Reading the 2025 Holiday AI Numbers Without Getting Carried Away

It is the morning after Thanksgiving, and two big research firms have data they want you to see.

Salesforce tracked 1.5 billion shoppers through Cyber Week and found that AI and agents influenced 20% of all global online orders: $67 billion in sales driven by personalised recommendations and conversational customer service. Adobe Analytics, which covers more than a trillion visits to US retail sites, reports that traffic from generative AI tools rose 693% year on year during the 2025 holiday season. On Thanksgiving itself, AI-referred visitors converted at 54% higher rates than everyone else. On Black Friday, 38% higher.

These are real numbers from credible primary sources. They are also numbers that reward some care before you start rewriting your Q1 strategy around them.

What "AI influenced" actually means

The Salesforce methodology counts any shopping journey where AI contributed a touchpoint. That includes someone who asked ChatGPT for gift ideas before opening a browser tab. It includes the personalisation engine that served a relevant recommendation at 11pm. It includes the AI chatbot that answered a delivery question at checkout. All of those count as AI influence, and that is probably the right way to measure it.

But it means the $67 billion figure is not "$67 billion worth of purchases completed by AI agents acting autonomously." The agentic commerce piece (a shopper instructs an AI, the AI researches, selects, and completes the transaction without further input) is real and growing, but it is still a small fraction of that total. Salesforce's own data shows the broader AI influence figure dwarfs the narrower autonomous-agent figure.

This distinction matters because the two cases have different implications for retailers. A high "AI influenced" share means your recommendations engine, your chatbots, and your content quality all matter. A high "autonomous agent completion" share would mean something more structural about how transactions are authorised and executed. We are in the first situation, not the second. Not yet.

The conversion data is the interesting part

The Adobe numbers on conversion quality are the finding I find most substantive. AI-referred visitors converting 54% better on Thanksgiving is a meaningful signal, not rounding error.

The most plausible explanation is selection effects. Someone using an AI assistant to research gift options before Thanksgiving has already done the consideration work. They arrive at a retailer already intent on a purchase, already knowing roughly what they want. They are not browsing; they are buying. Adobe's data supports this: AI-referred visitors are 33% less likely to leave immediately, reflecting stronger intent on arrival.

This is consistent with what Adobe has tracked through 2025 more broadly: AI referral traffic shows lower bounce rates, higher time on site, and higher conversion than other organic sources.

What it probably does not mean is that AI is doing the selling better than other channels. It may mean that the people currently using AI tools for shopping discovery are disproportionately high-intent to begin with, and that as adoption widens, the average AI-referred visitor will look more like the average online shopper. The premium will compress as the population normalises.

A note on the 693% traffic figure

A 693% year-on-year increase sounds extraordinary. It is worth remembering that the 2024 baseline was at early-adopter levels. Large percentage increases from small bases are not the same as large absolute numbers.

Adobe's November figure was even higher: AI-driven traffic to retail sites up 769% year on year. December: 673%. The percentage growth rate is already beginning to moderate as the base grows.

By 2026 and 2027, as AI assistants become a routine part of how people approach shopping, the useful numbers will shift from year-on-year percentage changes to absolute traffic share. The growth rate tells you things are moving fast. The absolute share will tell you how important the channel actually is.

What this means in practice

The practical observation is that AI-referred traffic is now large enough to warrant its own lane in your analytics.

If you are not already segmenting AI-referral traffic separately from organic search, paid, social, and direct, you are missing a signal that your conversion data contains. If AI-referred visitors are converting at a materially higher rate, you want to know which AI sources are driving that traffic, whether your site experience is optimised for visitors who arrive already knowing what they want, and whether your content is structured in ways that AI assistants can actually read and cite.

That last point is where a lot of retailers are not ready. Adobe's companion research found that significant portions of US retail sites are not fully readable by AI crawlers. The parallel in the UK market is no different: a site that cannot be parsed cleanly by an AI assistant is a site that does not appear in AI-driven discovery at all, regardless of how well it performs on traditional search. That is a visibility problem, not an AI problem. It is fixed with structured data, clear taxonomy, and machine-readable product information — the same GEO and content work that has been building through 2024 and 2025.

The agentic question, honestly

One more thing worth naming. The story the vendors most want to tell about this data is the agentic one: AI is about to complete purchases autonomously on your behalf, and retailers who are not ready will lose the transaction entirely. That story is directionally true but temporally premature.

Salesforce's data is unambiguous that retailers using AI agents saw meaningfully faster sales growth this season — 59% higher growth rate than those that did not. The direction is right. But the population of consumers authorising AI to complete purchases end-to-end is still small. Most of what the data captures is AI as discovery and service tool, not AI as purchasing agent.

That will change. Probably faster than previous platform shifts. But the near-term question is not "is your checkout agent-ready" so much as "are your products findable and well-described in the AI interfaces consumers are already using." Get that right first. The agent-readiness question comes after.

The holiday data is evidence that AI commerce is genuinely arriving. It is also early enough that the shape of what it becomes is not fixed.

That is actually useful information, if you let it be.

---

Data sourced from Salesforce Cyber Week 2025 press release and Adobe holiday season 2025 report. Adobe figures cover US retail only; Salesforce figures are global.