You’ve probably seen some headline numbers by now declaring that AI shopping has finally arrived. Adobe Analytics reported that AI-driven traffic to U.S. retail sites surged 805% year-over-year on Black Friday.

Salesforce, tracking 1.5 billion global shoppers, announced that generative AI and agents drove $14.2 billion in online sales on Black Friday alone. Sensor Tower found that Amazon shopping sessions involving the Rufus chatbot that resulted in purchases jumped 100% compared to the prior 30 days.

The narrative is that AI shopping has finally arrived. The robots are here, and they're buying stuff. But Ian Simpson, SVP of Innovation & Strategy at Sensor Tower, had a word of caution about reading too much into these figures, which I’ll share later.

I'm writing this Monday afternoon watching Cyber Monday numbers roll in, still far too early for conclusions. But let’s look at some of the data that’s flowed in so far and where it may be pointing.

Did you know leading retailer media networks drive 85% of their ads revenue through mid- and long-tail advertisers?

Mirakl Ads provides full funnel ad  formats tailored to  both 1P and 3P advertisers, leveraging unique AI-capabilities that provide unprecedented levels of relevance and engagement. 

Retailers who want to capture ad spend from the long tail of 3P marketplace sellers use Mirakl Ads in their tech stack.

Adobe's number—805% growth in AI referrals to retail sites—sounds explosive. Salesforce says AI influenced $14.2 billion in Black Friday sales globally. Sensor Tower reports that Rufus-involved Amazon sessions that resulted in purchases jumped 100% versus the prior 30 days.

These numbers look huge until Sensor Tower’s Ian Simpson asks the obvious question: Didn't ChatGPT usage roughly quadruple this year?

If you're measuring AI referral traffic and the total addressable market of AI users grew 4x, of course referrals surged. That's not a shift in the daily shopping behavior of everyday customers—it's basic math about a larger base of users.

Simpson unpacked three other problems with the narrative in a post on LinkedIn:

So people are using AI tools, but are they shopping differently, or just using better search?

Consumers got maybe 48-72 hours to test these new features before the biggest shopping weekend of the year.

Scot Wingo has been tracking agentic commerce announcements on his Retailgentic blog. Look at the timing:

This is AI shopping's "first real test at scale"—the moment when supply and demand finally clicked like tap-to-pay did for mobile wallets. But we're not actually there yet. The supply arrived, but consumers barely got to try it before the shopping event was underway.

What we don’t know yet is if conversion numbers we're seeing this weekend from genuinely new behavior, or are early adopters testing shiny new features during a high-intent shopping window? We'll need to see January usage numbers when there's no sale urgency to know if behavior actually shifted.

James Taylor, founder of ad-tech firm Particular Audience dropped a post that captures the contradiction at the heart of retail's AI strategy. According to his analysis, 71% of Cloudflare-protected sites block GPTBot, and 80% block Claude. (Cloudflare protects about one in five websites on the internet.)

[Note: I'm skeptical of these exact percentages and the methodology here, but the directional point stands.]

The reason is straightforward economics: browser-based AI agents load full pages—every script, image, dynamic element. That's expensive to serve at scale, especially when many of those sessions don't convert to purchases. So retailers block AI traffic to save on server costs.

But here's the problem: if AI assistants can't access your product catalog, they can't recommend it. They just default to competitors who haven't blocked them.

Taylor argues that retailers need to stop treating AI agents like scrapers and start providing structured access through protocols like Retail-MCP. That would let them control and meter AI access while keeping frontend performance fast for humans.

Strip away the inflated percentages and what actually happened was more modest, Per Salesforce Data

Mobile became default 72% of global orders and 79% of U.S. traffic came from mobile devices. Mobile wallets hit 30% of U.S. digital orders, up from 28% last year. This is real behavior change, but it's been building for years.

Margins held. Average Selling Price grew 5% year-over-year even with discounting staying flat. Retailers protected their margins while consumers traded up.

Customer service automation doubled. Salesforce reports agentic customer service conversations surged 86% week-over-week, with agent-completed tasks like returns doubling. This is where AI is already delivering concrete value—handling the volume spike without hiring seasonal workers.