For twenty years, e-commerce optimization has rested on one assumption: a human lands on your product page, evaluates it, and decides to buy. Every tactic built on top of that, page speed, product photography, checkout flow, review widgets, exists to influence a person standing at that specific decision point.
That assumption is breaking in real time. AI agents inside ChatGPT, Perplexity, Google, and Amazon are now discovering, comparing, and in a growing number of cases completing purchases entirely on a shopper’s behalf, often without the customer ever visiting the retailer’s actual website. The consumer describes what they want. The agent finds it, evaluates the options, and can finish checkout without a single human eye landing on your storefront.
The Scale This Has Already Reached
AI-referred retail traffic grew 393% year over year in Q1 2026, and it converts roughly 42% better than traditional search traffic. ChatGPT alone now handles an estimated 50 million shopping-related queries a day. McKinsey projects agentic commerce will reach $3 to $5 trillion globally by 2030, and Morgan Stanley’s base case puts U.S. agentic commerce at roughly $190 billion by the same year, around 10% of all U.S. e-commerce, with a bull case as high as $385 billion.
The conversion advantage isn’t a fluke of measurement. By the time a shopper clicks through from an AI conversation, they’ve typically already compared options, narrowed preferences, and built confidence through the back-and-forth itself. They arrive with meaningfully higher purchase intent than a typical search visitor still weighing multiple open tabs.
“The consumer doesn’t visit your site, doesn’t see your checkout flow, doesn’t interact with your brand at all. The AI agent does everything.”
That’s the uncomfortable part for anyone whose marketing strategy still assumes a landing page is where the relationship with a customer begins.
The Platforms Are Already Splitting Into Different Games
It’s tempting to treat “AI shopping” as one channel to optimize for. The current data says otherwise, the major platforms are settling into genuinely different roles rather than competing head-to-head on the same behavior.
ChatGPT commands the largest audience by far, roughly 900 million weekly active users, but after early friction with its Instant Checkout feature, most conversion now happens by redirecting users to merchant sites rather than completing purchases natively. Perplexity, despite a fraction of that user base at around 45 million monthly users, converts at a notably higher rate and produces an average order value 57% higher than other AI platforms, largely because its user base skews affluent and further along in the decision. Amazon’s Rufus assistant, serving roughly 300 million users, drove an estimated $12 billion in incremental sales, operating almost entirely inside Amazon’s own walled garden rather than sending traffic anywhere external. Google has gone further still, enabling autonomous “Buy for me” purchasing directly on merchant websites through AI Mode and Gemini.
The practical read: a single-platform strategy is already a losing one. Merchants relying on just one protocol are estimated to be leaving roughly 40% of available agentic traffic on the table simply by not being structured for the other major agents.
Why This Isn’t Just a Bigger Version of SEO
The instinct is to treat this the way brands treated the arrival of mobile search, adapt the existing playbook, ship it faster. Agentic commerce is a different problem, because the agent isn’t a person browsing your site, it’s a system deciding, often in seconds, whether your product is even worth surfacing at all.
Structured data has become the baseline requirement rather than a nice-to-have. Products with complete Schema.org markup are roughly 6.4 times more likely to be selected by AI agents for recommendations, a gap large enough to functionally exclude an unstructured catalog from this channel entirely, regardless of product quality or price. Beyond markup, retailers now need clean, structured product feeds compatible with the open protocols emerging across the industry, OpenAI’s Agentic Commerce Protocol, Google’s Universal Commerce Protocol, and Shopify’s Storefront MCP among them, since each platform’s agent pulls from differently structured data sources.
Attribution is genuinely harder here than almost anywhere else in digital marketing. Traditional search at least gives you a click to trace. Agentic checkout can happen in seconds with no identity-capture step at all, meaning a retailer can see that a sale occurred and which platform facilitated it, without a clear picture of why the agent chose that product over a competitor’s, or whether the sale was truly incremental. That’s a materially harder measurement problem than the “dark funnel” attribution challenges brands have already worked through with channels like podcast advertising.
Where Adoption Is Actually Ahead Right Now
Category matters more than most brands currently assume. Beauty, fashion, and apparel are leading adoption, with retailers like Ulta and Glossier already live on the emerging commerce protocols, followed by grocery, while home goods, electronics, and B2B remain further behind. Vertical leadership tracks structured catalog quality more directly than category size or brand recognition, a smaller retailer with a clean, well-tagged catalog can outperform a larger competitor whose product data isn’t agent-readable.
Trust remains the real brake on full autonomy. Around 79% of consumers rank accuracy as their top priority when using an AI shopping assistant, yet only 17% currently trust AI enough to complete a purchase entirely on its own, most people still want a checkpoint before money actually moves. That gap suggests the near-term opportunity is less about full autonomous checkout and more about winning the discovery and comparison stage, where the agent is doing real work but a human still confirms the final decision.
A Practical Readiness Checklist
- Audit your product catalog for complete Schema.org structured data, since incomplete markup can functionally exclude products from AI agent recommendations regardless of quality
- Identify which commerce protocols matter most for your customer base, rather than assuming ChatGPT’s scale alone makes it the only priority
- Build clean, agent-readable product feeds separate from what’s optimized purely for human-facing search
- Set up attribution tracking through whichever protocol integrations your platform supports, so you can at least see which agent platforms are driving sales
- Treat agentic discovery as a new, measurable channel in its own dashboard, rather than folding it into existing organic or paid traffic reporting where its behavior gets diluted
- Prioritize the product categories and catalog sections most likely to see early agentic demand based on your specific vertical, rather than treating the whole catalog as equal priority
Why This Connects to Everything Else Reshaping Search Right Now
Agentic commerce isn’t really a separate trend from the shifts already reshaping how content and products get discovered. It’s the same underlying pattern we’ve covered in how agentic AI is making AEO non-negotiable and in Google’s new AI Overviews opt-out setting, applied specifically to the point where research turns into a transaction. In every case, the deciding audience is increasingly a system evaluating structured, verifiable signals rather than a person scanning a page, and the brands treating that as one connected shift, across content, search, and now checkout, are the ones building a genuinely durable advantage rather than reacting to each platform change individually.
Once your catalog meets the basic protocol requirements, the real, ongoing advantage comes from understanding how AI engines actually rank your specific products, which queries you’re winning, which you’re losing, and what changes actually move that needle over time, rather than treating compliance as a one-time project.
