AI shopping assistants are starting to deal directly with businesses

AI assistants are beginning to research products, compare prices and talk to businesses on behalf of shoppers. A new CallMiner report finds that 91% of organizations have concerns about using AI in customer service. For businesses, one risk stands out: if an assistant cannot read your product details, it may pass you over.

What changed

AI agents are software tools that act for a customer. They are increasingly comparing products and dealing with brands directly. As customer service shifts toward AI talking to AI, companies may struggle to verify that an agent is legitimate. They may also struggle to tell whether it accurately represents what the customer wants.

The worries

• 91% of organizations have concerns about AI in customer experience.
• 52% worry AI will spread false information to customers.
• 47% worry about security and compliance risks.

These figures come from CallMiner's Customer Experience Landscape Report 2026. The report also found that 62% of organizations are not making the most of their customer data, so useful insights go unused.

Why it matters for businesses

Eric Williamson, CMO at CallMiner, says brands must make sure AI agents can reach accurate information about what they offer. Product descriptions, pricing and availability need to be correct and organized so AI systems read them properly. If an agent cannot find or understand them, a brand can be overlooked before the customer ever considers it.

Williamson also points to generative engine optimization, or GEO. It means making sure your products show up in AI chatbots and AI-generated search results, alongside traditional search optimization.

People still matter

Agents can narrow the choices, but product quality, the experience after a purchase and brand trust still count. Williamson says, "the customer is still the customer." The report suggests simple, low-risk questions may suit automation. Complex or sensitive cases often work better when AI supports a human decision.

How to measure success

A finished automated chat is not the same as a solved problem. Williamson advises checking accuracy, resolution, repeat contacts, escalations, customer effort and sentiment. If a customer got a wrong answer or needed a person to fix things, the automation did not work. Nearly half of organizations use AI to assist agents in real time, and 43% use it to free staff for higher-value tasks.