Agentic Commerce by the Numbers: What Mid-2026 Data Shows

Agentic commerce statistics for 2026 fall into four evidence classes that most coverage prints side by side as if they were one thing: platform telemetry, vendor-reported program data, consumer surveys, and analyst forecasts. Shopify reports that AI-referred traffic to its merchants grew roughly 8× year over year in Q1 2026, with AI-referred orders up nearly 13×. A widely reported eMarketer forecast puts AI-platform retail ecommerce at about $20.6 billion in the US this year. Both statements can be true. They are not measuring the same thing.

That distinction is not pedantry — it is the difference between a budget that survives contact with reality and one that does not. The forecasts circulating in mid-2026 range from roughly $20.6 billion for one year in one country to $5 trillion globally by 2030. Divide those and you get a spread of about 243×. Almost none of that gap is disagreement about how fast AI shopping will grow. Almost all of it is disagreement about what counts as an agentic transaction.

This guide sets out the figures we could actually stand behind, in the order of how firmly they are sourced: Shopify and Adobe platform telemetry first, Salesforce’s vendor-reported peak-season data second, the analyst forecast landscape third, and the consumer trust data that acts as the honest counterweight to all of it. Where a primary source was paywalled or unreachable at the time of writing, the post says so rather than dressing a secondary citation up as a direct one.

Most agentic-commerce statistics you will encounter are either projections or stated intent. Shopify’s Q1 2026 figures are neither. They come from a platform observing sessions and orders across its merchant base, published in Shopify’s own enterprise blog post on May 11, 2026. The same pair of figures also appears in an earlier Shopify executive guide dated April 23, 2026 — the same publisher restating its own numbers, which is consistency rather than independent corroboration.

Secondary coverage ties the underlying commerce results to Shopify’s Q1 2026 Form 8-K filed with the SEC in early May 2026. We did not read the filing itself, so treat the SEC framing as corroboration of the reporting period rather than as our source for the numbers.

The quality figures matter as much as the volume ones. Shopify reports AI-referred sessions converting nearly 50% better than organic search on product detail pages, AI-referred orders carrying a 14% higher average order value, and AI-referred conversion beating organic SEO in 23 of the 25 merchant categories tracked — averaging 56% higher conversion within those categories. That last number is worth holding onto: it is a within-category average, not a blended headline, which makes it harder to explain away as a mix effect.

One arithmetic note we can add ourselves. Nearly 13× order growth against 8× traffic growth is a ratio of roughly 1.6. Both multiples are measured on the same base population over the same window, so the implication is direct: the conversion rate on AI-referred sessions improved year over year, rather than the channel simply delivering more of the same-quality visits. That is our computation from Shopify’s two published multiples, not a figure Shopify printed.

The honest caveat: this is one platform’s merchant base, using one platform’s definition of an AI referral. Shopify skews toward direct-to-consumer brands, and referral classification depends on how assistant traffic identifies itself. Strong evidence, bounded scope. If you sell on Shopify, our guide to making Shopify catalogs visible to AI chats covers the mechanics behind these referrals.

One platform’s telemetry is a data point. Two independent platforms showing the same direction in the same quarter is evidence. Adobe Digital Insights, drawing on its own retail analytics footprint rather than Shopify’s, reports AI-referred traffic to US retail sites growing 393% year over year across Q1 2026, with March 2026 alone up 269% year over year. Adobe published the analysis without a visible publication date at the time of writing; it references the company’s 2026 Q2 AI Traffic Report and carries a byline from Vivek Pandya, Director of Adobe Digital Insights.

The more interesting Adobe finding is the conversion reversal. In March 2025, AI-referred traffic converted 38% worse than non-AI traffic. In March 2026, it converted 42% better. That is an 80-point swing in the relative gap inside twelve months, and it is the single clearest signal that assistant traffic stopped being curiosity browsing and started being purchase intent.

Adobe separately reports AI-referred traffic during the 2025 holiday season (November and December) up 693% year over year. Keep that figure labelled by its window: it is a holiday-season measurement, not the Q1 2026 one, and averaging the two into a single “AI traffic is up 400 to 700%” claim would blur two different reporting cadences.

Read the chart as one claim, not seven. Across two independent datasets and several different metrics, AI-referred visitors engage longer, view more, convert better, and spend more per order than the baseline they are compared against. The magnitudes differ because the comparisons differ — Shopify benchmarks against organic search, Adobe against all non-AI traffic — but no metric in either dataset points the other way.

Salesforce’s Cyber Week 2025 release, published December 5, 2025, is the most-quoted agentic-commerce figure in circulation: AI and agents drove $67 billion in sales during Cyber Week (November 25 to December 1, 2025), influencing 20% of all global orders, across a dataset Salesforce describes as covering more than 1.5 billion shoppers in 89 countries. Global Cyber Week sales came in at $336.6 billion, up 7% year over year, with US sales at $79.6 billion, up 5%.

Two labels belong on that $67 billion, and most coverage applies neither. It is vendor-defined: “AI influenced” here includes AI product recommendations and conversational service, not only autonomous agent checkout. And it is vendor-reported: the company measuring the impact of AI commerce is also the company selling AI commerce software, and the figure has not been independently audited. Neither label makes the number wrong. Both change what you can compare it to.

A quick consistency check we can run ourselves: $67 billion against $336.6 billion in global Cyber Week spend is 19.9%, which lines up with the 20%-of-global-orders claim in the same release. The two figures are internally coherent, which is reassuring about the arithmetic even though it says nothing about the definition.

The operational figures in the same release are less headline-grabby and arguably more useful: agentic customer-service conversations on Salesforce grew 55% week over week during Cyber Week 2025, and agent-initiated actions such as returns and address changes surged 70% against the prior week. Those describe post-purchase service load, which is where most merchants will feel agentic commerce before they feel it in checkout.

Here is the table the rest of the internet does not print. Every major agentic-commerce forecast in circulation gets quoted as though the numbers were rival predictions about the same quantity. They are not. Once you add a scope column, the apparent disagreement mostly evaporates — and what is left is a sourcing problem, not a forecasting one.

The group headers below are the sourcing tier, and they are the most important column on the page. Only the McKinsey rows were read directly from the publisher’s own page. Everything else reached us through secondary coverage, and we have said so on each row rather than hedging once in a footnote.

The multiples in the last column are our own arithmetic against the $20.6 billion figure, and they exist to make one point visible: the top of the range sits about 243× above the base. They are not like-for-like comparisons. A 2026 US autonomous-checkout number and a 2030 global orchestration number differ on year, geography and definition simultaneously, so the ratio measures scope drift, not growth. Read it as a warning label.

Two smaller reconciliations fall out of the same table. The eMarketer figure implies a total US ecommerce market of roughly $1.37 trillion in 2026, since $20.6 billion at about 1.5% back-solves to that denominator — a sanity check that the share and the dollar figure are consistent with each other. And McKinsey’s US band sits between 18% and 33% of its own global band depending on which ends you pair, which is a plausible US share of global consumer ecommerce and suggests the two ranges were modelled together rather than bolted on.

Morgan Stanley’s narrow definition is the useful bridge. At $190 billion to $385 billion for US agent-driven ecommerce by 2030, it sits between 2.3× and 5.3× below McKinsey’s US band for the same year and country — which is roughly the size of the gap between “the agent executed the purchase” and “the agent shaped the purchase.” That gap is the whole argument, quantified.

Growth charts are only half the picture, and the missing half is the one merchants actually have to plan around. A Checkout.com study on agentic commerce, published in June 2026 and reaching us through a research aggregator rather than the primary, reports that 65% of US consumers would trust AI to compare prices while only 14% would trust it to place an order autonomously. The aggregator attributes the underlying data to a Checkout.com agentic-commerce tracker run with YouGov in May 2026.

That is a 51-point gap, or about 4.6× on a ratio basis. It is also the cleanest available explanation for why the narrow forecasts and the broad forecasts diverge so hard: the broad ones assume the gap closes, the narrow ones assume it mostly does not.

The 3%-versus-89% pairing at the bottom of that chart is the one to put in front of a board — with its sourcing label attached, because both figures reach us through the same aggregator citing Checkout.com rather than from the primary. Roughly 3% of transactions currently involve an AI agent while about 89% of merchants say they are actively preparing for agentic commerce. Preparation is running far ahead of consumer behaviour, which is a defensible position — the catalog and payments work takes quarters, and merchants who start when demand arrives will be late — but it is not the same thing as demand.

The survey numbers are also not mutually consistent, and that is informative rather than disqualifying. Adobe’s own consumer survey of more than 5,000 US respondents puts AI shopping usage at 39%, of whom 85% say it improved their experience and 66% believe AI tools return accurate results. IBM’s Institute for Business Value, working with the National Retail Federation, surveyed more than 18,000 consumers across 23 countries in Q3 2025 and found 45% using AI for help during their buying journeys, with 72% still shopping in physical stores — both IBM figures reaching us through a research aggregator rather than from the IBM primary, which we did not read. Different populations, different windows, different question wording. Do not average them.

Our own reading of the spread: usage of AI as a research tool is now mainstream and rising, delegation of the purchase decision is not, and the two will decouple further before they converge. Merchants who plan for research-stage agents in 2026 and 2027 and treat autonomous checkout as an option to be ready for rather than a volume assumption will be closer to right than either extreme. The discover-in-AI, buy-on-site pattern is what that looks like operationally.

Put two of the figures above next to each other and a specific, actionable problem appears. Shopify reports that more than half of AI-referred sessions — around 55% — start on a product detail page, against roughly 20% for organic search. Adobe, scoring US retail sites for machine readability, reports average scores of 75% for homepages, 74% for category pages, and 66% for individual product pages.

So agents arrive on product pages at nearly three times the rate organic search does, and product pages are the surface scoring lowest for machine readability by a 9-point margin against homepages. Roughly a third of product-page content is not machine-readable by that measure. The traffic is compressing onto exactly the pages that are least prepared for it.

Shopify’s own framing of why this matters is blunt. Its field CTO team writes that AI does not browse — it reasons across structured data, matches attributes to intent, and skips anything it cannot parse. The same guide adds that unlike a low-ranking Google result, there is no page two to scroll to; the agent simply moves to the brand that is easiest to understand. Both lines come from Shopify’s agentic commerce executive guide of April 23, 2026, and we are paraphrasing them rather than presenting them as pull quotes.

That sequencing — catalog first, protocol second — is the practical translation of every number in this post. If you want the step-by-step version, our readiness checklist for merchants walks the audit, and the UCP, ACP and AP2 protocol landscape covers the checkout layer once the catalog is clean. The same structured-data discipline drives visibility in AI answers outside of commerce surfaces.

Every statistic in this post belongs to one of four evidence classes, and the class determines how much weight it can carry. Most agentic-commerce coverage mixes all four in a single bulleted list, which is how a modelled 2030 projection ends up sitting beside a counted Q1 result as though they were peers.

Three questions applied to any agentic-commerce statistic will keep you out of trouble. What is being counted — autonomous checkout, agent-influenced revenue, referred sessions, or stated intent? Over what window and where — a quarter, a peak week, a full year, a 2030 projection; one country or worldwide? How firmly is it sourced — read from the publisher, widely reported behind a paywall, or reaching you through one aggregator?

Our forward read, stated plainly so it can be judged later: the telemetry class will keep compounding through 2027 because assistant-mediated research is becoming a default consumer behaviour, while the autonomous-checkout share stays small enough that the narrow forecasts look conservative only in hindsight rather than immediately. The interesting inflection is not consumer checkout at all — it is procurement, where the buyer is already a system and the trust question is a policy configuration rather than a psychological one. If the B2B prediction in the table above proves even half right, the volume story of agentic commerce will be a business-to-business story long before it is a consumer one.

Numbers earn their keep when they change a decision. Here is how the four evidence classes map onto four common situations, with the figure that should drive each one.

One more comparison worth making explicit for anyone benchmarking platforms rather than merchants: the telemetry above comes from Shopify and Adobe, but the surface where agents transact is determined by which standards each storefront and each assistant supports. Our platform-by-platform comparison covers where each stack currently sits, and our ecommerce team runs the catalog and readiness work described above.