Every agentic commerce headline this year has led with the traffic number, because it's the exciting one. Fewer have led with the conversion number, because it's the one that actually determines whether any of this traffic is worth chasing yet. A report published August 17 by ResearchAndMarkets, distributed via GlobeNewswire, puts both numbers in the same sentence for once, and the gap between them is the actual story.
The numbers behind the gap
The report frames the gap explicitly as an infrastructure problem rather than a demand problem: consumer appetite for AI-assisted shopping is real, growing fast, and running well ahead of what merchant systems are built to handle when an agent, rather than a person clicking through a UI, is the one attempting to complete a purchase. Kearney's AlphaWise survey adds independent corroboration on the demand side: 60% of shoppers expect to use an AI agent to shop within the next 12 months, and 73% already report familiarity with the tools involved. This is not a speculative future channel. It's a channel with real, measured traffic that most merchant stacks currently mishandle at the point of transaction.
| METRIC | FIGURE | WHAT IT IMPLIES |
|---|---|---|
| Traffic growth | 805% | Demand is arriving now, not in some hypothetical future |
| Adoption | 39% of consumers | Not a niche behavior; approaching mainstream in specific categories |
| Conversion vs. affiliates | 86% worse | The infrastructure gap, not the demand gap, is the current bottleneck |
| Conversion potential vs. actual | 4.4x higher potential | The size of the opportunity sitting behind the infrastructure fix |
That 4.4x figure is the one worth sitting with the longest, because it reframes the whole story from a warning into an opportunity with a number attached. The report's own framing treats the performance gap, potential conversion versus actual conversion, as the size of the infrastructure opportunity currently sitting unclaimed on the table. By 2030, the same research projects nearly half of online shoppers will use AI agents, accounting for roughly a quarter of their spending, adding an estimated $115 billion to US ecommerce. Merchants that close the infrastructure gap early are positioning for a multi-year runway, not a one-quarter trend.
Why traffic and conversion split this hard
The mechanical reason behind the gap is straightforward once you separate discovery from transaction as two genuinely distinct problems, which is exactly what most merchant stacks currently fail to do. An AI agent can discover your product easily, because product discovery mostly runs through the same retrieval and citation mechanics we've covered extensively on the GEO side: structured, extractable content that a retrieval system can parse gets surfaced. Discovery was never the hard part. Transaction is, because completing a purchase requires an agent to read real-time inventory, resolve pricing and variant selection, and execute payment through an interface built for a human clicking buttons, not a machine calling an API.
This is the same pattern we've documented in Kevin Indig's 100-product agent study, where AI agents answered pricing questions correctly only 79% of the time against 92-93% for less volatile fields like integrations or security. Pricing and checkout are exactly the parts of a commerce flow most likely to be dynamic, personalized, or gated behind interactive elements an agent can't reliably parse, which means the failure concentrates precisely where a transaction actually needs to happen.
The protocol layer merchants actually need
The infrastructure fix isn't abstract; it already has a name, or rather four separate names right now, because the market hasn't converged on one standard yet. Google and Shopify co-developed the Universal Commerce Protocol, launched January 11, 2026 at NRF with backing from Etsy, Wayfair, Target, and Walmart. OpenAI and Stripe built the Agentic Commerce Protocol around ChatGPT's shopping features. Google separately backs the Agent Payments Protocol for payment authorization specifically. And Anthropic's Model Context Protocol provides the data-connectivity layer, giving agents access to real-time inventory, pricing, and product detail rather than a static, possibly stale product feed.
None of these four competing protocols has won outright yet, and merchants shouldn't wait around for consolidation before acting, because the report's underlying point is that the gap exists right now, independent of which protocol eventually dominates. A merchant whose product data is clean, current, and structured to be machine-readable is positioned to plug into whichever protocol or protocols end up mattering for their specific customer base. A merchant whose product data still assumes a human is reading a rendered page has work to do regardless of which standard wins.
It's worth naming why merchants ended up here, because none of this infrastructure gap was a strategic decision. Ecommerce platforms spent the last decade optimizing checkout for a human on a screen: fewer clicks, better mobile responsiveness, saved payment methods, one-click reorders. Every one of those investments made sense for the traffic that existed at the time and made the checkout flow more, not less, dependent on a human navigating a visual interface. An agent showing up to that same flow isn't encountering neutral ground. It's encountering a decade of UX decisions that actively assumed it would never be the one making the request.
That history matters because it explains why this gap won't close on its own, the way some infrastructure gaps eventually do simply through normal platform updates. Shopify, WooCommerce, and the other major platforms are actively building agent-facing capability now, which is genuinely good news, but a merchant running a heavily customized checkout, the kind most mid-market and enterprise ecommerce operations have, inherits none of that platform-level progress automatically. Custom checkout logic, third-party payment integrations, and bespoke variant selectors all need their own, separate audit against agent-readability, regardless of what the underlying platform ships.
What 'agent-ready' means for a product page
“The merchants winning this transition aren't the ones with the flashiest product pages. They're the ones whose backend already answers the questions an agent needs answered to complete a purchase without a human in the loop.”
Building the fix in the right order
The temptation is to treat protocol adoption, picking UCP or ACP and integrating it, as the first move. It shouldn't be. Per the framework we laid out in the enterprise AI agent readiness framework, task completion for an agent depends on machine-readability under load and fallback exposure before it depends on which specific protocol sits on top. A merchant that bolts a transaction protocol onto product data that's still inconsistent, stale, or rendered client-side has built a fast path to a broken transaction, not a working one.
This ordering matters because the 86%-worse conversion figure almost certainly isn't evenly distributed across the funnel. Discovery-stage agent traffic is probably performing close to expectations, since that's the layer most GEO work already targets. The collapse is concentrated at the transaction layer specifically, which means the fix is concentrated there too. Spending another quarter optimizing discovery content while checkout remains agent-illegible is optimizing the part of the funnel that was never actually broken.
There's a competitive-timing argument for moving on this now rather than waiting for the standards fight to settle, and it's worth being explicit about it. Early technical debt in a new channel compounds the same way it does everywhere else: the merchants who wire up clean, machine-readable product and inventory data now are the ones who can plug into whichever protocol wins with a small integration project. The merchants who wait for a winner to emerge, then discover their underlying product data still isn't agent-legible, are looking at the same foundational audit this piece describes, just starting later, with more competitors already through it, and less of the current 4.4x conversion-potential gap left to capture for themselves.
Do this next: pull whatever agent-traffic data you currently have, even if it's just user-agent logs, and separate it into discovery-stage visits versus transaction-attempt visits. If you can't make that split today, that's the actual first gap to close, because you cannot fix a conversion problem you can't yet locate inside your own funnel. Our generative engine optimization team is building exactly this kind of agent-transaction instrumentation into ecommerce engagements now, ahead of the point where every competitor in a given category has already closed the same gap and the advantage of moving early on it has quietly disappeared for good.
See where you are cited today
A free snapshot audit of your rankings and AI citations before we ever talk.
Tyler leads work at the intersection of SEO and generative engines at Something Inc., helping B2B brands get ranked and cited across every major AI engine.