Nearly Half of US AI Users Admit Lying to Customer Service Bots
A TELUS Digital poll finds nearly half of regular U.S. AI users have bent the truth to AI customer service systems to reach a human, get a refund, or resolve issues faster.
A new poll from TELUS Digital finds that nearly half of regular U.S. users of AI tools admit to bending the truth when interacting with AI-powered customer service systems. The survey indicates that a significant share of consumers have misrepresented information, exaggerated a problem, or otherwise gamed an AI agent in order to get the outcome they wanted — such as reaching a human representative, securing a refund, or resolving an issue faster.
The findings point to a broader pattern of friction between consumers and automated service channels, suggesting that many people see dishonesty as a practical workaround rather than an ethical lapse when dealing with a bot rather than a person.
For companies racing to deploy AI across customer service, this is a signal that deployment alone doesn't guarantee trust or compliance from the people on the other end. If a large share of users feel compelled to misstate facts to get past an AI system, that system is failing at its core job: understanding and resolving the customer's actual need efficiently and fairly.
The behavioral implication is bigger than a single poll. When customers believe a system is rigid, slow, or indifferent, they route around it — and dishonesty becomes a rational response to a poorly designed interaction, not a character flaw. Leaders building AI-driven service need to treat this as a design and trust problem, not just a technology rollout.
— Nearly half of regular U.S. AI users surveyed say they have bent the truth to get past AI customer service systems.
This finding isn't really about honesty — it's about what happens when a system forces people to choose between truth and a timely resolution.
Most organizations will read this poll as a cautionary tale about AI gatekeeping, and miss the real lesson: customers lie to systems they don't trust to listen. The fix isn't harsher verification or smarter lie-detection — it's designing AI service flows that let people state their actual problem plainly and still get a fair, fast outcome. Any operator who finds users are gaming their bot should treat that as a direct measure of how broken the escalation path is, not a reason to make the bot harder to fool.
