A 2026 industry survey found that 57% of people say they'd trust a business less for using AI to answer its phone, up from 53% a year earlier — and that talking to an AI is closing in on hold time as callers' single biggest phone-service complaint (AnswerConnect, 2026). A separate dataset covering more than 1.4 million real AI-handled calls found the opposite: 99% of callers reacted positively or neutrally once the call actually happened (NextPhone, 2026). Both numbers are real. The gap between them, not either one alone, is what matters for a business deciding whether to put AI on its phone line.
What the survey actually found
The numbers come from a survey OnePoll ran for AnswerConnect in May 2026 — 6,000 adults across the U.S., U.K., and Canada, split roughly evenly by region (AnswerConnect, 2026). Trust is trending the wrong way year over year: 57% say they'd trust a business less for relying mainly on AI customer service (up from 53%), 85% would rather reach a real person than AI (up from 83%), and 59% call AI agents frustrating to deal with on a customer-service call (up from 54%). The sharpest line in the data is the crossover: hold time, the traditional top phone complaint, dropped from 35% to 31%, while naming an AI agent or chatbot as the most frustrating part of calling a business rose from 29% to 30%. On the current trend, AI overtakes hold time as the #1 phone complaint within a year or two.
| What people were asked (OnePoll/AnswerConnect) | 2025 | 2026 |
|---|---|---|
| Would trust a business less for using AI to answer calls | 53% | 57% |
| Would rather speak to a real person than AI | 83% | 85% |
| Find AI agents frustrating on a customer-service call | 54% | 59% |
| Name hold time the most frustrating part of calling a business | 35% | 31% |
| Name talking to an AI/chatbot the most frustrating part | 29% | 30% |
Worth saying plainly: AnswerConnect is a human-staffed answering service — a direct alternative to an AI receptionist, not a neutral bystander. That doesn't make the numbers wrong. OnePoll is a real market-research firm and n=6,000 across three countries is a serious sample, not a landing-page pop-up survey. But a company that sells live human answering is exactly the company with a reason to commission, and publish, the study that shows trust in the alternative getting worse. Read the trend, not just the headline.
What 1.4 million real AI-handled calls found
NextPhone, an AI receptionist vendor, published a different kind of number: an analysis of 1,446,980 business calls its own AI handled across 2,074 businesses in 2025, including 89,577 full conversation transcripts run through LLM-based sentiment classification (NextPhone, 2026). The result: 99.0% of callers expressed positive or neutral sentiment during the call itself, and only 1.0% expressed anything negative. Same caveat in reverse — this is a vendor grading its own product on its own unaudited data, not an independent study. But it's a record of what happened on real calls, not a survey about a category.
Two flawed datasets, pointed at two different moments
Neither number is the full picture, and that's the actual finding. The survey measures how people feel about "AI answering the phone" as an idea — a category they've read headlines about, maybe argued with a bad chatbot over, long before they've dialed a specific business today. The call data measures a specific conversation that either worked or didn't, in the moment. Stated opinion and lived experience diverge constantly in consumer research, and phone AI looks like a clean example: distrust the category in the abstract, tolerate or even prefer the individual interaction when it's fast and it works. A business betting its phone line on either number alone is betting on the wrong one — the category trend if it ignores its own call quality, or its own good transcripts if it assumes that immunizes it against a caller who already walked in skeptical.
The one number from the survey worth building around
Buried in the same AnswerConnect data is the more useful stat: 85% say it should be clearly indicated when they're interacting with AI rather than a person (AnswerConnect, 2026). That's not a legal question — Montana has no law requiring it, and most states that do regulate AI phone disclosure carve out ordinary business customer service. It's a design preference, stated by the same people who say they distrust the category, and it's fully within a business's control regardless of which way next year's survey trend runs. A caller told upfront, in one plain sentence, that they're talking to an AI assistant is being treated like an adult. A caller who figures it out three questions in, or never quite figures it out, is the caller who ends up in next year's 57%.
Where this gap actually bites: rural Montana and Northwest tourism corridors
The caller most likely to hang up on a bad AI interaction is the one with the least already invested in reaching that specific business. That's a real pattern in the Flathead Valley and across the Northwest, not a hypothetical: a lot of Montana service businesses are two- or three-person operations where the person capable of answering the phone well is usually out running a job, not sitting at a desk, and a meaningful share of calls into tourist-corridor towns like Whitefish or Coeur d'Alene come from someone with zero prior relationship to the business — a traveler off Highway 2, not a ten-year regular in Kalispell who already trusts the name on the door. That caller has no track record to fall back on if the first thirty seconds feel like a maze. They just hang up and call the next result.
- Disclose it's AI once, early, in one plain sentence — not buried in a disclaimer, and not hidden.
- Give an immediate, ungated way to reach a person if the caller asks — a real transfer, not another menu layer.
- Design the conversation for the actual call type your business gets most, not a generic script borrowed from a vendor demo.
- Read your own transcripts and sentiment, not a national survey or a vendor's aggregate — your call mix decides which number is closer to true for your phone line.