In the first quarter of 2026, AI-driven visits to U.S. retail websites increased 393% year over year.1 That number is dramatic enough to make headlines, but the stranger story is what's happening beneath it: a growing share of e-commerce revenue is being generated by purchases that leave no trace in the analytics tools merchants use to understand their business. Orders arrive. Sessions don't. The funnel, as traditionally instrumented, shows nothing.
Dina Kowalski-Bright runs e-commerce for Copper & Thyme, a mid-market specialty food retailer with roughly 2,400 SKUs, a Shopify storefront, and a brand built on editorial content, recipe integration, and the kind of visual merchandising that makes people feel something about harissa paste. She's been in digital commerce since 2014 and has survived enough paradigm shifts to be appropriately tired of the phrase "paradigm shift." But what she's seeing in her dashboard this spring is, by her own admission, different.
We should note: Dina is a composite character, assembled from the real strategic tensions facing mid-market e-commerce directors in 2026. She is fictional. Her problems are not.
You've described what you're seeing in your analytics as "ghost orders." What does that actually look like?
Dina: Around November, we started getting orders — fulfilled, paid, real revenue — with no corresponding session in GA4. No product page view, no add-to-cart, no checkout funnel entry. Just an order. Tagged from ChatGPT or Gemini. I spent two weeks with our analytics vendor trying to fix what I assumed was a tracking bug before someone explained there was nothing to fix. The purchase happened inside a conversation. The customer never visited our site.2
How much revenue?
Dina: Small. Maybe 3% of total. But it's growing at a rate that makes the absolute number feel beside the point. Shopify says AI-attributed orders grew 11x between January 2025 and March 2026.3 And here's the part I keep circling back to: the conversion rate on AI-referred traffic is now 42% better than paid search.4
So the channel I can barely measure is outperforming the channel I spend the most money optimizing. I don't have a playbook for that.
Your brand was built on storytelling. Recipe content, origin stories, the whole editorial apparatus. What happens to that when the buyer is software?
Dina: Yeah. This is the part that makes me want to lie down on the floor.
We spent five years building this content engine. Four hundred recipes. A James Beard-nominated food writer doing quarterly features. Beautiful photography. And an AI shopping agent does not care about any of it. It queries structured product feeds. It evaluates attributes: price, availability, dietary tags, shipping speed. If our harissa paste has five structured attributes and a competitor has twenty, the agent doesn't deliberate. It just picks them.5
I had this moment in February where I realized the thing I'm best at — brand storytelling — might be the thing that matters least in the channel growing fastest. Fun realization to have at 2 PM on a Tuesday.
So do you optimize for agents or for people?
Dina: Everyone frames this as a choice, and I think the framing misses what's actually happening. The agent is becoming the first decision layer. It determines which brands earn consideration at all.6 If your product data isn't structured for agents to read, you don't make the shortlist. Then the human never sees you either.
So it's agents first, then people. Which means product data quality is now the most important brand asset we have. More important than photography. More important than content. More important than the website redesign we just spent $180K on.
That's a hard sell internally.
Dina: A terrible sell. Try telling your CMO that the most strategic investment this quarter is cleaning up product feed attributes. Try explaining that "machine-readable harissa metadata" is the new brand equity. People look at you like you need a vacation.
But most brands score below 25 out of 60 on agentic readiness audits.7 That gap is enormous. The protocol access is basically free now — UCP is open, Shopify integrated it automatically in March.8 But protocol access without clean data is like having a storefront on the busiest street in the world with nothing on the shelves.
The Universal Commerce Protocol went live in January. What does that actually change for someone at your scale?
Dina: On paper, everything. A merchant builds to the protocol once and every AI agent that speaks UCP can discover your inventory, check stock, apply discounts, complete checkout.9 A small Shopify store gets the same protocol-level access as Walmart. That part is real.
In practice, it surfaces a problem we've been ignoring for years: our product data is a mess. Inconsistent across channels, missing attributes, pricing that doesn't sync in real time. We were getting away with it because humans are forgiving. Humans click around, read the description, figure it out. Agents are literal. If the data says "out of stock" because someone didn't update a feed, you're invisible. Not penalized. Invisible.
What about brand loyalty? You have a subscription program, repeat customers who love you. Does that survive agent mediation?
Dina: I oscillate between optimism and dread on this one, sometimes within the same sentence. UCP has an identity linking capability where a customer buying through Google AI Mode can carry over their membership pricing, loyalty perks, all of it.10 Good. But it has to be set up deliberately. It doesn't happen by default. If it's not configured, that customer buys through an agent, we fulfill the order, and they never enter our CRM. We served them and learned nothing.
The deeper worry is that personal AI agents will start managing shopping lists. Comparing prices automatically. Switching retailers based on structured attributes. Loyalty in that world has nothing to do with emotional connection to a brand. It's about whether your data feed matches the user's dietary profile better than the next option.
That sounds bleak for a brand builder.
Dina: It sounds bleak if you think brand lives on the website. But I'm starting to wonder whether brand has to migrate. Bain found that consumers trust brands' on-site AI agents three times more than third-party agents.11 There's still a trust premium attached to the name. It just expresses differently.
Maybe brand becomes the thing that makes a customer tell their agent, "Buy the Copper & Thyme one." That's a different kind of loyalty. Earned upstream, expressed as a constraint on the agent's decision space.
I don't know if that's right. I'm genuinely working this out as I go. But I know that sitting still is the worst option. AI platforms account for maybe 1.5% of total retail e-commerce this year.12 Tiny. But the brands that figured out Google SEO in 2010 had a massive advantage over those who waited until 2018. I think this is that moment. And I'd rather be wrong and early than right and late.
The interview was conducted over video call. Dina's office featured a wall of Copper & Thyme product packaging and a whiteboard with DATA QUALITY circled three times in red marker, which she acknowledged was "either my strategic roadmap or a cry for help."
Footnotes
-
Adobe Analytics, Q1 2026 data covering more than one trillion visits to U.S. retail sites. https://www.gncrypto.news/news/ai-driven-traffic-us-retail-sites-393-q1-2026/ ↩
-
AI shopping agents complete purchases through API calls, not browser sessions — no JavaScript fires, no cookies are set, no thank-you page loads. https://www.digitalapplied.com/blog/ai-agent-commerce-revenue-attribution-guide-2026 ↩
-
Shopify data, January 2025–March 2026. https://shopify.engineering/ucp ↩
-
Adobe Analytics, March 2026. In March 2025, AI referrals converted approximately 38% worse than standard channels; by March 2026, AI traffic was converting 42% better. https://www.gncrypto.news/news/ai-driven-traffic-us-retail-sites-393-q1-2026/ ↩
-
Structured data density determines agent selection. If a competitor exposes more semantic attributes, the agent selects them through data availability, not deliberation. https://www.thron.com/en/blog/insights/universal-commerce-protocol/ ↩
-
Microsoft Industry Blog, February 2026. https://www.microsoft.com/en-us/industry/blog/retail/2026/02/09/how-agentic-commerce-is-becoming-the-new-front-door-to-retail/ ↩
-
Agentic readiness audits score brands across six dimensions: structured data quality, machine-readable product feeds, transparent pricing signals, review/trust signals, brand authority signals, and API accessibility. https://clickvoyant.com/ai-shopping-agents-the-strategic-guide-to-agentic-commerce-in-2026/ ↩
-
UCP announced at NRF 2026 on January 11, 2026; Shopify's Agentic Storefronts activated automatically on March 24, 2026. https://developers.googleblog.com/under-the-hood-universal-commerce-protocol-ucp/ ↩
-
Mastercard, January 2026, on UCP's single-standard integration model. https://www.cahoot.ai/universal-commerce-protocol-agentic-commerce/ ↩
-
UCP Identity Linking capability, the only feature in UCP's stable release as of May 2026. https://www.getpassionfruit.com/blog/what-is-google-s-ucp-update-carts-catalogs-and-loyalty-in-ai-shopping ↩
-
Bain & Company, consumer trust data on brand-owned vs. third-party AI agents. https://www.deptagency.com/insight/how-cpg-brands-will-drive-discovery-loyalty-commerce-in-2026/ ↩
-
eMarketer projects AI platforms will account for 1.5% of total retail e-commerce sales in 2026, or $20.9 billion. https://www.emarketer.com/content/ai-shopping-tools-gain-traction-retailer-pushback-could-cloud-2026-progress ↩