Nearly Half of US AI Users Admit Lying to Chatbots to Reach a Human
A TELUS Digital poll of 1,000 US AI users finds 46% have exaggerated or lied to automated customer service to reach a human, and 30% abandoned purchases over AI-only support.

Why it matters
- 46% of regular US AI users have given exaggerated or inaccurate information to automated customer service to reach a human, per a TELUS Digital poll of 1,000 adults.
- 77% of respondents said guaranteed human support would make them more likely to choose a company.
- 30% of respondents abandoned a purchase after finding AI was the main customer service option.
- 57% of 18-to-29-year-olds admitted lying to reach a person, versus 32% of those aged 62 to 75.
- 23% named getting the AI to understand the problem as the single most burdensome part of automated support.
Forty-six percent of regular US AI users say they have deliberately given exaggerated or inaccurate information to an automated customer service system just to reach a human being. That figure comes from a new poll of 1,000 US adults released by TELUS Digital, a customer experience (CX) services provider — and it lands at a moment when enterprises are shifting support workloads to AI at speed, betting on cost savings the same customers appear increasingly unwilling to absorb.
The headline number deserves scrutiny in both directions. It measures self-reported behavior, not observed behavior, and respondents may over-report clever defiance or under-report it. Still, the pattern is consistent across the survey: customers are not passively accepting automated support, and a measurable share are gaming it.
What did the poll actually find?
TELUS Digital commissioned an online Pollfish questionnaire of US adults aged 18 to 75 who regularly use AI assistants such as ChatGPT, Gemini and Claude. The questions covered customer service interactions across retail, travel, financial services and healthcare. The sample is users, not the general population — so the data describes the behavior of people already comfortable with AI tools, which arguably makes their dissatisfaction more telling.
The core numbers:
- 46% have given exaggerated or inaccurate information to get past automated customer service and reach a person.
- 77% said they would be more likely to choose a company that guarantees human support.
- 89% have tried at least one workaround when automated support fell short.
- 57% chose an AI-powered customer service tool because they expected it to be the fastest option.
- 30% have abandoned a purchase after finding AI was the main customer service option.
- 56% avoid AI customer service at least half the time when another channel is available.
Two of those figures sit in tension. Fifty-seven percent pick AI tools expecting speed, yet more than half dodge those same tools when an alternative exists. The poll does not resolve whether the speed expectation fails in practice or applies only to simple queries; it records the contradiction without explaining it. What it does establish is that AI-first support is simultaneously the preferred entry point and the most avoided channel, depending on the situation.
Where does the friction come from?
Asked to name the single most burdensome part of using AI in customer service, respondents pointed first at comprehension. 23% said the biggest problem was getting the system to understand the issue at all. Another 22% cited having to repeat the same information, and 15% said the main burden was difficulty reaching a human — or having no option to do so.
Smaller complaints trailed behind: 10% cited inaccurate or contradictory answers, 9% pointed to waiting for the AI to respond, and another 9% highlighted navigating menus or irrelevant response options. Smaller shares said they were unsure whether the AI could resolve the issue, or felt the problem was taking more time than it was worth.
In plain words: the top three complaints all describe a system that neither listens nor hands off cleanly.
How are users working around the systems?
Only 12% of respondents said they give up altogether when automated support falls short. Everyone else adapts. The survey breaks the adaptations down:
- 47% simplified or rewrote their explanation.
- 42% asked the same question in a different way.
- 35% explicitly asked for a human agent.
- 28% restarted the chat or opened a new session.
- 26% repeated the same request several times.
Bending the truth was not a fringe tactic. Nearly half of respondents admitted doing it at least once, and 33% said they had done so more than once. The behavior skews sharply by age: 57% of respondents aged 18 to 29 said they had given exaggerated or inaccurate information to reach a person, against 32% of those aged 62 to 75.
Read that gap carefully. The youngest cohort — the demographic enterprises most want to serve through digital channels — is also the most willing to manipulate those channels. Automation designed to deflect contact may instead be training its most valuable users to lie to it.
What is the commercial risk?
The survey quantifies consequences beyond annoyance. The 30% who abandoned a purchase when AI was the main support option represent direct revenue exposure. Another 26% said they would give feedback or file a complaint after being forced into a workaround, and 17% said they would lose trust in the company's ability to address their needs.
Jamie Timm, Global SVP of Service Delivery and Operations at TELUS Digital, framed the findings as evidence of design failure rather than technology failure.
"Nearly half of the people we polled said they've bent the truth to get past an automated system, and most said they have used other workarounds, which suggests gaps in how some customer experiences are built," Timm said.
"Those gaps matter because revenue may be at stake," Timm continued. "Some respondents said they have walked away from a purchase when AI was the main way to get help, and most said guaranteed access to a person would make them more likely to choose one company over another. Keeping customers and reducing frustration calls for AI that understands requests and resolves issues, alongside human agents who are easy to reach and have real-time guidance and context when a customer needs a person."
Timm added: "Customers shouldn't have to use a workaround to get help."
One caveat belongs here. TELUS Digital sells exactly the omnichannel AI-plus-human integration the poll's findings recommend — the company explicitly says it "helps enterprises bring the two together in a cohesive omnichannel experience, with the consulting, engineering and operations to run it." The data may be sound; the messenger has a commercial stake in the conclusion that enterprises need help fixing it.
How should companies split the work?
Timm outlined a four-way model TELUS Digital uses to decide where automation belongs in CX operations:
- Self-service: simple tasks customers can handle themselves.
- Automation: routine requests AI can resolve end to end.
- Human-led with AI support: complex issues where a person steps in, backed by AI.
- Proactive outreach: the company contacts the customer first based on their circumstances and preferences, before they need to ask.
"With enterprises, we look at the interactions their customers have and build an AI strategy designed to support each type," Timm said.
The poll does not test whether that model works, and TELUS Digital offers no outcome data for it in the release. What the survey does establish is the direction of pressure: with 77% of regular AI users saying guaranteed human access would influence their choice of company, AI-first support has become a competitive variable customers can see — and one a measurable minority will lie to escape.
Source: itbrief.com.au
More from James Calloway
Show full bio
News editor covering people, launches and day-to-day practice in artificial intelligence and machine learning at AI In Context.
722 articles