. AI answers do not surface keyword-matched ads in any recognisable form. They generate natural language responses that incorporate product recommendations, and increasingly, those recommendations can be sponsored.
Google is testing new ad formats integrated directly into AI Mode responses, announced May 2026. Google AI Max for Shopping, launched April 2026, offers text customisation, flexible user engagement and optimal format selection. Amazon's Rufus exited beta for Sponsored Products and Brands Prompts in March 2026, operating on a billable CPC model within the conversational interface.
These are not experiments. These are the foundations of a new advertising model in which the ad unit is not a slot on a page but a recommendation in a conversation. The advertiser no longer buys a keyword. They buy inclusion in a generated answer.
The advertiser no longer buys a keyword. They buy inclusion in a generated answer.
Inflexion
Product data becomes media infrastructure
An AI shopping agent does not read a website. It reads structured data: product feeds, schema markup, merchant centre listings, API responses. If a product's attributes are not machine-readable, the product does not exist in the AI's consideration universe.
Google's Shopping Graph contains over sixty billion product listings. When Gemini searches for a product, it queries that graph, not the open web. Amazon Rufus draws from Amazon's own catalogue. OpenAI's ACP (launched September 2025) connects ChatGPT directly to merchant product feeds. The infrastructure of commerce is becoming the data feed, not the product page.
This is why Google launched the Universal Commerce Protocol in January 2026, with checkout rolling out across the US, Canada, Australia and the UK. UCP is a standardised checkout API that allows any AI agent to complete a purchase on any participating merchant's site without the shopper leaving the conversation. Amazon's Buy for Me, launched April 2025, already does this: the agent buys from a brand's own website on the shopper's behalf.
If a product's attributes are not machine-readable, the product does not exist in the AI's consideration universe.
Inflexion
The infrastructure of commerce is becoming the data feed, not the product page.
Inflexion
Brand before the prompt
An AI agent can only recommend a brand it recognises from its training data, its knowledge graph or its retrieval corpus. If a brand is not present in the sources the model trusts, it will not appear in the answer. This is not an SEO problem. This is a brand-awareness problem mediated by a language model.
The implication is that brand building has become an AI visibility strategy. The brands mentioned in authoritative editorial coverage, verified retailer pages, community discussions and creator content are the brands that appear in answers. A brand that has invested in awareness, earned media and structured presence across the web is a brand the model can find.
Gartner's January 2026 survey found that only 11% of US consumers are willing to let AI make purchase decisions entirely, while 31% are willing to let it narrow choices for household supplies. Trust in AI recommendations is still being earned. Brands that build recognition and credibility outside the AI channel bring that trust into the AI answer when the model selects them.
If a brand is not present in the sources the model trusts, it will not appear in the answer. This is not an SEO problem.
Inflexion
Brand building has become an AI visibility strategy. The brands the model can find are the brands that appear in the answer.
Inflexion
The validation economy: trust is the currency
The AI recommendation is not the end of the journey. It is the beginning of a validation phase in which the shopper checks, compares and confirms before buying. Gartner's November-December 2025 survey found that 54% of AI shoppers had to double-check the accuracy of all information provided, and 62% said the AI-provided information was a waste of time.
These numbers describe a trust gap that commerce infrastructure must close. The brands that win in this environment are those that not only appear in the AI answer but also provide the validation content the shopper needs to say yes: reviews, creator content, retailer listings, comparison data. Each validation surface is a moment where a brand can be confirmed or rejected.
This creates what can be called a validation economy, in which the brands that own the best evidence across reviews, social proof, editorial coverage and retailer presence capture the conversion. Trust is not a brand attribute. It is a media asset that must be built, maintained and surfaced at the moment of doubt.
Trust is not a brand attribute. It is a media asset that must be built, maintained and surfaced at the moment of doubt.
Inflexion
Each validation surface is a moment where a brand can be confirmed or rejected.
Inflexion
Retail media after the search result
US retail media spend is projected to reach $71.09 billion in 2026, according to eMarketer. The growth has been driven by the logic that the retailer owns the transaction and can monetise the path to it. But if the transaction moves to an AI agent that does not land on the retailer's search results page, the economics of retail media shift.
Amazon Rufus now shows Sponsored Products and Brands within the conversational AI interface. The ad unit is not a search result slot. It is a recommendation in a dialogue. Google UCP and OpenAI ACP create checkout experiences that occur outside the retailer's owned environment entirely. The retail media impression that used to happen on a product listing page now happens inside an AI conversation, and the measurement framework must follow.
The retailer's control over the shopper's attention erodes as AI agents intermediate. The sponsored slot on a search results page is replaced by the sponsored mention in an AI answer. The auction moves upstream, from the search query to the training data and retrieval corpus that determine the AI's recommendation.
The auction moves upstream, from the search query to the training data and retrieval corpus that determine the AI's recommendation.
Inflexion
From buying impressions to influencing decisions
The payment infrastructure is following the AI agent. Visa launched Intelligent Commerce Connect in April 2026, working with OpenAI ACP, Google UCP and Mastercard's Multi-Party Payments protocol. Mastercard Agent Pay, launched in 2025, is already working with Microsoft Copilot Checkout. The checkout that was once a secure page on a retailer's site is becoming an API call from an AI conversation.
OpenAI's Instant Checkout, launched alongside ACP, was retired in March 2026 after the company pivoted ACP to focus on product discovery rather than full checkout. The direction of travel is clear: every major card network and AI platform is building the infrastructure for agent-initiated payments. The purchase decision is being unbundled from the purchase execution.
Marketing is no longer about buying impressions that lead to a checkout page. It is about influencing the AI recommendation before the payment API is called. The media buy becomes a decision-influence buy. The impression becomes a recommendation signal.
The checkout that was once a secure page on a retailer's site is becoming an API call from an AI conversation.
Inflexion
Measurement breaks. Then it reassembles differently
Last-click attribution was already inadequate before AI. In an AI-mediated journey, it becomes meaningless. The shopper may discover a product through a ChatGPT recommendation, validate it through a creator video, check reviews on a retailer site, then purchase through an agentic checkout. No single touchpoint is the last click.
The measurement problem is compounded by the fact that AI agents do not provide referrer headers or session IDs in the way browsers do. When a user purchases through Google UCP or Amazon Buy for Me, the platform knows the attribution. The brand does not. The data asymmetry between AI platforms and advertisers widens.
This will drive investment in new attribution models: multi-touch measurement that accounts for brand presence in AI training data, citation frequency in AI answers, validation surface engagement, and downstream conversion. The brands that invest in connected measurement will understand the new journey. Those that wait for platforms to provide the answers will operate blind.
When a user purchases through Google UCP or Amazon Buy for Me, the platform knows the attribution. The brand does not.
Inflexion
The data asymmetry between AI platforms and advertisers widens. Those who wait for platforms to provide the answers will operate blind.
Inflexion
Category changes everything
Not all categories are equally affected by AI-mediated commerce. The categories where AI has the strongest impact are those with structured attributes, high research intensity and repeat purchase behaviour. Categories like beauty, consumer electronics, home goods, groceries and apparel are early movers because their product data maps naturally to AI recommendation logic.
Categories with low differentiation or commodity pricing may compress further as AI agents optimise for price and availability. Categories with high differentiation or strong brand equity may see a premium on brand recognition as a recommendation signal. The AI agent's recommendation algorithm becomes a category structure: it determines which attributes matter, which brands get surfaced and which products win.
McKinsey found that 50% of consumers now use AI-powered search generally. But the usage varies wildly by category. In categories where the AI answer is trusted, the brand that owns the answer owns the sale. In categories where the AI answer is distrusted, the validation economy matters more. Category strategy becomes AI strategy.
What marketers need to change
The evidence in this report points to seven areas where marketing operating models must adapt:
Invest in structured product data. If your product attributes are not machine-readable, your products do not exist in AI discovery. Feed quality is the new SEO.
Build brand recognition as an AI signal. Brand awareness is not a soft metric. It determines whether the model recommends you. Invest in owned, earned and paid media that builds the brand into the AI's knowledge base.
Own the validation surfaces. Reviews, creator content, retailer pages and community discussions are where the AI recommendation is confirmed or rejected. Invest in them as conversion channels.
Prepare for agentic checkout. UCP, ACP, Buy for Me and Agent Pay are not experiments. They are infrastructure decisions being made now. Ensure your payment and checkout systems can participate.
Rethink retail media. If the transaction moves off the retailer's search results page, the retail media buy must move to AI-native ad formats. Rufus, AI Mode and AI Max are the new shelf.
Invest in connected measurement. Build attribution models that span AI citation, validation engagement and agentic purchase. The platform will not provide this data. You must build it.
Treat category as strategy. The degree of AI mediation varies by category. Understand how AI affects your specific vertical and build the response from that understanding.
The emerging AI commerce system
Putting the pieces together, an AI commerce system is emerging with distinct layers. Each layer has its own economics, its own competitive dynamics and its own implications for brands.
The system is not complete. Many pieces are still being built. The direction is clear enough that waiting for stability carries worse odds than acting on what is known.
Brands that invest now in structured data, brand recognition across AI sources, validation surface ownership, connected measurement and AI-native advertising will be positioned for the system that arrives. Brands that wait for the search engine to come back will find it has become an answer engine, which has become a shopping agent, which has become the store.
Brands that wait for the search engine to come back will find it has become an answer engine, which has become a shopping agent, which has become the store.
Inflexion
Inflexion PerspectivesAnalysis from the Inflexion team on AI search, media and measurement.
