Semafor ran a piece this month that should be taped to every Chief Digital Officer’s monitor. This coming holiday season, brands will have a new type of shopper to court: AI agents. The number underneath it is the part that matters. Cheq, a cybersecurity company that verifies AI shopping agents, recorded a 100x surge in agentic traffic over a 45-day period following the release of consumer AI agents like Instinct and Meta’s Muse.
A hundred times. In six weeks. Right before peak.
Here is the uncomfortable follow-on. Agents still make up a small fraction of online shoppers, but they are growing quickly, especially amid a constant drumbeat of consumer agent releases. Small fraction, exponential slope, arriving in the one window where you have zero slack to investigate anomalies. That is the setup for Holiday 2026.
The traffic you cannot see is still traffic
The mechanic is simple and it breaks your dashboards. An agent does the research, the comparison, and in a growing number of cases the purchase, inside a conversation. AI assistants embedded in ChatGPT, Gemini, and Perplexity increasingly function as personal shoppers that find products, compare prices, and complete purchases without customers ever visiting a retailer’s website.
Your analytics stack assumes a human with a browser, a session, and a scroll depth. An agent satisfies none of those assumptions. It may hit your product feed, call a checkout API, and complete an order without ever rendering the page your funnel metrics are built around.
The volume is real and it is already here. AI-driven visits to US retail sites grew 693% year over year during the 2025 holiday season, with 769% growth in November and 673% in December, per Adobe Analytics. Off a small base, yes. AI-driven sessions still sit below 0.2% of total ecommerce traffic, but they are growing faster than any other channel. Fast growth off a small base is exactly the profile that goes from rounding error to board slide in one season.
And the forward-looking piece is where it gets strategic. Salesforce expects AI chat agents to originate about 20% of holiday ecommerce traffic. If a fifth of your traffic routes through a channel you cannot label, your January numbers are fiction.
The Cost of Doing Nothing is a blind January
I run this calculation with clients constantly. The CODN here is not a missed sale during Cyber Week. Agent conversion is still catching up to legacy channels, so the near-term revenue leak is manageable. The real cost lands in late January, in the postmortem, when someone asks a question you cannot answer: how much of our holiday revenue was agent-originated, and did those orders behave differently?
If you did not instrument before Black Friday, that meeting is a shrug. You will debate a channel you never measured, using proxies and vibes. Meanwhile the operators who labeled their agent traffic will walk into 2027 planning with a hard number and a behavioral profile.
That profile is worth capturing, because agent-referred demand is not like your other traffic. Customers arriving via AI agents are 10% more engaged than traditional visitors, reaching retailers further down the funnel with stronger intent to purchase. You want to know which of your holiday orders carried that signature. You cannot know it retroactively.
Instrument these four things before Black Friday
This is the forward-deployed checklist. None of it requires a platform migration. All of it requires a decision this week.
First, classify agent sessions at the edge. Build a labeling layer on user agent strings, verification signals from services like Cheq, and traffic patterns that no human produces. Tag the session the moment it arrives.
Second, stamp agent-originated orders in the order record itself. Finance needs to separate agent revenue from human revenue in the general ledger, not reconstruct it from logs in February. One boolean field, decided now, saves a quarter of arguing later.
Third, promote your structured feed and checkout API hits to first-class analytics events. In agent-mediated commerce the behavioral data stream starts at the add-to-cart moment; the discovery, the browse, and the refined preferences all live inside the assistant. You will never recover the upstream behavior, so capture everything downstream of the handoff with precision.
Fourth, write the taxonomy down and freeze it before peak. Define what counts as an agent, what counts as agent-influenced versus agent-completed, and publish it to every team that touches a report. A number only means something in January if it meant the same thing in November.
Some of your peers will punt. Amazon and eBay have blocked agents to reinforce control over their ads businesses and checkout experiences. That is a strategy available to platforms with that kind of leverage. For everyone else, the agents are coming through the door whether you measure them or not.
Holiday 2026 is the first peak with agents in the checkout path. The retailers who instrument this month get a map. The rest get a story they tell themselves about a channel they never saw.