AI in customer service: 64% would rather skip it (Gartner)

The most quoted survey on AI in customer service shows resistance with a specific reason behind it. In a Gartner survey of 5,728 customers, fielded in December 2023 and released in July 2024, 64% said they would prefer that companies did not use artificial intelligence in customer service, and 53% said they would consider switching to a competitor if they found out a company was going to use AI for service (Gartner, 2024). The most cited concern, at 60%, was that it would become harder to reach a person. In 2026, the same firm found that only 27% of customers would try a chatbot again after a bad experience and that 87% consider access to a human agent essential when a company uses generative AI in service. Added to the academic studies on chatbot aversion and to the complaints filed on Brazil’s Consumidor.gov.br, these numbers sketch the operation customers accept: AI to solve simple requests fast, a visible human for everything else and transparency about who is on the other side.

At Iasbeck&Co we implement CRM and AI agents in B2B operations, and almost every customer service project arrives with the same question from the executive team: will customers accept talking to a bot? I wrote this for people who lead customer service, CS and post-sale, for the executives who approve the budget and for the HR team that trains the staff. The practical design of triage, draft replies and handoff rules is in AI in customer service. Here the subject is what customers say, and what they do, when they run into AI along the way.

Which studies measure what customers think of AI in customer service?

Gartner publishes its customer surveys in press releases that give sample and fieldwork period, without the full questionnaire. Two peer-reviewed marketing studies measure behavior, and the Consumidor.gov.br bulletin, from Brazil’s National Consumer Secretariat, shows what Brazilians take to the government when service fails.

Source	Year	Sample and method	Main finding
Gartner	2024	5,728 customers, December 2023	64% would prefer companies did not use AI in service; 60% fear it will be harder to reach a person
Gartner	2026, with 2025 data	5,801 customers in the US, January and February 2025	54% trust human agents more than AI for product or service recommendations; 32% trust AI more
Gartner	2026	3,566 B2B and B2C customers, February and March 2026	49% would use a chatbot, 7% did; 27% would try again after a bad experience; 87% see access to a human as essential
Luo, Tong, Fang and Qu, Marketing Science	2019	Field experiment with more than 6,200 customers randomized to a sales call from a chatbot or a human	Disclosing the bot before the conversation cut purchase rates by more than 79.7%
Crolic, Thomaz, Hadi and Stephen, Journal of Marketing	2022	Five studies: real data from an international telecom company and four experiments	With angry customers, a humanized chatbot lowers satisfaction, firm evaluation and purchase intention
Senacon, Consumidor.gov.br Bulletin	2025	2,328,863 complaints closed in 2025	Service/SAC accounted for 11.91% of complaints, with no split for automated service

What did Gartner’s 2024 survey measure?

The July 2024 press release is the origin of the 64% figure, and the chart that comes with it says more than the headline. The question on concerns had 5,140 answers, because Gartner excluded those who said they did not know what AI technology means, and each person could pick more than one option. The same release notes that 60% of service leaders were under pressure to adopt AI in their function. Analyst Keith McIntosh sums up the customer’s fear: generative AI becoming one more obstacle between them and an agent.

Customer concern about AI in customer service	Share
It will become more difficult to reach a person	60%
AI will displace people’s jobs	46%
AI will provide the wrong answers	42%
My data will be less secure	34%
AI biases will not treat customers equally	25%
Other	3%

Source: Gartner, 2024, State of the Customer Survey, 5,140 respondents.

For anyone designing a service operation, the first row of the table is worth more than the 64%. Customers fear losing their path to a person, a fear built over years of phone IVRs and chat menus that end in “sorry, I didn’t get that,” and wrong answers only come third. Gartner itself recommends in the release that the chatbot say it will connect the customer to an agent when it cannot solve the issue and hand the conversation to the human from the point where it stopped. The 53% who would consider switching measure intention in a hypothetical question, which usually runs above real switching, and they show that customers link AI in service to risk before they meet your company’s chatbot.

What changed between 2024 and 2026?

Gartner’s later surveys show the resistance moving from opinion into behavior, with questions that change from edition to edition and do not form a comparable series. They work as a counterweight to the firm’s forecasts on how much agents will resolve on their own, discussed in the article on the CRM market.

Trust: a human for recommendations

In a survey of 5,801 customers in the United States, run in early 2025 and cited in an April 2026 release, 54% said they trust human agents more than AI for product or service recommendations, against 32% who trust AI more (Gartner, 2026). The firm uses the figure to argue for human involvement in complex, high-stakes or advisory interactions. In a B2B operation, a recommendation is the moment the customer asks which plan to buy or whether to move to another module, and that is where the expansion sale begins.

Use: plenty of willingness, little practice

The February and March 2026 survey, with 3,566 B2B and B2C customers, showed the gap between saying and doing. Forty-nine percent said they would have been willing to use a chatbot if the company had provided one, and only 7% used a chatbot or digital assistant in their most recent service interaction (Gartner, 2026). Analyst Eric Keller explains the difference through past experience: when a chatbot misunderstood the issue, gave generic information or made it harder to reach a person, customers choose another channel. In their latest interaction, customers used third-party generative AI tools about three times more than company chatbots. Customers do use AI, the one they chose, outside the company’s channel.

The second chance is short

For anyone about to launch an agent, the heaviest number in the release is another one: only 27% of customers would try a chatbot again after a negative experience. Gartner calls the effect a leaky bucket, because one poorly resolved interaction drives the customer away even if the chatbot improves later, and recommends prioritizing reliability over reach. In the same survey, 87% said access to a human agent is essential when a company uses generative AI in customer service. The firm’s guidance is to start with issue types that have a proven resolution rate, tell customers what the chatbot can do and widen the scope once performance is stable.

Why do customers react badly to chatbots, according to academic studies?

Gartner’s surveys measure what customers declare. The two studies below measure what they do, with a comparison group, and explain the mechanism behind the distrust.

Disclosing the bot cut purchases

The most cited field experiment on the subject is by Xueming Luo, Siliang Tong, Zheng Fang and Zhe Qu (Marketing Science, 2019). More than 6,200 customers were randomized to receive highly structured outbound sales calls from a voice chatbot or from human workers. Without knowing they were talking to a machine, customers bought from the chatbot as much as from proficient workers and four times more than from inexperienced ones. When the chatbot disclosed it was a machine before the conversation, purchase rates fell by more than 79.7%, calls got shorter and customers became curt.

The authors point to the mechanism: people who know they are talking to a bot see it as less knowledgeable and less empathetic, even when its objective performance matches a human’s. The negative effect shrank when disclosure came later in the conversation and among customers with prior AI experience. I do not read the study as a license to hide the bot. I read it as evidence that the bias weighs on results and fades with good experiences, and in a B2B relationship, where the same account deals with the company for years, a customer who later finds out they talked to a machine without knowing carries that distrust into the renewal.

Angry customers and human-looking chatbots

Cammy Crolic, Felipe Thomaz, Rhonda Hadi and Andrew Stephen studied humanized chatbots, with a name, an avatar and human-like language (Journal of Marketing, 2022). In a large real data set from an international telecom company and in four experiments, they found that when customers enter the conversation angry, the humanized chatbot lowers satisfaction, overall firm evaluation and later purchase intention. For customers who were not angry, the negative effect did not appear. The mechanism is inflated expectations: the human-looking bot promises more than it delivers, and the angry customer punishes the gap.

For the operation, the study puts the customer’s emotional state inside the routing rule. Complaints, disputed charges and service outages arrive angry more often, and in those cases the sentiment classification at triage should send the customer down the shortest path to a person.

And in Brazil, what do the complaints about service say?

I found no official Brazilian source that separates complaints about automated service from the rest. The Consumidor.gov.br 2025 Bulletin, from the National Consumer Secretariat (Senacon), records 2,328,863 complaints closed in the year. The Service/SAC problem group accounted for 11.91% of them, behind billing and disputes, with 52.14%, and contract and offer, with 12.56%. In some segments service tops the list: among health plan operators, SAC requests left unresolved, unanswered or answered late were the most cited problem, at 13.17%, and among content and internet service providers, difficulty of contact or slow service reached 13.81%. The record does not say whether the request went through a bot, an IVR or an agent.

On the rules side, Decree 11,034 of 2022 regulates customer service (SAC) for companies in services regulated by the federal Executive, such as telecom, electricity, health plans and aviation. It requires SAC access 24 hours a day, phone service with a human for at least eight hours a day and complaint and cancellation options in the first menu. The decree also forbids making initial access to an agent conditional on the consumer providing data first. For companies outside that scope, including many B2B operations, the text is a reference for what Brazilians have learned to expect.

How do you turn the research into a service design?

The sources above combine into a routing rule by type of request. AI pays off where the customer wants speed and the answer can be checked, and the human pays off where judgment, emotion or money are at stake.

Situation	Who handles it	Basis in the data
Simple, repetitive request, such as status, duplicate invoice, access and hours	AI agent resolves it, with a visible option to talk to a person	Gartner recommends starting with issues that have a proven resolution rate
Recommendation of plan, product, module or migration	A human, with AI preparing the account history	54% trust humans more for recommendations (Gartner, 2025)
Complaint, disputed charge, service outage	A human as soon as possible, with no humanized bot in the way	Crolic and colleagues (2022): anger plus a humanized chatbot worsens satisfaction and purchase intention
Cancellation	Visible option from the first menu, with an agent	Decree 11,034 for regulated services; 60% fear not being able to reach a person
Customer who already had a chatbot failure	Straight to a human on the next contact	Only 27% would try again after a bad experience (Gartner, 2026)
Any conversation with an agent	Identification as an automated assistant and handoff to a human with context	87% see access to a human as essential (Gartner, 2026)

In a B2B operation, the agent that sends a duplicate invoice frees the analyst for the expansion conversation, and the agent that handles a strategic customer’s complaint badly creates work for the account executive. The cadence that connects those pieces is in B2B post-sale.

What these data do not tell you

All three Gartner surveys reach the public through press releases, with questionnaire and country breakdown in paid reports; only the 2025 survey states the country, the United States. The 2023 fieldwork predates most generative AI agents in production, and the 2026 survey confirms the distrust with different questions. Gartner sells advice to service leaders, and the conclusions in the releases point to the practices it recommends to those clients.

Almost all the numbers measure stated preference or intention. The 64% and the 53% are not observed churn, and the 2026 survey shows saying and doing drifting apart: 49% willingness against 7% use. The samples lean toward end consumers, and a B2B buyer, with an account manager and a contract, may react differently.

The academic studies measure behavior with a comparison group, which makes them stronger for talking about cause. Luo and colleagues measured outbound voice sales calls in 2019, and Crolic and colleagues studied chatbots from the generation before today’s language models. The mechanisms they describe should still hold, but no one has yet measured, with the same rigor, how strong the effect is with today’s agents.

The Consumidor.gov.br bulletin counts only those who complained on the public platform, with banks making up 42.84% of complaints, and it does not report the channel where the problem started. I found no Brazilian survey with an open method on preference for AI or for humans in customer service. The percentages about “Brazilians who prefer human service” that circulate in the press usually come from software vendor surveys without a published questionnaire and sample, and I left them out.

What leaders do with these numbers

1. Classify requests by type and risk. List the 20 most frequent contact reasons of the last three months and mark, for each one, whether the answer can be checked, whether it involves money or a contract and whether it usually arrives with an angry customer.
2. Start with what the agent does well. Turn on automated replies for simple, repetitive reasons after measuring the accuracy rate in shadow mode. Gartner recommends reliability before reach, and only 27% give a second chance.
3. Keep the human visible. Make sure the option to talk to a person is on every screen of the conversation, set a maximum number of messages before transfer and keep complaint and cancellation one step from the start.
4. Route by sentiment. Send the customer that triage flagged as angry straight to a human.
5. Identify the agent and pass on the context. The agent introduces itself as an automated assistant, says what it can solve and hands the human a summary of the conversation. The design of what each agent reads and executes is in AI agents.
6. Treat the first failure as an alert. Log in the CRM the conversation the chatbot did not resolve and send that customer’s next contact straight to a person.
7. Measure behavior beyond satisfaction. Track use of the automated channel, transfers to humans, repeat contact within seven days on the same subject and satisfaction in transferred cases.

The role of HR and L&D

Agents now receive customers who already tried the bot and did not get a solution, with less patience. Training needs to teach agents to read the summary the bot handed over, to avoid asking again for what the customer already provided and to recover the conversation in the first minutes. Supervisors learn to review samples of the agent’s conversations and adjust the handoff rules. The effect of AI on the agent’s own productivity is covered in generative AI and productivity, and the format of the learning tracks by audience is in AI training for companies.

Customers accept AI that solves things fast and hands the conversation to a person when needed, and a company only delivers that with written rules and a team that knows how to work alongside the agent. That is the work of the in-company applied AI training, built on your company’s tickets and service process, with groups for leadership, service and CS. It delivers the prompt library by area and the agent architecture, with what each agent reads, what it executes on its own and who approves the rest, including the rules for handing off to a human.

Frequently asked questions

Do customers prefer to be served by a human or by AI?

In Gartner’s survey of 5,728 customers, 64% said they would prefer companies did not use AI in customer service, and the main concern was the difficulty of reaching a person. Preference changes with the type of request: in a 2025 survey in the United States, 54% trusted humans more for recommendations, and 32% trusted AI more.

Where does Gartner’s 64% figure come from?

From a Gartner press release dated July 9, 2024, based on a survey of 5,728 customers run in December 2023. The same release includes the 53% who would consider switching and the chart of concerns, answered by 5,140 people who knew what AI is.

Should I tell customers they are talking to a bot?

Yes, knowing that disclosure has a cost: in the experiment by Luo and colleagues, it reduced conversion in outbound sales. Gartner recommends telling customers what the chatbot can do and guaranteeing the handoff to a human, and a customer who later finds out they talked to a machine without knowing carries that distrust into the whole relationship.

When should a chatbot hand the conversation to a human?

When the customer asks, when the agent fails to understand the request after one or two attempts, when triage detects anger and when the subject involves a complaint, cancellation, disputed charge or purchase recommendation. Customers who already had a chatbot failure should also go straight to a person on their next contact.

Is there a rule in Brazil that requires human service?

For companies in services regulated by the federal government, Decree 11,034 of 2022 requires phone service with a human for at least eight hours a day and complaint and cancellation options in the first menu. For other companies, the decree is a reference for what consumers expect.

How do you prepare the service team to work with AI?

Train agents to receive customers coming from the bot, reading the conversation summary and without asking again for what the customer already provided. Train supervisors to review samples of the agent’s conversations and adjust the handoff rules, and measure the effect through repeat contact within seven days and satisfaction in transferred cases.