Retailers are racing to win agentic attention ahead of Black Friday and Cyber Monday. The question they should be asking is whether they can convert that attention into actual sales.
Every retail conversation seems to have a new audience: agents. The advice is familiar by now. Structure your product data so large language models can find it, understand it and recommend it. Let’s imagine the strategy works. Your product appears in an AI-generated answer. A shopping assistant visits your site, compares products, creates an account, or moves toward a transaction. You’ve captured a new kind of attention.
Then what? Are you ready to receive the traffic you’ve worked so hard to attract?
Attention Is Not the Same as Readiness
The first mistake is to assume agentic attention behaves like human attention.
A human visitor arrives with a browser, a device, a set of intentions and a degree of context that retailers have spent decades learning to interpret. An agent may arrive on behalf of a person, another system or no legitimate customer at all. It may be gathering information, comparing options, executing a task, testing a credential or probing for a weakness.
From the outside, those interactions can look remarkably similar. That is the central tension in the agentic internet: the traffic that looks like a promising new customer may also be automated abuse wearing the shape of a promising new customer.
The question is not simply whether an interaction claims to be ChatGPT, Claude or another recognized agent. Names and user-agent strings are not identity. They can be copied, manipulated or used as camouflage. A request presenting itself as a familiar AI assistant may be a legitimate agent acting for a real buyer. It may also be a spoofed system, a scraper or a fraud operation using AI because it knows businesses are eager to welcome machine-driven demand.
And the landscape is already more varied than the phrase “AI agent” suggests. Cloud agents, agentic browsers and browser extensions can all reach the same digital property, sometimes through nearly indistinguishable paths.90% of agentic traffic observed by HUMAN is stealth by default. The traffic does not arrive carrying a reliable badge that tells a retailer what it is, who authorized it or what it intends to do.
Influencing an Order Is Not the Same as Completing One
Salesforce reported that AI and agents influenced 20% of all global orders during Cyber Week 2025. That number has shaped a lot of planning for this season, and the verb matters: influencing an order and completing one are not the same thing.
DoorDash offered the clearest illustration. On the company’s third-quarter earnings call, CEO Tony Xu was asked about agent-based ordering. His answer: “we’re seeing maybe several hundred orders in total from some of these third-party agents.” Several hundred, at a company that completes roughly three million deliveries a day.
That tracks with what latest trends in agentic traffic: across September, 60.4% of agentic activity went to ecommerce sites, 81% of it landed on search and product pages, and 2.4% reached checkout. Meta’s Muse, which arrived at number one in its first month, sent 76.5% of its traffic to ecommerce and converted 2.6% of it.
So agents are arriving in volume and browsing intensively, but they are very rarely buying. For retailers heading into Black Friday and Cyber Monday, that gap is the whole story, because agentic traffic is already shaping the numbers you use to judge performance without showing up in the revenue those numbers are meant to explain. Page views inflate. Measured conversion falls. Neither movement tells you anything about demand.
Routing Matters More Than Welcoming
The next phase of agentic commerce will not be defined by whether retailers let automated traffic in. It will be defined by whether they can route different kinds of traffic intelligently.
Different participants need different experiences:
- A human buyer may need a rich, persuasive journey
- A legitimate agent may need clear product data, structured policies, inventory information and a reliable path to act on its user’s behalf
- A training crawler may be governed by a content-access policy
- A suspicious interaction may deserve additional verification
- A malicious operation should be stopped
This is why “block or allow” is becoming an inadequate operating model. It treats every automated interaction as interchangeable at exactly the moment automation is becoming more diverse, more commercially important and harder to distinguish by appearance alone.
It also requires marketing, security and fraud teams to share more than a meeting. Without shared definitions and shared accountability, marketing celebrates a surge in agentic traffic that security experiences as an attack wave.
Four Actions for Retailers Before Black Friday
None of this has to wait for a new platform or budget cycle.
- Establish a baseline now, while the week is still ordinary. If you don’t know what share of your traffic is agentic today, you will have nothing to read your Cyber Week numbers against.
- Separate agentic activity in your reporting rather than letting it blend into the totals, so conversion and engagement stay interpretable when volume spikes.
- Decide in advance which agents get which access, and be specific: browsing, account creation, authentication, checkout. Those are different permissions, and peak week is a poor time to be making the call live.
- Agree who owns the decision when marketing and security disagree. That conversation is uncomfortable in November and nearly impossible in December.
The Takeaway
Traditional retail metrics were built around human visibility. Sessions, add-to-carts and conversion rates remain useful, but become harder to read when a growing share of activity is mediated by an agent. A page view may be a crawler collecting competitive pricing. A completed transaction may be legitimate commerce, an unauthorized purchase, or a test of stolen payment credentials.
Being mentioned by an AI system isn’t the same as being chosen by a customer, and being chosen by an agent isn’t the same as being ready to serve, verify and retain the customer behind it.
In the agentic internet, attention is only the beginning. The advantage will belong to the retailers who can recognize what arrives, route it responsibly, convert it safely and tell the difference between activity and growth.
Sarah Acker is SVP and Global Head of Marketing at HUMAN Security, where she leads worldwide marketing strategy spanning brand, creative, communications and product marketing. She writes regularly on agentic traffic, measurement and digital trust.





