Horizon Media released an agentic commerce study on June 3 that identifies a trust problem behind AI-assisted shopping. Sixty-eight percent of respondents distrust an AI agent when a purchase lacks human input, and 60% are uncomfortable sharing the data an agent needs to act.
Adoption is already real: nearly one in five respondents have used an AI agent to make a purchase, rising to 23% among Gen Z. Yet a successful transaction does not guarantee loyalty. Thirty-one percent say an autonomous purchase would make them less likely to return, even when the purchase succeeds.
Our read: brands face a trust tax when they automate the transaction but leave accountability unclear. Agentic commerce can reduce buying effort and still weaken the relationship if customers cannot see who approved the action, how their data was used, or what happens when the agent gets it wrong.
Key Takeaways
- Horizon Media surveyed 5,630 U.S. adults across a multi-phase agentic commerce study.
- Sixty-eight percent distrust AI agents making purchases without human input.
- Thirty-one percent may be less likely to return after an autonomous purchase, even when it succeeds.
- Trusted brands and money-back guarantees raise willingness to adopt AI shopping agents.
- Brands need post-purchase trust and accountability metrics beside conversion.
What Horizon Media Found About AI Shopping Trust
The official Agentic Commerce report page describes a proprietary, multi-phase study of 5,630 U.S. adults. The findings show interest and hesitation at the same time. Consumers see convenience in delegated shopping, but they do not automatically trust the agent or the brand behind it.
Brand familiarity helps. Forty-four percent are more likely to adopt an AI shopping agent when it comes from a trusted brand. Guarantees help too: 48% say a money-back guarantee would increase adoption. Those figures suggest that the strongest agentic commerce asset may be an existing promise the customer already understands.
The accountability gap is sharper. Only 6% feel fully confident in privacy protections, and 72% want clearer accountability. A purchase can be technically correct while still leaving the customer unsure about consent, data use, returns, or responsibility.
Why a Successful Purchase Can Still Cost Loyalty
Most commerce reporting ends at conversion. An agent found the item, completed payment, and reduced friction, so the event looks successful. Horizon’s 31% finding shows why that scorecard is incomplete. The customer may accept the result and still feel that control was taken away.
This changes how brands should measure agentic commerce. Our coverage of Mondelez’s agentic commerce strategy focused on visibility, citation, and sentiment inside AI systems. Horizon adds a fourth question after the transaction: did the autonomous experience increase or reduce the customer’s willingness to return?
The issue is not limited to consumer retail. A B2B buyer may allow an agent to renew a software seat, reorder supplies, or select a service tier. The order can be accurate while violating an internal preference, skipping a relationship owner, or creating uncertainty about who authorized the spend.
The Trust Controls Brands Need Before Delegated Buying
Brands cannot control every external shopping agent, but they can publish clear commercial terms and build accountable transaction paths. Google’s agentic commerce announcement presents a Universal Commerce Protocol intended to help agents and businesses work across discovery, buying, and support. Protocol support matters, but customer trust still depends on the rules a brand exposes through it.
- Visible authorization: Show what the agent can buy, the spending limit, and when human approval is required.
- Data boundaries: Explain which personal, account, and purchase data the agent can access and retain.
- Clear accountability: Name who handles a wrong order, disputed recommendation, failed delivery, or unwanted renewal.
- Easy reversal: Give customers a plain cancellation, return, or money-back route.
- Post-purchase confirmation: Summarize what the agent chose, why it chose it, and what the customer can change.
Payment infrastructure is part of that control layer. AWS Bedrock AgentCore Payments shows how technology providers are preparing for non-human transactions. Brands should pair payment capability with explicit authorization and dispute rules before asking an agent to act.
How Brands Should Measure the Trust Tax
Start by separating agent-assisted and agent-autonomous purchases. A customer who asks an agent for options but approves the final order has a different experience from one who delegates the full transaction. Combining those journeys hides where trust falls.
Next, measure beyond conversion. Track returns, cancellations, support contacts, disputed purchases, repeat purchase, brand preference, and the number of customers who reduce agent permissions after an order. Compare those results with human-approved purchases of similar value and complexity.
Catalog accuracy remains essential because a trusted transaction cannot start with bad product data. Our guide to Feedonomics agentic catalog exports explains the need to inspect pricing, configuration, availability, and channel access across AI surfaces. Horizon’s research adds the next check: whether the customer still trusts the brand after the agent acts on that information.
Finally, test the guarantee and accountability message before scaling. Horizon’s findings suggest both can change adoption. A brand should know whether clearer approval rules, privacy language, and reversal options improve willingness to use the agent without causing avoidable support demand.
Agentic commerce will not be won by the fastest checkout alone. The durable advantage is a transaction the customer can understand, approve, reverse, and trust enough to repeat.
Frequently Asked Questions
It is the loyalty and adoption cost a brand may face when an AI shopping agent completes a purchase without enough human control, privacy confidence, or accountability. The transaction can succeed while trust declines.
Horizon Media reports that 68% distrust AI agents when purchases lack human input. Sixty percent are also uncomfortable sharing the data required for an agent to act.
Horizon found that 44% are more likely to adopt an agent from a trusted brand, while 48% say a money-back guarantee would increase adoption. Clear accountability and privacy protections also matter.
Measure conversion alongside returns, disputes, support contacts, repeat purchase, permission changes, and customer trust. Separate agent-assisted purchases from fully autonomous purchases so the results remain useful.
