- Tension: Brands are watching their checkout numbers hold steady and reading that as safety, when the actual contest for who gets chosen has already moved somewhere they can’t see.
- Noise: The commerce press keeps promising that AI is about to check out on your behalf, which distracts from the quieter and already-real shift happening earlier in the journey.
- Direct Message: Owning the transaction is worthless if you weren’t the brand the AI decided to recommend. The contest already happened, upstream of the click.
To learn more about our editorial approach, explore The Direct Message methodology.
Seventy percent. That’s the share of AI shoppers who, according to new research from Zeta Global, still prefer to complete a purchase directly on a brand’s own website rather than transacting through an AI agent. Read quickly, that number sounds like reassurance. The storefront survives. The direct relationship holds. Whatever anxiety marketers have about AI disintermediating the checkout experience, the data says: not yet, and maybe not even close.
Read slowly, it says something else entirely.
The seventy percent that looks like nothing changed
Zeta’s study, based on a survey of 2,000 U.S. adults who reported using AI to make a purchase within the past three months, is the second installment in the company’s ongoing AI shopping research series. The headline finding is that agentic commerce, AI acting with real autonomy on a shopper’s behalf, is still emerging rather than arrived. Consumers aren’t yet handing over their credit cards to a chatbot en masse.
But the same research makes clear the purchase step was never where the real decision happened. As Zeta put it in its release, “agentic commerce may reshape how consumers discover and evaluate brands before it reshapes where transactions occur.”
AI is taking over discovery and evaluation now, while the checkout box stays the same shape it’s always been. The transaction is the last domino, and it was already going to fall the way the earlier ones fell.
That’s the part quarterly reports don’t capture. A brand can hold its conversion rate steady, hold its site traffic steady, and still be losing, because the shopper arriving at checkout was never really choosing among competitors once they got there. They were confirming a choice AI had already narrowed for them, often down to one option, several steps earlier.
Where the actual decision gets made
Fifty-four percent of AI shoppers told Zeta they’d choose a brand’s own personalized AI experience over a general-purpose AI tool, a figure that climbs to 58% among consumers aged 18 to 45. That preference matters because it means the discovery layer isn’t a neutral, single chokepoint that every brand fights over equally. Brands that have built their own AI-native experience, their own recommendation layer, their own agent, are pulling shoppers into a channel they control before a third-party AI tool ever gets the chance to make the recommendation for them.
Brands without that layer are, by default, ceding the discovery moment to someone else’s model. Once a general-purpose AI tool has made the recommendation, brand-owned checkout is just where the shopper goes to formalize a decision that already happened somewhere else.
David A. Steinberg, Zeta’s co-founder and CEO, framed the stakes plainly in the company’s release: “There’s no question consumers are increasingly trusting AI with shopping decisions,” he said. “The more important question now is whether brands are positioned to be discovered, recommended, and ultimately selected by AI.”
That is a different question than the one most marketing dashboards are built to answer.
Parents are the early signal, and the tell
The clearest preview of where this goes sits with parents. Parents with children under 18 are outpacing non-parents on every measure of AI delegation Zeta tracked: 43% would let AI make purchases within a set budget versus 27% of non-parents, 43% would let AI auto-reorder household essentials versus 31%, and 74% say AI has surfaced a brand they wouldn’t otherwise have considered, compared with 66% of non-parents.
Parents are a reasonable leading indicator here because they shop under time scarcity and repetition, the exact conditions that make delegating a decision to AI feel less like a risk and more like relief. What they’re modeling now, other segments tend to follow once trust catches up. And notably, 60% of parents say they’d pick a brand’s own AI experience over a general one, versus 49% of non-parents, which is the same discovery-layer preference showing up even more sharply in the group moving fastest.
Pamela Lord, President of Customer Relationship Management at Zeta Global, put it this way: “As AI becomes a more influential layer in the purchase journey, brands need to understand how they show up in AI-driven recommendations and create experiences that are useful enough to earn the next click.”
What GEO actually asks marketers to give up
Zeta is pitching its own answer to this, a generative engine optimization product meant to give brands visibility into how they’re surfaced across AI recommendation systems. Whether or not marketers buy that particular tool, the underlying ask is the one worth sitting with: GEO isn’t SEO with a new acronym. SEO optimizes for a page being findable. GEO requires being the kind of brand an AI system decides is worth recommending in the first place, which depends on relevance signals, first-party data quality, and trust markers that live upstream of any single page or campaign.
That’s a harder problem than keyword targeting, because it isn’t something a marketing team can fix with better copy. It requires the underlying data and reputation to actually be good, not just presented well.
The number nobody’s reporting on yet
Ninety-five percent of AI shoppers cross-check AI recommendations before buying, and 44% say they’ve hit inaccurate AI-generated shopping information along the way, according to Zeta’s data. That skepticism is, for now, a small mercy: it means the human is still checking the machine’s work. But it also means brands are being evaluated on two fronts simultaneously, whether the AI recommends them, and whether the human’s cross-check confirms that the AI was right to.
Most marketing organizations still report against channel performance and site conversion. Almost none of them report against whether their brand is the one AI systems reach for when a shopper hasn’t named them yet. That’s the metric this data points toward, and it’s the one nobody’s built a quarterly slide for.
The checkout page isn’t where brands are losing customers — it’s where they find out, a little late, whether they’d already won.
