Agentic commerce in 2026: what the numbers actually say
AI referred traffic to US retail grew 393 percent in a year and now converts better than every other channel. The pages doing the selling are the least readable ones on the site.
For two years the honest answer to whether AI shopping mattered was that nobody could tell yet. The traffic was growing off a tiny base and it converted badly, which is what a novelty looks like. That answer expired sometime in the first quarter of 2026.
393% growth in AI referred traffic to US retail sites, Q1 2026 against Q1 2025. 42% better conversion from AI referred traffic than non AI traffic, March 2026. 37% higher revenue per visit from AI referrals than every other channel.
Those figures come from Adobe Analytics, which tracks more than a trillion visits to US retail sites, so they are measurement rather than forecast. The growth number is the one that gets quoted. The conversion number is the one that should change what you do.
The number that flipped
In March 2025, traffic arriving from an AI assistant converted 38 percent worse than traffic from paid search, email and organic. Twelve months later the same channel converted 42 percent better. That is an eighty point swing in a year, and it is the single most important fact in this market.
The reason is not that the models got smarter about shopping. It is that the question changed. A shopper who types a product name into a search box is often browsing. A shopper who describes a constraint to an assistant, budget, delivery date, compatibility, has already done the thinking and is close to buying. The assistant hands you a filtered, late stage buyer. Adobe measures the rest of that behaviour too: those visitors spend 48 percent longer on the page, view 13 percent more pages, and show a 12 percent higher engagement rate than everyone else.
A channel growing at 393 percent that also converts 42 percent better is not an experiment to watch. It is a channel your competitors are already being ranked in, whether or not anyone at your company has looked at it.
The gap is on your side of the wire
Adobe also published a visibility scorecard across retailer sites in the same report, and this is where it gets uncomfortable. The page types that score best on machine readability are the ones with the least commercial value. Returns and exchanges, contact us and FAQ pages all score above 80 percent. Homepages average around 75 percent.
Product detail pages score 66 percent, the worst of any category. The pages that carry your prices, your stock and your reason to be chosen are the least readable pages on your site. Your returns policy is beautifully structured and your product catalogue is a rendering problem.
There is a mundane engineering explanation for this. Returns and FAQ pages are static text rendered on the server. Product pages are the most dynamic thing a retailer owns: price experiments, personalised offers, live inventory, all of it fetched by JavaScript after the page loads. A human never notices. An agent typically makes one HTTP request with no browser attached and gets a page with no price on it.
What the plumbing looks like now
Two protocol efforts moved from announcement to production in the last year, and they are not the same thing.
ACP, the Agentic Commerce Protocol from OpenAI and Stripe, has been live inside ChatGPT since September 2025. Because it was built into Stripe checkout infrastructure, it arrived with scale attached. More than a million merchants are reachable through it, including Etsy sellers and Shopify brands such as Glossier, Vuori and Spanx.
UCP, the Universal Commerce Protocol, was announced by Google in January 2026 with Shopify, Target, Etsy, Walmart and Wayfair behind it. In March 2026 Google added onboarding through Merchant Center along with cart support and catalogue access. Klarna has since adopted it.
The strategic point is not which one wins. It is that both of them turn a merchant into something an agent can transact with directly, instead of something an agent has to read and then send a human to. Once a category has enough participants on that footing, being the store that still requires a person to click through checkout is a ranking penalty rather than a neutral choice.
What a serious merchant should take from this
Treat AI referral as a measured channel, not a curiosity. If your analytics does not separate it, you are flying blind on the fastest growing and best converting source you have.
Fix product detail pages first. Server render price, currency, availability and an identifier, and publish them as Product structured data. This is the cheapest work with the highest leverage, and it helps ordinary search too.
Find out whether your platform already published a commerce manifest for you. If it did, you have parity with every competitor on the same platform and no advantage over any of them.
Publish the boring trust signals. Return policy, delivery windows and business identity as structured data. When several stores can all be bought from, these are the tiebreakers.
None of this requires a bet on how large agentic commerce gets by 2030, where credible forecasts range from 190 billion dollars to five trillion depending entirely on how the term is defined. It only requires noticing that the traffic is already here, it already converts better than anything else you have, and it is currently landing on the least readable pages you own.
Sources
AISmith Team
Author