A new Capacity survey finds a third of consumers have already sent an AI agent to handle a service request, and nearly half say AI has contradicted a company representative or its policies. When the two disagree, customers are about as likely to believe the machine.
Key Takeaways
For most of the history of customer service, the company held the rulebook. The agent read the refund policy off a screen, and the customer either accepted the answer or asked for a supervisor. That asymmetry is ending. Customers now arrive with their own AI tools, which have read the terms of service, summarized the return window, and sometimes drawn a different conclusion from the one the company’s agent is about to give.
Capacity’s Agentic Consumer Report, released October 27, surveyed 1,000 U.S. consumers. 35% have used an AI agent to handle a customer service request at least once, and 49% of Gen Z and Millennial respondents have done so. Counting any AI tool, including general-purpose chat assistants, 55% have used AI to help with a service request.
The finding that should worry operations leaders is what happens next. 44% of consumers say AI has contradicted a company representative or the company’s policies. And when the AI and the representative disagree, consumers are about equally likely to trust either one. A company’s own agent, reading from the company’s own systems, now starts many conversations with no more credibility than a chatbot the customer brought from home.
That changes how customers behave once they are on the line. 34% say AI leads them to ask more specific questions during service interactions. 23% would ask representatives to explain or cite company policies because of what AI told them, and 20% would escalate issues more quickly. 36% would use AI to research return, refund, or cancellation policies before they make contact. David Karandish, founder and CEO of Capacity, said businesses “need to rethink how they measure success” as that behavior spreads.
A contradiction can come from either side. A customer’s AI may misread a policy or blend it with a competitor’s. But companies often publish one version of a policy on the website, another in the help center, and a third in the agent knowledge base. Any AI reading those sources will surface the inconsistency, and the frontline agent is the one left to defend it.
Customers are not inclined to give the company the benefit of the doubt. In Zingtree’s Beyond Basic CX study, conducted by CMSWire INSIGHTS among 1,040 U.S. consumers in July and August, 67% believe companies design support around their own cost savings and efficiency. Half had at least one complex issue that was never resolved, and 37% of those unresolved issues failed because the company could have solved the problem but its process was too difficult or slow. 68% would abandon AI support after two failures or fewer.
Patience is thin on the human side too. A Consumer Cellular survey of 482 adults aged 50 and older, reported by Telecom Reseller on October 20, found more than eight in ten would consider leaving a company if an automated system trapped them without a way to reach a human. 47% of existing customers had switched providers or seriously considered it because of a poor automated service experience. Asked what matters most in a service interaction, 43% chose complete resolution and 43% chose speaking with a real person. Only 4% chose speed.
Put together, the research describes a customer who arrives better informed, more skeptical, and quicker to escalate, and who is judging the company on whether it resolves the problem rather than how fast it answers. The same Telecom Reseller report notes a 2026 Gartner survey of 3,566 customers in which 87% considered access to a human agent essential when companies use generative AI for customer service. The human is not going away. The human is now the person who has to settle the argument between two machines.
That puts new weight on something most service operations treat as housekeeping: whether every published policy, help article, and agent script says the same thing. If a customer’s AI and the company’s agent read the same consistent source, they are far less likely to disagree. If they read different versions, the contradiction is built in before the call starts.
The shift is subtle but real. Customer service used to be a conversation between a customer and a company. Increasingly it is a conversation between two sets of software, refereed by a person in the middle. The companies that fare best will be the ones whose policies are consistent enough that both sides of that conversation read the same answer.
Guide
When a customer calls with an AI-drafted reading of your refund policy, the conversation becomes a negotiation. This guide covers what operations leaders must change so agents resolve those high-stakes voice calls on the first contact.
Download
Guide
If 44% of customers have watched AI contradict a company, the first question is whether your own policy data is clean and governed enough for any AI to read correctly. Gartner sets out the data foundations that decide it.
Download
Guide
Customers’ AI tools now repeat and remix whatever a company publishes, so mixed messages surface faster than ever. This guide explains why enforcing one consistent standard across everything a company produces has become a competitive necessity.
Download
Zingtree finds 50% of consumers repeat themselves after an AI-to-human transfer, and nearly 9 in 10 tell Alchemer AI has hurt their service experience. The fix sits in the handoff, not the bot.
Ricoh finds 90% of decision-makers believe their tools and systems are connected, while 76% of employees say their jobs would be easier if those tools worked better together. The gap is now slowing AI adoption.
A survey of 2,527 enterprise leaders finds 74% have shut down or rolled back a live AI customer agent, and the best-governed programs roll back most often. Nearly all are raising the budget anyway.
