This Is the First Agentic Shopping Season
AI agents have started buying on shoppers’ behalf. Here’s what that means for the brands they buy from.
This holiday season, AI agents are buying things on our behalf. The numbers are small, but they are large enough for me to call this the first agentic shopping season.
Our data shows purchases made through AI agents have about quadrupled since April, and they are growing roughly 43% month over month. The growth is fast, from a very small base. I shared those numbers with Asa Fitch at The Wall Street Journal because they mark a line we had not crossed before. Agents have moved from browsing to buying.
Adoption is still early. Only about 3% of the ecommerce brands we work with let agents complete transactions today. I expect the share to rise quickly once shoppers see agents get the order right.
The Traffic Came First
Agents showed up on websites well before they started buying. In April, AI agent requests reached 88% of human organic search visits across the sites we track. Most recently, our September AI Market Pulse found that AI platforms accounted for 2.25% of search referral traffic, up from 1.95% in August. ChatGPT drove 96.3% of those AI referrals.
Most sites are still not ready for agents. Our research found 97% of the top 20,000 websites do not offer agents a lighter format such as Markdown, which agents can process with less effort.
Every Model Shops a Little Differently
Each AI model evaluates a potential purchase in its own way. As I shared with Asa, that makes it more complicated for marketers, but it also makes it very interesting.
We see the difference in our own retail research. Google AI Overviews directed 30% of transactional citations to a major retailer domain, compared with 15% for ChatGPT. The same retailer can show up very differently depending on which AI engine a shopper uses.
Travel tells a similar story. In our latest holiday travel research, Hilton leads hotels in Google AI Overviews, while Marriott leads in ChatGPT. We published the full category rankings on October 21.
The Advantage People Underestimate
Most of the attention goes to ChatGPT, and the AI referral numbers explain why. But look at search overall. Our September AI Market Pulse shows Google still accounts for 86.3% of search referral traffic.
That lead is why I think Google has a big advantage in agentic commerce that people don’t fully appreciate.
To complete even a basic purchase, an agent has to handle payments and taxes. Google already has the wallet and the tax infrastructure. Its Universal Commerce Protocol also builds on the product feeds retailers already send. Payments, tax and product feeds will matter as much as the model itself.
The Real Battle Is Over the Customer
Every retailer I talk to wants to protect the same asset: the customer relationship they spent years building. This season, an agent is starting to stand between that customer and the store. When an AI platform decides which products a shopper sees, the retailer risks losing control of that relationship.
Amazon blocks outside agents. My read is that it wants to protect its direct relationship with every Prime household, and I understand why. For routine purchases, an agent can compare price, availability and delivery across retailers in seconds. Instant comparison puts real pressure on loyalty and could pull the relationship toward the AI platform. Brands need to give shoppers a reason to choose them, even when an agent does the comparing.
Established brands still have an advantage. Years of reputation, recognition and customer trust give them a strong foundation in AI search. That history also has a downside: AI can surface information about a brand from years ago, whether it reflects the company today or not.
What I Would Do Before Black Friday
Start by making your key pages easy for agents to read. When our customers serve agents a lean version of their pages, agents tend to come back more often. Arm, for example, more than doubled verified AI agent traffic to its optimized pages within weeks of turning on BrightEdge Agent Edge. That is one customer’s result, and it shows what happens when agents can reach the right information.
Agents ask practical questions. Is it in stock? Can it ship tomorrow? Do loyalty points apply? Put those answers on the page, and keep product ID, price and availability consistent in your structured data.
Finally, market to people and machines together. Shoppers search, ask AI and send agents to shop for them. Your brand has to appear at each of those points: in search, in AI answers and in front of agents. We’ve added the full readiness checklist in our guide to how AI shopping agents choose brands.
One Platform for Search, AI and Agents
At BrightEdge, we are bringing Google search, AI search visibility and AI agent activity together in one platform. That gives retailers a clearer picture of how shoppers and agents find their products, where visibility gaps exist and what needs attention.
BrightEdge Agent Edge serves verified AI agents a lightweight version of each page, without rebuilding the site or changing the CMS publishing workflow. AI Agent Insights shows which agents visit your site and where they run into errors. AI Catalyst tracks how AI answers mention and cite your brand, right next to your Google performance.
Agent Shopping Will Grow Category by Category
I don’t expect shoppers to hand every decision to AI agents at once. Delegation will vary by category, price point and urgency, and big-ticket purchases will require more thought. Still, agentic shopping is here, and every brand will have to manage it.
Google remains central to product discovery, and AI assistants now give shoppers more ways to compare and choose brands. That means brands need visibility in Google and AI search, and agents need access to accurate product information.
The crown jewel hasn’t changed. The customer is still the prize, and this season an agent may be the one walking through your door. We are tracking agentic commerce every week at brightedge.com/agentic-commerce, and I would love to hear what you are seeing.