The CMO's Guide to the New AI Funnel

AI-referred traffic to U.S. retail sites grew 393% year over year in Q1 2026, and Adobe found that it converted 42% better than non-AI traffic in March.¹ The numbers are still small, but the pattern is strategic: more shoppers are letting AI systems research, compare, and narrow the field before they ever reach a merchant site.

That changes the job for marketing leaders. The old funnel was built around human clicks, cookies, sessions, and retargeting. Agentic commerce moves the decisive moment upstream, into a consideration layer most commerce teams cannot yet see.

Twenty Years of Funnel Infrastructure

The digital commerce funnel took two decades to instrument. Cookies enabled session tracking. Google Analytics revealed where shoppers dropped off. Attribution models identified which channels drove purchases. Cart abandonment flows pulled hesitant buyers back in. Every stage of the journey, from discovery to checkout, became measurable and optimizable.

That infrastructure was built on a core assumption: the shopper is a person using a browser, leaving a trail of clicks, page views, and behavioral signals at every step. For twenty years, the assumption held.

A New Traffic Source Breaks the Model

AI is now the front door to commerce for a rapidly growing share of shoppers. They ask an AI assistant what to buy, get a recommendation, and click through to a retailer's site. And this traffic converts: Adobe's data shows AI-referred visitors convert at 42% higher rates than those arriving from paid search, email, or other traditional channels.¹

But here's what makes this different from every prior channel shift. When a human shopper visits your site through Google, they carry a cookie. They have an account. They build a browsing history. Your analytics stack knows who they are, what they looked at, and where they fell out of the funnel.

When an AI agent visits your site, it carries none of that. It has no cookies, no account, and no browsing history. It arrives, evaluates your catalog, compares you against competitors, and shapes a recommendation for the human who asked. Your analytics platform registers a visit, but it can't tell you what the agent compared you against, what it prioritized, or why it chose someone else.

The fastest-growing, highest-converting traffic source in commerce is also the one your measurement stack understands least.

To be clear: AI traffic is still a small share of total retail visits today. But the per-visit economics are already disproportionate, with better conversion and deeper engagement than traditional channels.¹ Mobile commerce looked similar in 2012: a fraction of total volume, an obvious trajectory. The brands that waited for mobile to be "big enough" to justify investment lost years to the ones that built for it early. AI traffic is following the same curve, and the time to get ahead of it is now.

Agents Don't Shop Like People

This would matter less if agents were simply faster versions of human shoppers running through the same decision process at machine speed, but they are not. Agents shop in a fundamentally different pattern.

A human shopper typically visits one site, browses a few pages, and makes a decision. An agent does the opposite. HUMAN Security's 2026 benchmark found that 77% of agentic AI activity occurred on product and search pages, and that retail and e-commerce led agentic traffic by vertical.³ The agent is not following your funnel. It is running its own process across the competitive set.

That process produces different outcomes than a human would reach. Researchers at Columbia Business School and Yale tested how AI agents make purchasing decisions and found that different AI models favor different products, even when given the exact same choices.⁴ Some models weight price more heavily. Others over-index on ratings or where a product appears on the page. The preferences vary so much from model to model that the same product catalog can produce completely different "best picks" depending on which AI the consumer happens to be using.

MIT Media Lab researchers found something similar: when they changed product details like prices, ratings, and persuasive cues, agents shifted their recommendations in ways that were measurable and consistent, but different from how a human shopper would respond to the same changes.⁵ In short, the decision-making patterns that marketers have spent decades studying in human shoppers do not apply to agents.

The human behavioral data that powers every personalization engine and recommendation system in commerce today will not predict what an agent does. Agents are a new shopper class that needs its own behavioral dataset, and that dataset does not exist yet.

Generative Engine Optimization Gets You Found. It Doesn't Get You Chosen.

The industry's current response to this shift is Generative Engine Optimization, or GEO: making your site readable to AI systems so your products show up in AI-generated recommendations. It matters. Adobe's research found that roughly a quarter of homepage content and a third of product page content across U.S. retail sites is not yet machine-readable.¹ If an AI can't parse your catalog, you won't be in the answer.

But readability is a floor, not a strategy. Being included in an AI's research is not the same as winning its recommendation. An agent might evaluate five merchants, compare prices and ratings across all of them, and surface two or three to the consumer. The merchants who didn't make the shortlist never know they were in the running. There is no cart abandonment email equivalent, no retargeting pixel, no exit survey. The agent evaluated them and moved on, leaving no signal behind.

GEO solves discoverability, but it doesn't solve engagement. And in a world where AI agents are compressing the funnel from dozens of clicks to a single shortlist, engagement at the point of evaluation is where the outcome is decided.

What CMOs Should Measure First

The first job is not to rebuild the entire commerce stack. It is to create a scorecard for the part of the funnel that agents now control: whether your brand is found, understood, compared fairly, and included before the human shopper arrives.

Why Engagement, and Why Now

Marketing teams have a hundred priorities. Here is why this one should move up.

Among AI-active shoppers, 53% use LLMs to decide where to buy.⁶ The agent narrows the field. The human picks from what's left.

This is the only channel where something other than the consumer is deciding which merchants make the shortlist. In every other channel, the shopper browses, compares, and narrows the options themselves. In this one, the agent does that work and presents a curated set of choices. If your store isn't in that set, the consumer never sees you. The purchase decision still belongs to the human, but the agent controls who gets a seat at the table.

Think of it this way. GEO is having a clean window display: your products are visible, your prices are readable, and the agent can see what you sell. That's necessary, but it's passive. Engagement is having someone at the door who recognizes the shopper, reads what they're looking for, and helps them find it faster than the store next door. One is a static page; the other is an active response to a visit that's happening right now.

Agents are the new window shoppers. They're walking past your storefront dozens of times a day, comparing what's in your window to the shop next door, and making a recommendation before your door even opens. Most merchants today offer every agent the same static catalog, regardless of what the agent is looking for or how it got there. The merchants who recognize an agentic visit in real time, understand what the agent is evaluating, and respond with relevant content, pricing, or offers during the visit itself will be the ones the agent recommends.

The difference is measurable. Research on generative engine optimization found that content changes can boost visibility in generative-engine responses by up to 40%.⁷ That isn't about having readable HTML alone. That's about what an AI system can extract, cite, and trust when it evaluates your surface. A well-timed offer, a cleaner product comparison, a faster path to the information the agent needs: these are the kinds of interactions that change the output.

There is a second reason the timing matters. AI influence is still undercounted in attribution today: Contentsquare's 2026 benchmark put AI-referred traffic at 0.2% of total visits, even as that channel grew 623% year over year.⁸ That will change as autonomous checkout infrastructure matures, but the relationships between agents and merchants are being established right now. The merchants who learn how agents evaluate, compare, and recommend during this window will have a durable advantage when the transaction layer arrives. Everyone else will be starting from zero against competitors who have been building for years.

The Three-Layer AI Funnel Readiness Framework

The practical starting point is not autonomous checkout. It is the consideration layer: recognizing agentic visits, seeing what they evaluated, and learning when your brand was considered but not chosen. The playbook has three layers.

The commerce funnel was built for human shoppers. AI agents are already reshaping it. The brands that treat agentic engagement as a priority today will define the playbook. The ones that wait will be playing catch-up in a channel that has already chosen its favorites.

Footnotes

1. Adobe Digital Insights, "Adobe report: U.S. retailers see surge in AI traffic, but many websites are not entirely readable by machines," April 2026.

2. Eight Oh Two, "2026 AI & Search Behavior Study."

3. HUMAN Security, "2026 State of AI Traffic & Cyberthreat Benchmark Report."

4. Allouah, Besbes, Figueroa, Kanoria, and Kumar, "What Is Your AI Agent Buying? Evaluation, Biases, Model Dependence, and Emerging Implications for Agentic E-Commerce."

5. Cherep, Ma, Xu, Shaked, Maes, and Singh, "A Framework for Studying AI Agent Behavior: Evidence from Consumer Choice Experiments," ICLR 2026.

6. Rithum, "A brand your customer had never heard of just won the sale," April 2026.

7. Aggarwal, Murahari, Rajpurohit, Kalyan, Narasimhan, and Deshpande, "GEO: Generative Engine Optimization," KDD 2024.

8. Contentsquare, "What is AI-referred traffic, and why is it worth watching?", April 2026.

9. SparkToro and Datos, "New Research: We analyzed 332 million queries over 21 months to uncover never-before-published data on how people use Google," December 2024.