64% Prefer No AI: What the Gartner Customer-Service Survey Means
The “64% prefer no AI” statistic is not a myth to wave away. In a Gartner survey published in July 2024, 64% of customers said they would rather companies did not use AI in customer service. A separate 53% said they would consider switching if they learned a company planned to use AI for service. Gartner surveyed 5,728 customers in December 2023. The finding is a meaningful warning about trust, access to people and wrong answers.
What the survey does not tell you is that 64% of callers reject every AI receptionist, that 64% will leave your business, or that a particular vendor's voice agent solves the concern. Gartner did not run a controlled test of Trillet or modern phone receptionists. Its principal reported concern was that AI could make it harder to reach a person. The sensible response is to design and measure a caller journey that respects that concern, not to dismiss the respondents as confused by old phone menus.
This article puts the original finding beside newer first-party research, then gives small businesses a practical way to evaluate AI answering against their current phone experience.
What did the Gartner 64% survey actually ask?
Gartner's 2024 press release reports a customer-service preference, not an AI-phone-call satisfaction rate. Respondents were asked about companies' use of AI in customer service broadly. The release does not say that the 64% had only used legacy IVR, that they were all picturing a scripted chatbot, or that they failed to recognize newer voice systems. Those explanations appeared in earlier versions of this article without evidence and should not be used to discount the finding.
The related 53% statistic also needs its verb: respondents would consider switching if they found out a company was going to use AI for customer service. That is not an observed 53% churn rate. “Would consider” describes a risk worth understanding, not a forecast that half of a small business's customers will leave after one call.
Gartner reported three leading concerns: losing access to a person, job displacement and incorrect answers. A business owner cannot solve all three simply by choosing a nicer voice. They can address the first and third with a clear human route, honest disclosure, bounded tasks and regular call review. Whether customers accept that design is something to test with their own callers.
Is the statistic about bad AI or all AI?
It is about people's stated preference toward companies using AI for customer service. It is not evidence that every AI implementation is equally disliked, but it is also not evidence that the respondents were only rating “bad AI.” A later, better-resolved interaction might change some opinions; the original survey does not measure how much.
More recent research keeps this nuance. In a February–March 2026 Gartner survey of 3,566 B2B and B2C customers, 50% said GenAI made service interactions easier, while 87% said it was essential to have an option to reach a human when GenAI was used. Both views can coexist: a customer may appreciate a quick automated answer and still insist on an escape route when the question becomes complex.
That newer Gartner survey did not establish a Trillet satisfaction or conversion rate either. Its practical message is to avoid making AI a compulsory obstacle. Offer it where it can complete a suitable task or collect useful context, and make the path to a person understandable before the caller becomes frustrated.
Does the survey account for voicemail as the alternative?
The 64% result did not compare a particular AI receptionist against your business's voicemail, missed-call rate or staffed coverage. For a small business, the real alternative may be a person answering, a callback, voicemail, an answering service or no successful connection at all. Which one is best depends on your actual route and staffing.
Do not replace the Gartner finding with unsupported claims that “virtually nobody prefers voicemail,” that 80% of callers always hang up, or that each missed call costs a fixed dollar amount. Those figures were used in the old article without a reliable comparable population or method. A missed call is not automatically a lost sale; an answered call is not automatically a completed booking.
Instead, check your own last month of calls:
- How many relevant inquiries reached a person, reached voicemail, rang out or were returned later?
- How many callers left a message or called back, and how many became bookings?
- Which requests were routine enough for an agent, and which needed a person from the start?
- What would an additional answered minute cost after the plan allowance, plus messaging and forwarding?
That baseline makes a pilot meaningful. You can compare a tested AI route with a real current route, rather than pitting a perfect hypothetical AI against a caricature of voicemail.
What does current research say about AI satisfaction?
COPC's AI customer-experience study, published in January 2026, surveyed more than 1,000 people across six countries who had recently interacted with AI customer service. It found 74% satisfied with their most recent interaction and more than 90% satisfied when the AI resolved the issue without further steps. It also found that failed resolution could be associated with a large fall in Net Promoter Score and that AI-to-human handover was a frequent weak point. The sample covered chat, voice and messaging, not only phone receptionists.
These numbers cannot be set against Gartner's 64% as though they were opposite answers to the same question. Gartner asked about preference toward companies' AI use; COPC asked about satisfaction with a recent interaction. The samples, countries, time periods and channel mixes differ. A satisfied user can still prefer to speak with a person next time, and an AI skeptic can still appreciate a resolved routine request.
Metrigy's 2025–26 consumer study adds a sobering check. Among 503 adults surveyed in the United States and Canada, 84.9% preferred a human agent; 80.1% still preferred a human even when assured either option would resolve the issue. That does not mean 80.1% will reject your specific receptionist. It does mean “people only care about resolution” is too simple.
The best reading of all three sources is that task completion, transparency and human access matter, while human preference remains real. None offers a universal satisfaction benchmark for Trillet or any other AI receptionist.
Does selection bias make the 64% figure unreliable?
The published Gartner release gives the survey size and fieldwork period, but does not support the old article's claim that a “vocal minority” of unhappy users dominated the sample. It is possible to ask reasonable questions about any survey's sample, wording and geographic relevance without inventing a direction or size for bias. We should not call a reported 64% a vocal minority when 64% is the majority of that survey's respondents.
The same caution applies to newer studies. COPC surveyed recent AI users across six countries; Metrigy surveyed North American adults; neither is a random sample of your Australian customers calling your particular business. Their findings are evidence about broader attitudes and experience design, not a substitute for a local pilot.
If a vendor claims its callers have an 85%, 92% or 100% satisfaction rate, ask for the population, channel, survey response rate, time period and definition of success. Completed bookings alone are a narrower denominator than every attempted call. A trustworthy number should include failed connections, abandonment, follow-ups and requests that should have been escalated.
What is the actual cost of avoiding or adopting AI?
There is no single “cost of avoiding AI.” A firm with a responsive staffed line may gain little from automation. A solo operator with regular after-hours demand may benefit from a carefully scoped answering route. A regulated practice may face additional privacy, safety and contracting work that changes the economics.
For an ordinary inbound small-business workflow, Trillet's current AI receptionist pricing is $49/month for 150 included voice minutes, then $0.20 per additional voice minute. At 300 voice minutes, the simple plan-plus-voice calculation is $49 + 150 × $0.20 = $79 for that month, before optional SMS, phone-number or carrier-forwarding charges. The D2C offer includes a 28-day money-back guarantee on the plan fee, not a free trial or a guarantee of customer satisfaction.
To estimate value, use a conditional scenario. Suppose your call log shows 20 relevant missed inquiries per month, a tested route helps complete two additional inquiries, and each completed inquiry contributes $100 after direct costs. That would be $200 in contribution before the receptionist bill, staff follow-up and other expenses. The calculation is not a claim that every missed call is worth $100 or that AI caused both completions; run it with your own data and compare to a control period if possible.
An agent can also create cost if it routes poorly, misstates a policy or delays a needed person. Include those failure modes in the pilot, not just the subscription fee. For a fuller cost picture, see how much an answering service costs.
How should a small business use the 64% finding?
Treat it as a design test. Before a live rollout, answer the concerns Gartner actually identified:
- Can a caller reach a person? Decide which calls may transfer, which require owner follow-up and what happens if the destination does not answer. Test the route on the real line. A transfer function's presence is not a guarantee that every transfer succeeds or carries full context.
- Can the agent give a wrong answer? Give it approved, current hours, prices, services and booking rules. Test exceptions and require a message or escalation when the information is uncertain.
- Does the caller know what is happening? A clear introduction that the caller is speaking to an AI assistant can set expectations. COPC reported higher satisfaction among respondents who knew they were interacting with AI, but that is an observed association, not proof disclosure alone caused the difference. Check applicable recording and notice requirements for your workflow.
- Does the system work on this phone line? Test busy, no-answer, after-hours, carrier voicemail and calendar behavior. The approximately five-minute Trillet setup claim is for an initial draft, not for a completed production deployment.
- What do your customers say? Review a sample of real call records, complaints, bookings and human follow-ups. Ask a simple feedback question where appropriate and compare like-for-like periods.
Trillet can handle ordinary inbound reception tasks within its public D2C scope, including supported booking through Google Calendar, Cal.com (including Outlook through Cal.com) and GoHighLevel Calendar. Other CRM or practice-system connections are DIY through the platform API, not automatic native D2C integrations. Email call summaries are available; SMS is optional and separately billed.
Keep the scope narrow for high-stakes calls. The public Terms of Use do not authorize HIPAA-covered patient information on the $49 self-serve plan; that requires Agency or Enterprise with an executed BAA and applicable Order Form. Emergency or clinical decisions also require an expressly agreed regulated deployment rather than a general receptionist script.
Frequently Asked Questions
Will 64% of my customers leave if I use an AI receptionist?
No study cited here measured that outcome. Gartner found that 64% of surveyed customers preferred companies not use AI for service and 53% would consider switching if a company planned to use it. Neither percentage is an observed churn rate for your business. Pilot a suitable call flow and watch your own customer feedback and repeat contact.
Is the Gartner 64% statistic wrong?
No. The statistic and survey size are correctly reported by Gartner. What is wrong is claiming it only captured legacy IVR or that a modern voice agent makes it irrelevant. The result describes a real preference; it does not test a specific AI receptionist.
Do people become satisfied if AI resolves the issue?
COPC reported satisfaction above 90% among its resolved AI customer-service interactions, across several channels. That is an association in a different sample, not a guarantee for an AI receptionist or proof that everyone prefers AI when it works. Metrigy's later study found most North American respondents still preferred a human even when assured of equal resolution.
Should I tell callers they are speaking with AI?
Clear disclosure helps callers understand the interaction and when to ask for a person. COPC found higher reported satisfaction among those who knew they were using AI, but a business should also check its notice, recording and sector obligations. Do not claim the agent is a person.
How can I start without replacing my current receptionist?
Consider a tested busy/no-answer or after-hours route for low-risk inquiries, or use the AI as a first point of contact only for defined tasks. Keep a human escalation or follow-up process. Trillet's direct receptionist starts at $49/month for 150 voice minutes, with $0.20 per extra minute and a 28-day plan-fee money-back guarantee; configure and test the actual route before relying on it.
The 64% figure is a reason to listen carefully to customers, not a reason to ignore AI or dismiss their preference. A good decision is measurable: what was answered, what was resolved, what reached a person, what failed, and what callers thought of the result.
