I've spent most of my career on the retail side of search, at a scale where a rounding error is a real number. So when a channel shows up claiming several hundred percent growth, my first instinct is to ask what the denominator was.
Here's the honest picture on agentic commerce in 2026, including the parts that undercut the hype and the part that should genuinely worry you.
The growth is real
AI referral traffic to US retail sites grew 393% year over year in Q1 2026, with March alone up 269%. Adobe reported 693% YoY growth across the 2025 holiday season, and AI-referred traffic converted 31% better than non-AI sources. Adobe's dataset covers over a trillion visits, which makes it the most methodologically solid thing available.
Triple Whale tracked 606,489 AI citations across ecommerce queries between January and March 2026. Traditional channels produced 164 million referral transactions over the same window. That is the ratio nobody puts in the deck.
AI is the fastest-growing channel in retail and it is not yet a primary channel. Those are not in tension. Treating it as either "the future, reallocate now" or "noise, ignore it" both misread the same data.
The attribution problem is the actual story
This is the part I'd escalate if I were still sitting in a retail analytics review.
Roughly 70.6% of AI referrals are invisible in standard GA4 setups, misclassified as direct traffic. The mechanism is boring and completely deterministic: paid ChatGPT accounts don't pass referrer data. Gemini in Deep Research mode doesn't either. No referrer means the session lands in direct, next to bookmarks and typed URLs.
The practical consequence: if you're evaluating whether agentic commerce matters to your business by looking at your AI referral line in GA4, you're seeing roughly three out of every ten sessions and concluding the channel is a third of its actual size.
Watch your direct channel instead. Direct traffic rising while brand search stays flat is the signature. It's not proof, but it's the cheapest diagnostic available and almost nobody runs it.
Checkout is lagging discovery, badly
OpenAI launched Instant Checkout on 29 September 2025 via the Agentic Commerce Protocol, co-developed with Stripe — Etsy sellers first, then Shopify merchants, 4% merchant fee. Google answered at NRF on 11 January 2026 with UCP, built with Shopify and endorsed by 20+ retailers including Home Depot, Lowe's, Best Buy, Macy's, Visa and Mastercard.
So we have two competing standards and a protocol war, which is where this always goes.
But the buying behavior hasn't followed the infrastructure. Walmart reported 3x lower conversion for in-chat purchases versus redirecting users to its own site, and responded by integrating its own assistant, Sparky, into ChatGPT rather than leaning further into in-chat checkout.
Discovery moved to the new surface years before checkout did. That's not a failure of agentic commerce — it's the exact shape mobile commerce took, and mobile turned out fine.
The more interesting Walmart number: ChatGPT traffic drove roughly 2x as many new customers as traditional search channels. That's the argument for caring about this now. Not revenue today — acquisition of people who don't already know you.
The conflicting conversion research
Worth being straight about this, because it gets cited selectively in both directions.
| Source | Finding | What it likely measures |
|---|---|---|
| Adobe Digital Insights Jan 2026, 1T+ visits |
AI-referred traffic converted 31% better than non-AI sources | Holiday retail, broad channel comparison. Largest dataset available. |
| Kaiser & Schulze Peer-reviewed, 973 retailers |
ChatGPT referrals beat paid social, fall below direct, organic and email | Rank-ordering against all channels, not just an AI/non-AI split. |
| Walmart | 3x lower conversion for in-chat vs. redirect to site | In-chat checkout specifically, not AI referral generally. |
These reconcile cleanly once you notice they're measuring different things. AI referrals are high intent relative to discovery channels and lower intent than channels reached by people who already know you. And in-chat checkout is a separate question from AI-driven discovery — one is working, the other isn't yet.
Anyone quoting one of these three without the other two is building an argument, not describing a market.
What I'd actually do about it
Fix attribution before you fix anything else. Build a custom channel grouping that captures chatgpt.com, perplexity.ai, claude.ai, gemini.google.com and copilot.microsoft.com. Then track your direct channel against brand search as a proxy for the invisible 70%. You cannot make a resourcing decision on a number you know is wrong by a factor of three.
Treat your product data as acquisition infrastructure. If a buyer's first interaction is a conversation rather than a category page, then catalogue structure, specification completeness, and documentation quality are doing the work your PDP used to do. That's a merchandising and data-hygiene project, not a content project.
Don't reallocate budget yet. At sub-0.2% of sessions, moving real money out of channels that work is not defensible. Build the measurement, fix the data, watch the trend. The cost of being six months late here is low. The cost of being wrong about a reallocation is not.
One live constraint worth tracking: in March 2026 a US federal judge in San Francisco issued a preliminary injunction blocking Perplexity's Comet browser from making purchases on Amazon. Amazon argued the agent misrepresented automated traffic as human. However that resolves, it establishes that whether agents are permitted to transact on a given platform is a legal question, not just a technical one.
Want your AI citation baseline?
Free 5-prompt snapshot through Sable Search — your brand across ChatGPT, Claude, Perplexity and Google AI Overviews versus your top competitor. One page, 48 hours.
Get the free snapshot ↗Frequently asked questions
AI referral traffic to US retail sites grew 393 percent year over year in Q1 2026, with March alone up 269 percent. Adobe reported 693 percent growth during the 2025 holiday season. The growth is real but comes off a very small base — AI-driven sessions still sit below roughly 0.2 percent of total ecommerce traffic.
Roughly 70 percent of AI referrals are misclassified in standard GA4 configurations. Paid ChatGPT accounts do not pass referrer data, and Gemini in Deep Research mode does not either. When the referrer is stripped, the session lands in the direct bucket alongside bookmarks and typed URLs.
Adobe data covering over a trillion visits found AI-referred traffic converted 31 percent better than non-AI sources during the 2025 holiday season. Peer-reviewed work from Kaiser and Schulze found ChatGPT referrals convert better than paid social but below direct, organic search and email. Both can be true — AI referrals are high intent relative to discovery channels and lower intent than channels reached by existing customers.
ACP is an open standard for AI-led purchasing co-developed by Stripe and OpenAI and released in September 2025. It powers ChatGPT Instant Checkout, which launched 29 September 2025 with Etsy and Shopify merchants at a 4 percent merchant fee. Google announced a competing standard, UCP, at NRF on 11 January 2026 with Shopify and more than twenty retailers.
Sources
- Adobe Analytics / Adobe Digital Insights. AI traffic to US retail, Q1 2026 and January 2026 holiday report.
- Elogic Commerce. ChatGPT commerce and agentic shopping statistics, 2026 — GA4 misclassification analysis.
- Triple Whale. AI citation tracking across ecommerce queries, January–March 2026.
- Kaiser, M. & Schulze, C. The economics of LLM referral traffic in ecommerce: evidence from 973 retailers. SSRN, 2025.
- OpenAI / Stripe. Agentic Commerce Protocol, September 2025.
- Commercetools. Agentic commerce enterprise guide, 2026 — Walmart conversion and new-customer figures.