Visa published a Trust Index for agentic commerce this week, and the headline number tells you everything about where we are with AI agents in business: 72% of US consumers have already used an AI assistant for shopping, but only 23% trust AI to actually manage their payments.

That 49-point gap is not a technology problem. It’s a trust problem. And for any business building AI agents that touch customer money, it’s the thing worth understanding before you go further.

What the Research Found

The Visa Trust Index for Agentic Commerce surveyed 2,065 US consumers in late May 2026. The findings land at an important moment — AI agents capable of browsing, comparing, adding to cart, and checking out are now real products, not demos. Consumers are already interacting with them. What they’re not ready to do is hand over the keys to their wallet.

Key numbers from the research:

• 72% of US consumers have used an AI assistant for shopping tasks
• 23% trust AI to manage their payments autonomously
• 61% would trust Visa specifically to handle AI-initiated transactions
• The most trusted brand for agentic payments was Visa, ahead of both tech companies and traditional banks

The implication: consumers draw a hard line between AI as a research and discovery tool versus AI as a financial decision-maker. Getting information is one thing. Spending money is another.

Why the Trust Gap Exists

This isn’t irrational consumer resistance. It reflects a genuine change in what’s being asked.

When you use an AI assistant to research products, the worst case is a bad recommendation. You still decide to buy. But when an AI agent completes a transaction on your behalf, you’ve delegated financial authority. Errors, unauthorised charges, and fraud scenarios are now the agent’s responsibility — or rather, they feel like they are.

Three things are driving the hesitation:

No transparency into agent reasoning. Most consumers have no visibility into why an AI agent made a particular purchase decision. The “show your work” layer that would build confidence doesn’t exist in most current implementations.

No clear accountability chain. If an AI agent makes a bad purchase, who fixes it? The merchant? The platform? The AI provider? Consumers don’t know, and that uncertainty is expensive.

No trusted intermediary until now. The 61% trust figure for Visa is significant precisely because it shows consumers will extend trust to AI payments — but only when a recognised financial brand they already trust is standing behind the transaction.

The Business Design Problem

For companies building customer-facing AI agents — whether for e-commerce, service booking, or subscription management — the Visa data suggests a clear product design principle: separate the AI shopping layer from the AI payment layer, and make the payment layer explicitly human-controlled until trust builds.

That’s not a permanent limitation. It’s where the market is right now.

The businesses that will win in agentic commerce are those that earn trust incrementally. Start with AI agents that recommend, inform, and summarise. Let the customer confirm the purchase. Track the accuracy of those recommendations over time. Build a record that the agent’s choices would have been good ones. Then, once that track record exists, offer opt-in payment delegation to customers who want it.

This is how trust gets built in financial services — slowly, with evidence, and with a clear exit available at every step.

What This Means for Voice AI Specifically

The trust gap is particularly sharp for voice AI deployments. A customer talking to a voice AI employee for service queries or product information is one thing. Asking that voice agent to process a payment or amend a subscription is another — and the Visa data confirms customers are not ready to treat those as the same interaction.

For businesses deploying voice AI in customer-facing roles (contact centres, booking systems, frontline support), the practical implication is: separate the conversation from the transaction. The voice agent handles dialogue. Payment happens through a channel the customer already trusts — a secure link sent to their phone, a portal they log into, a representative who steps in for the financial moment.

This isn’t a failure state. It’s a design choice that matches where consumer confidence actually is.

What Visa’s Position Tells You About the Future

Visa’s 61% trust figure isn’t just interesting market research — it’s a strategic signal. Visa has spent 60 years building trust in payments infrastructure. That trust now transfers to agentic contexts, which is why Visa has moved aggressively into the agentic commerce space with its Agentic Ready programme and token-based payment infrastructure for AI agents.

The implication for businesses is that partnering with established financial infrastructure providers — rather than building proprietary payment agents — is probably the faster path to consumer adoption. Consumers already know how to trust Visa. They don’t yet know how to trust your AI agent.

The Bigger Picture

The Visa Trust Index captures something that often gets missed in discussions about AI agent deployment: adoption is not just a capability question. The technology to let AI agents shop and pay on behalf of consumers exists today. What doesn’t exist yet is the earned trust to make mass adoption work.

Businesses that treat trust as an afterthought — shipping capable agents and assuming consumers will figure out how to feel comfortable — are likely to hit adoption walls. Businesses that treat trust as a design input from day one, building transparency, accountability, and fallback mechanisms into every agent interaction, have a cleaner path.

The 72% who already use AI for shopping are your early adopters. The 23% who trust it for payments are your leading indicators. The gap between them is the work.

---

Enterprise DNA helps businesses build and deploy AI agents that drive real outcomes — from customer-facing voice agents to back-office automation. If you’re thinking about where AI fits in your customer experience stack, start a conversation with the team.