AI Agents Are Shopping This Black Friday: Can Retailers Tell Who Sent Them?
Key Takeaways
- •AI shopping agents influenced $14.2 billion in Black Friday sales last year, establishing themselves as a significant sales channel rather than a fringe experiment.
- •Malicious bot traffic jumped 135% during the same holiday season, meaning the days with the most valuable agent-driven sales also bring the most hostile automation.
- •Traditional defenses such as CAPTCHAs, rate limits, and device fingerprinting assume fast, precise traffic is malicious, an assumption that agentic commerce breaks and that risks turning away real customers with real credit cards.
- •Proof of Human systems like World ID let people verify once in person with the Orb device and let their agents present zero-knowledge proof they act for a unique human without revealing identity.
- •Personhood credentials could enable per-human purchase caps on limited releases, reduce checkout friction for verified shoppers, and spare retailers from collecting additional personal data.

On Black Friday, a growing share of the shoppers reaching checkout pages will not be human. They will be AI agents that people have dispatched to compare prices, track restocks, and complete purchases faster than any person could click. Last year, AI shopping agents influenced $14.2 billion in Black Friday sales. That is no longer a fringe experiment — it is a sales channel.
Over the same holiday season, malicious bot traffic jumped 135%. Scalpers, card-testing scripts, fake account farms, and promotion abusers arrived at the same door as the helpful agents. From the retailer's side of the screen, the two look nearly identical: both are automated, both move fast, and both skip the homepage banner. Both figures land in the same holiday window, which is what makes the overlap so difficult: the days when agent-driven sales matter most are also the days when malicious automation is most active.
That leaves retailers with an unattractive choice.
The Block-or-Allow Trap
For two decades, the standard defense has treated automation itself as the threat. CAPTCHAs, rate limits, device fingerprinting, and behavioral checks all rest on one assumption: humans are slow and messy, while bots are fast and precise. Anything fast and precise, therefore, gets stopped.
Agentic commerce breaks that assumption. A shopper who instructs an assistant to "buy the best-reviewed air fryer under $100 before it sells out" wants that assistant to be fast and precise. Block it, and the retailer has turned away a real customer with a real credit card — one who will simply buy from a competitor that let the agent through. Let every bot through, and the retailer takes on the fraud, chargebacks, and inventory hoarding that come with it.
Neither option works in a season measured in billions.
A Better Question
The useful question is not "is this a bot?" It is "who sent it?" An agent buying a gift for one parent is a customer. A thousand agents spun up by a single operator to empty a limited drop is an attack. The software might be identical. What differs is how many real people stand behind the traffic, and whether each one is acting once or posing as a crowd.
This is where Proof of Human comes in. A person verifies once that they are a real, unique human and receives a credential they can carry. When their agent reaches a checkout, it can present a cryptographic proof that it acts for a verified person, without revealing who that person is.
World ID, from the World network, is one working example. People verify once, in person, with World's Orb device. After that, the credential lets them prove they are a unique human using zero-knowledge proofs, so a site can check the claim without learning who they are. The merchant gets no name, no email, and no ID scan — just a yes-or-no answer to the one question it needs answered.
What Changes at Checkout
For retailers, that answer enables policies that were not possible before. A limited release can be capped at one unit per human instead of one per account — a limit that fake accounts have long rendered meaningless. Agents backed by a verified person can skip the friction built for anonymous traffic, while unverified automation still gets checked closely. New-customer promotions reach new people rather than a single operator using 500 email addresses.
For shoppers, it removes a frustration they rarely mention but always notice: being treated as a suspect. They get fewer puzzles, fewer "are you a robot?" loops, and agents that are not thrown out halfway through a purchase.
Because the proof reveals nothing beyond "human, unique," it also avoids the usual cost of stronger verification. Retailers do not collect more personal data they must then protect, and customers do not hand over more than they want to.
The Season Ahead
Agents will not replace holiday shoppers, but they will act for them more and more. Retailers can keep chasing automation one bot at a time, or they can start asking the question that separates a customer from an attack. The signs to watch are concrete: whether limited releases shift from per-account to per-human caps, and whether checkout flows begin accepting a personhood credential alongside the CAPTCHA.
This Black Friday, plenty of the traffic at checkout will be bots. The retailers that do well will be the ones that can tell which bots a real person sent.
This article was first published by Metaverse Post.