4 Ways Merchants Can Reduce It

AI agents make up less than 1% of site traffic, which could lull merchants into a false sense of security. While the number is small, it’s growing rapidly: Agentic orders on Forter’s network increased 7.2x from the first week of August to the last week of October. As traffic increases, agentic risk naturally becomes a bigger concern for merchants.

Agentic risk is the exposure merchants take on when AI agents shop on behalf of customers. Surveying digital commerce leaders at IMPACT NYC, we found that 69% consider it to be their biggest concern. Because the landscape is evolving at a breakneck pace, the best strategy is adaptable and iterative rather than complete.

Read on to learn four ways to capitalize on agentic commerce while mitigating risk.

1. Continuously re-evaluate your approach to agentic traffic

Merchants have spent years trying to keep agentic traffic off their sites. Bots were considered universally bad, scraping prices and stuffing credentials. Now they may be among your best customers. Blocking agents outright trains them to understand that they cannot transact on your site.

Start by evaluating whether your site allows agentic traffic and whether you can distinguish the malicious bots from AI shopping agents. The latter often look identical to (extremely fast) human shoppers. You should also be able to tell who’s behind an agent, as fraudsters can use them just as easily as legitimate shoppers.

Models change constantly and so will the best approach, which makes regularly retesting and working across internal teams monumentally important. Treating this as something to set and forget opens you up to a major agentic risk: Leaving your site vulnerable to an army of agents exploiting a loophole.

2. Treat agents like customers

Agentic commerce protocols work like a paved road for agents — but merchants have to adopt them. Scraping agents, such as Muse and Instinct, don’t require any opt in, scrolling your site, often without announcing themselves. They still have to be able to find and navigate your site.

Optimizing for discoverability, many merchants realize they didn’t have enough clean product data. That means taking a step back to fix the underlying data platforms.

The data gap highlights a risk problem. With incomplete or inconsistent product data, agents may buy the wrong item or miss a final-sale condition. When a customer disputes the charge, the liability most often falls to the merchant. 

Audit your product data, terms, and conversion flows the way an agent would navigate them. Rules built for human shoppers don’t always hold up against an agent, which may offer to place repeat orders to get around per-cart SKU limits or reverse-engineer cart dollar limits, for example.

3. Set one policy for humans and agents

Policy abuse used to be concentrated among a small subset of customers. AI has lowered the barrier to entry for return abuse, chargebacks, and class-action claim-filing. Many merchants think about agents at the transaction, but customers are also sending them to handle tedious customer service, claims, and return tasks. Think end-to-end, including the post-purchase experience.

The merchant should set clear policies, rather than leaving them open to interpretation and allowing the agent to decide what’s acceptable.  Your policies should be the same for humans and their agents, ensuring they’re treated the same on returns, warranties, and similar matters. If an agent is representing a human, it should be held to the same rules.

Once you set policy, automation can make decisions at scale. Decide in advance where an agent’s authority ends, before and after the sale. Some merchants accept orders but don’t fulfill them without customer confirmation, while others scale the friction based on order value. You’ll also need to decide whether agents can reach customer service, or whether certain tasks should route to the human shopper. Where human intervention is needed depends on your own risk tolerance.

A customer’s history should shape that call. If their pattern shows a sketchy history of Item Not Received (INR) claims, for example, you may want to manually review their agent’s activity.

4. Define KPIs to measure agentic risk

You can’t manage agentic risk you can’t see. Start by isolating agentic traffic, and tracking fraud rates and velocity for that segment. Because agentic traffic is comparatively low, blended numbers may hide a spike in AI agent fraud.

By isolating that traffic segment, you gain a better understanding of how agents shop and return. Conversion by page shows where they succeed and where they get stuck, and what that means for merchandising. Tracking the approval rates of agentic orders also ensures reliability and security. Chargebacks are also worth watching. If you look at dispute rates for agentic orders separately, you can see how frequently customers claim not to have authorized an agent’s purchase: something else blended numbers will conceal.

Different teams often measure success differently. Align early on shared metrics, so that they’re already established during a fire drill.

See live data from the first 60 days of agentic orders

How do AI agents move through checkout? What are typical authorization rates for agentic transactions? Are chargeback rates higher on agentic transactions?

On November 12, 2026, at 1:00 PM ET, join Forter VP of Payments Galit Michel for The First 60 Days of Agentic Transactions, where she’ll answer those questions and more with live data from the first 60 days of Muse, Instinct, and Grok Bot transactions.