AI Agents: 74% Positive Feedback, Conditional Trust

The public is more favorable towards AI agents than a year ago and shows a high level of receptiveness after demonstrations. However, trust remains conditional: it depends on the tasks, channels, and explicit guarantees. Respondents cite escalation to a human and clear disclosure of AI as prerequisites, while the loyalty of shopping assistants raises doubts.

Shopping: The Loyalty of AI Assistants Called into Question

The expansion of AI agents to commercial uses raises questions about their impartiality. One respondent indicates that they assume an AI shopping assistant prioritizes the financial interests of the company over those of the customer. Another fears that the agent is programmed to favor certain products or purchasing choices. This concern about loyalty leads companies to reassure users about the purpose of the technology. Even though the general sentiment is improving and trust is progressing, it remains conditional.

Trust Varies by Channel: Chat in Front, Social Media Behind

The trust placed in an AI agent depends on the interaction channel. Users are most inclined to trust it via chat, a channel they are most familiar with for this type of exchange. Conversely, they are less willing to grant full trust on social media, where they are used to interacting with humans. The phone is the most polarizing channel: some users who never use AI find it frustrating to talk to an agent, while those who use it frequently often prefer an agent to take their call. This polarization reflects the level of familiarity with AI.

What Users Demand: Accountability and Human Recourse

The main expectations concern the agent's ability to stand by its responses and demonstrate judgment in nuanced situations. Some users perceive AI agents as isolated, without human supervision, and want guarantees regarding the agent's accountability and the ability to quickly access a human if needed. Many also believe that AI strictly applies the rules, even when an exception would be justified. The two most cited levers for establishing trust are the ability to easily escalate to a human and clear information, from the outset, that they are interacting with AI.

More than Half of Users Limit Their Trust to Simple Tasks

A relative majority of users only trust AI for simple or routine tasks: 54% report this. The notion of complexity varies among individuals. Some doubt the AI's ability to demonstrate human discernment, worry about the handling of their personal data, or question the agent's ability to keep track of all information exchanged during a prolonged conversation. Users are more concerned about how their problem will be resolved than about the mere accuracy of the response. For nuanced or emotional requests, trust remains to be built.

Exposure to the Tool: 74% Positive Opinions After Demonstration

The perception of AI agents has improved compared to last year for customer service. 49% of respondents describe positive experiences, which is nine points higher than in 2025. Each respondent was exposed to a video showing an AI agent resolving a real request, then answered the same questions again. After this demonstration, the share of positive opinions reaches 74%. Respondents also believe that the technology has improved. Trust, although on the rise, remains conditional and can evolve. The authors of the report emphasize that customer experiences can be designed to strengthen trust, using levers that are within the reach of companies.