Shoppers Bring Their Own AI to Customer Service, and It Contradicts the Brand

For a decade, brands have argued about whether to put AI in front of their customers. The customers have now settled part of the argument themselves. A growing share of people arrive at a service conversation having already asked a general-purpose assistant what the return policy says, what they are owed and how to phrase the complaint. Sometimes that assistant is right. Often enough, it disagrees with the company.

That changes what a service interaction is. It is no longer a customer asking the brand for an answer. It is a customer checking the brand's answer against one they brought with them, and two new surveys released this month suggest the brand does not automatically win that comparison.

The Second Opinion Is Already in the Room

Capacity's Agentic Consumer Report, released on October 27, surveyed 1,000 US consumers. Some 55% have used an AI tool, such as an agent or a large language model, to help with a customer service request, and 35% have used an AI agent for one at least once. Among Gen Z and Millennial respondents that figure rises to 49%, and 46% of respondents with children aged 25 or younger at home have used a personal AI agent, against 29% of those without.

The headline finding is the friction. Some 44% said AI had contradicted a company representative or its policies. When the two conflict, Capacity reports that consumers are equally likely to trust the AI or the representative. The company's own staff, in other words, no longer carry automatic authority over a policy the company wrote.

The behaviour shift is visible in the conversation itself. Some 34% say AI leads them to ask more specific questions, 23% would ask a representative to explain or cite company policy, and 20% would escalate faster when using AI tools. More than a third, 36%, would use AI to research return, refund or cancellation policies, and about 40% have used, would use or plan to use a personal agent during peak seasons with long wait times. David Karandish, Capacity's founder and chief executive, argued that businesses “need to rethink how they measure success, moving beyond speed and efficiency.” Capacity sells service automation, so the framing suits it, but the contradiction figure is a problem for any brand regardless of vendor choice.

Interest Is Running Far Ahead of Use

Lorikeet's Agent Economy report, fielded between September 30 and October 8 among 1,966 US consumers, puts the trend in proportion. Some 66.4% are interested in an AI agent that deals with companies for them, 18.7% very interested. But only 13.9% said an AI tool had contacted a company on their behalf, and only 6.3% said the agent contacted the company itself, which Lorikeet treats as an upper bound of roughly one in 16.

The early users are the signal worth watching. Among them, 76% have done it more than once and 26.6% do it weekly or daily. Their most common jobs are looking things up (62.0%), drafting messages they then sent (50.0%) and contacting the company directly (44.9%), with returns, refunds, cancellations, claims and rate negotiations among the tasks named. Consumers are most comfortable delegating the hunt for a better price, at 55.5%, and split almost evenly on making a complaint, at 41.8% comfortable against 40.8% uncomfortable.

That pattern matters for brand teams. The first jobs consumers hand to agents are the ones where they suspect the company is not on their side: price, refunds, cancellations and complaints. An agent arriving with that brief will test every inconsistency between what the website says, what the help centre says and what the representative says.

People Still Want a Person, on One Condition

None of this means customers want to talk to machines instead of people. Verint's State of Customer Experience 2026 research, based on 5,000 US consumers surveyed earlier this year, found that 61% of customers prefer a human agent, a preference up 5% year on year. Yet 69% of those who prefer humans said they would switch to automated self-service if it could fully resolve their issue, and 79% of consumers say they would switch to a competitor after a single terrible experience.

Read together, the three studies describe the same customer. They will use whatever gets the problem solved, they increasingly bring their own AI to check the answer, and they punish inconsistency hard. A brand whose chatbot, help pages and frontline staff give three different readings of the refund window is now handing its customer's assistant the evidence to escalate. Verint's respondents also ranked not having to repeat information when transferred to a human as the fourth most important factor in a good experience, which puts the hand-off between bot and person squarely inside brand perception.

Making the Brand's Answer the One That Holds

• Publish one canonical version of every policy. Returns, refunds, cancellations and price matching should live in a single source that the website, help centre, chatbot and agents all draw from, so an outside assistant finds the same answer everywhere.
• Write policies for machines as well as people. With 36% of consumers willing to research policies through AI, plain, structured, unambiguous wording decides what a third-party assistant tells your customer before they ever contact you.
• Train representatives to cite, not assert. When 23% of customers will ask for the policy behind an answer, staff need quick access to the exact clause rather than a paraphrase.
• Measure contradiction, not just handle time. Track how often a customer arrives with a conflicting AI answer, which policies trigger it, and how those contacts resolve. That is a brand consistency metric, not only a service one.
• Carry context across the hand-off. Customers rank not repeating themselves highly. Pass the full conversation and customer record from the bot to the human so the second answer builds on the first.
• Prepare for peak season now. About 40% expect to lean on personal agents when wait times rise. Stress-test the holiday returns policy against a general-purpose assistant before your customers do.

The arrival of consumer-side AI does not take customer service away from the brand. It raises the price of being inconsistent. When nearly half of consumers have already seen AI contradict a company, the brands that hold their customers' trust will be the ones whose answer is the same in every channel, whichever assistant comes asking first.