Do customers mind talking to AI on the phone?

What the surveys actually measure

There is a lot of confident writing about whether customers accept AI on the phone, and much less solid data. The best public sources measure general attitudes to AI, not reactions to a call with a home service company. They are still useful, as long as you read them for what they are.

Pew Research Center's 2025 survey found that 50% of US adults are more concerned than excited about the increased use of AI in daily life, up from 37% in 2021^[1]. In the same survey, 57% rated the societal risks of AI as high, compared with 25% who rated its benefits as high^[1]. The Stanford AI Index reports that only 39% of people in the United States see AI products and services as more beneficial than harmful^[2].

The same Pew work also found that nearly three-quarters of US adults would let AI assist them at least a little with day-to-day tasks^[1]. Put together, the picture is of a public that is skeptical about AI in the abstract but willing to use it for practical help, especially when it is clear what the tool is doing and who is responsible for it.

The comparison callers are really making

When owners ask whether customers mind talking to AI, they often picture the comparison with their best CSR on a quiet Tuesday morning. That is rarely the choice. The real alternative at 9 p.m. on a Saturday, or at 10 a.m. in the first heat wave of the summer, is voicemail, a busy signal or an answering service that takes a message.

Homeowners are already struggling to reach anyone. In Angi's 2025 survey, 53% of homeowners reported difficulty finding pros^[3]. And callbacks are unreliable: eight in ten Americans say they generally do not answer calls from unknown numbers^[4]. A caller who gets a clear, useful answer on the first call, even from software, is usually better served than one who leaves a message that turns into a missed callback.

So the useful question is not whether callers prefer software to a person. It is whether a caller who would otherwise reach voicemail gets a better outcome. That is something you can test in your own data within a few weeks, by comparing booked jobs and callback workload for the hours the system covers with the same hours before launch.

What makes callers comfortable or annoyed

Most of these are not about AI at all. They are the same things that make callers happy or angry with a human receptionist. The difference is that a software system does them the same way every time, for better or worse, so your instructions decide the experience.

Disclosure: how and when to say it is AI

People want to know. In Pew's survey, 76% of US adults said it is extremely or very important to be able to tell whether pictures, videos and text were made by AI or by people^[1]. Some states also have disclosure rules for automated or AI interactions. Whatever the law requires where you operate, the business case points the same way: disclose clearly, and never deny being AI when asked.

• Up front: Thanks for calling [company]. I am the company's AI assistant and I can take your details, check our schedule and reach the on-call technician if it is urgent.
• Short version: Hi, you have reached [company]'s automated assistant. How can I help?
• When asked directly: Yes, I am an AI system working for [company]. I can take care of this now, or arrange for a team member to call you back by [time].

Up-front disclosure costs a few seconds and avoids the moment of feeling tricked that callers remember. It also sets expectations about what the system can do, which reduces frustration when it says it cannot answer a question. Our guide to AI voice call disclosure rules covers the federal and state picture.

Trust: the impersonation problem

Callers are primed for fraud. Nine in ten US adults think people often or sometimes pretend to be someone else to steal personal information by phone or online^[5]. A smooth voice asking for personal details can trigger exactly that suspicion. Design calls to reassure:

1. Say the company name in the first sentence, and the city or service area if it helps callers recognize you.
2. Never ask for card numbers, bank details, Social Security numbers or passwords on the call.
3. Tell callers which number the technician or office will call back from.
4. Offer a person or a callback whenever the caller sounds uncertain.
5. Make sure what the system says matches your website and your trucks, so the caller can check.

Calls that should reach a person quickly

Acceptance also depends on what the system does not try to handle. Callers forgive software that hands over quickly on difficult calls; they do not forgive software that keeps them in a loop when they are scared or angry. Decide in advance which calls go straight to a person or to your on-call path:

• Anything involving gas, carbon monoxide, smoke, burning smells, sparking or active water near electricity.
• Callers who say they are elderly, disabled or alone and without heat or cooling in extreme weather, if your rules treat that as urgent.
• Complaints about a recent visit, refund requests and anything that mentions a lawyer, an insurer or a review.
• Commercial accounts and property managers with service agreements, if they expect a named contact.
• Any caller who asks for a person twice.

Tell your team what the system handles and what it passes on. When CSRs and technicians know the rules, they can explain them to customers who ask, and they stop treating every AI-handled call as a problem to redo. That internal acceptance matters as much as the customer's, because a team that distrusts the system will quietly work around it.

How to measure acceptance in your own calls

1. 01

   Early hang-ups

   Track the share of callers who hang up in the first ten to fifteen seconds. A rise after launch suggests the greeting is the problem.

2. 02

   Requests for a person

   Count how often callers ask for a human, and at what point in the call. Clusters show you which questions the system handles badly.

3. 03

   Completed intakes

   The share of calls that end with full details and a booking, request or on-call alert is the best single measure of usefulness.

4. 04

   Callback connection rate

   For calls that end in a callback promise, check how many callbacks reach the customer.

5. 05

   Complaints and reviews

   Tag any complaint or review that mentions the phone experience, positive or negative.

Compare these numbers with the same weeks before launch where you can. Acceptance is not a feeling; it shows up in behavior.

Example: reading the first month of data

Example: a fictional cleaning company starts AI answering for evenings and weekends. In the first two weeks, about one caller in eight hangs up within ten seconds. The owner listens to those recordings and finds that the greeting runs for twenty seconds before asking a question. She shortens it to one sentence plus a disclosure and the early hang-ups drop by half. Requests for a person cluster around one topic, questions about moving-out deep cleans, so she adds approved answers for those. By the end of the month most calls end with a booked or requested visit, and the two complaints received are both about the price of a service, not the phone experience.

Some callers will always prefer a person, including some older customers. Give them an easy route, and see our guide on serving senior customers for wider tips.