Will your customers hang up on an AI receptionist? What the data actually says

Some will. In Gartner's survey of 5,728 customers, 64% said they would prefer companies didn't use AI in customer service. Look at what worries them, though, and the objection narrows: the concern they raised most was that AI would make it harder to reach a person. A later Gartner survey put a number on the remedy, with 87% saying a company using generative AI for service must still offer a way through to a human agent. The top complaint is about being trapped.

An AI receptionist is a voice agent that answers your phone, holds an ordinary conversation, takes the caller's details and books the job. The question is whether the people ringing you will put up with one.

Do customers actually mind an AI answering the phone?

Many say they do. Far fewer say AI actually made things worse.

Start with what people say. 64% of Gartner's respondents, surveyed in December 2023, would prefer companies kept AI out of customer service. Deloitte's Dutch Digital Consumer Trends survey of 2,000 consumers aged 18 to 75, fielded in August 2025, found 58% would be less inclined to use a customer service handled only by AI. And in Invoca's 2026 buyer report, 693 US consumers who had made a high-stakes purchase in the past year were asked who they would prefer when both options are equally available: 59% picked a human representative, 11% an AI assistant and 21% a combination of the two.

Now the experience. In the same Invoca research, the share of buyers who said AI made their buying experience worse fell from 29% in 2025 to 18% in 2026, while 46% said it made things better. The Nationale Voice Monitor, now in its seventh edition and run by research firm Markteffect with AI company Y.digital, asked 1,016 Dutch adults and found more than seven in ten had already had customer contact with a voice assistant. The same survey found only 12% said the questions they put to a chatbot or voice assistant were answered well.

Attitudes to brands' own AI use are hardening on another front. Fractl's 2026 study of 1,008 US consumers, on AI in search and brand trust, found the share who say heavy AI use would reduce their trust in a favourite brand doubled in a year, from 20% to 40%. The study looks at how brands use AI in marketing and search.

None of that says customers love AI. In Invoca's own sample, 59% would rather have a human, yet only 18% say AI made their purchase worse.

What are they really objecting to?

Being cut off from a person, first of all.

That is the finding inside the Gartner headline. The primary concern customers raised was that AI would make it harder to reach a person, with job losses and wrong answers among the other worries. "Many customers fear that GenAI will simply become another obstacle between them and an agent," said Keith McIntosh, senior principal, research, at Gartner.

Two years on there is a number on it. Gartner surveyed 3,566 business and consumer customers in February and March 2026 and found 87% say it is essential for a company using generative AI to offer a way through to a human agent. In the same survey, 50% said their interactions are easier when companies use generative AI. Both are true at once, and together they are the design brief.

You have probably been on the wrong end of this yourself. Six minutes of menu, then hold music, then something that cannot understand the word "boiler". The resentment is real, and it lands on exactly what McIntosh describes: a system standing between the caller and a person.

That makes it a design problem, and design problems have fixes. An agent that answers quickly and offers a callback the moment someone asks for a person is aimed straight at the thing Gartner's respondents said they feared most.

Does hiding the AI work?

In one field experiment, yes. In the EU, the provider of the AI now has to build it so your callers are told anyway.

The evidence for hiding is uncomfortably strong. Luo, Tong, Fang and Qu ran a field experiment on more than 6,200 customers, published in Marketing Science in 2019. Undisclosed chatbots were as effective at winning a purchase as proficient human staff, and four times as effective as inexperienced ones. Disclose the bot before the conversation starts and purchase rates dropped by 79.7%. Customers turned curt and calls got shorter.

Three reasons not to read that straight across to your phone:

• Those were outbound sales calls. A lender's own call centre rang existing borrowers to sell them a loan renewal. Your caller dialled you, in the middle of a problem, and wants it fixed. Luo and colleagues themselves name the more dynamic inbound call as a setting their results still have to be tested in.
• The penalty shrinks with familiarity. The authors found prior experience with AI mitigates the disclosure effect. That paper is from 2019; in the Nationale Voice Monitor, more than seven in ten Dutch adults have since had customer contact with a voice assistant.
• In the EU, the disclosure is no longer optional. Article 50 of the EU AI Act requires the providers of AI systems that interact directly with people to build them so those people are informed they are dealing with AI, unless it would be obvious anyway. It has applied since 2 August 2026. Do not wait for a reprieve: the AI Omnibus pushed the AI Act's high-risk deadlines out to December 2027 and August 2028, and the only Article 50 duty it gave extra time was the separate one to mark AI-generated content, for systems already on the market. The duty to tell people they are talking to an AI kept its date.

So let it say so in one line at the start, and get on with the job. Luo's numbers say that has a cost. In the EU it is also what a compliant product has to do.

Which calls does an AI actually get away with?

Short, straightforward ones, on the evidence there is. For anything emotional or genuinely unusual, we would put a person on it.

Here the evidence comes from real callers rather than a survey. Wang and colleagues studied a large telecoms firm that replaced its automated phone menu with voice AI, and published the result in Production and Operations Management in 2023: demand for a human rose at first, then complaints fell persistently. On simple requests complaints fell for experienced and inexperienced callers alike, and the drop was larger among older customers and long-time users of the old menu system. The same study found that when the speech recognition failed, both demand for a human and complaints went up.

For a plumbing or heating round, this is how we would split the calls:

The call	Who should take it
New job, address, rough symptom, book a slot	AI, comfortably
Out-of-hours emergency triage: is the water off, how bad, how soon	AI, then you
Rescheduling, "are you still coming"	AI
An angry customer about work you already did	You, immediately
A quote negotiation or an unusual property	You

Age is the caveat to take seriously. Pew Research Center, surveying US adults in February 2026, found 31% of 18-29s were extremely or very confident in their ability to use AI chatbots, against 6% of over-65s. Older callers also lean on the phone more in Invoca's data: 52% of US Boomers prefer calling when they need help, against 41% of all respondents. Which is exactly why the way out has to be one sentence away, every time.

You are not choosing between an AI and a human

If nobody else answers your phone, you are choosing between an AI and nobody.

The figures above answer different questions. Invoca's offered people a human as the alternative; Gartner's and Deloitte's are about AI in customer service in general. Your caller at 11pm on a Sunday gets your voicemail, and under 3% of callers sent to voicemail leave a message. If you employ someone to answer, the comparison is fair and the trade-offs are worth working through properly. If you don't, the question that matters is an AI or nobody.

Gartner's 64% is about AI in customer service in general. Your caller at 11pm is choosing between an AI and a ring-out.

Invoca's research also found 75% of US buyers have hung up because they were on hold too long, up from 50% a year earlier. The customer who abandons an AI is annoyed. The one who reaches voicemail, going by Invoca's call data, almost never leaves you a message to call back.

How to answer with an AI without losing the caller

Five things we would do, in this order:

1. Say it in the first sentence, then stop. "You've reached the AI assistant for Jansen Plumbing" is a disclosure. Thirty seconds of terms and conditions invites the caller to hang up.
2. Always offer the way out. If someone asks for a person, the agent should take a message and flag it for a callback. Being unable to reach a person is the concern Gartner's respondents raised first.
3. Give it one job. Capture the details and book the slot. An agent that tries to negotiate a bathroom refit is taking on the part of the call where a mistake costs real money.
4. Keep the callback promise. A flagged call that nobody rings back is a broken promise: you took their details and their patience first.
5. Listen to the recordings for a fortnight. You will hear where callers hesitate, and that is the part to fix first.

In SwiftGuard, the opening line says it is an AI assistant and that the call is recorded, and the agent takes a message for anything it cannot sort out itself, which lands in your dashboard for you to ring back. Booking straight into your calendar needs a connected Google Calendar.

The number that settles it

Run your own count. Two weeks of missed calls, then ask yourself how many of those people would have preferred a slightly stilted conversation at 2am to a beep and a dial tone.

That is the real question, and your own call log holds the answer. When you want to see what answering them costs, the plans are here, from €69 a month ex VAT.