One survey says 35% of Americans prefer to reach a company by phone. Another, of more than 27,000 consumers, says 87% rank the phone as their preferred channel. Both were published within about a year of each other. Both are real. If you are deciding where to put your support effort next year, that gap in customer service channel preferences data tells you more than either number does.
We read a lot of customer service channel preferences research while building ChatterMate, because the first question a new team asks us is "which channels should we turn on?" Our honest answer used to be "the ones your customers use." That turns out to be a weaker answer than it sounds. What people say they prefer, what they actually use, and what they want for a specific problem are three different things. This post walks through the data we trust, where it contradicts itself, and a way to choose channels that doesn't depend on picking the right survey.
Written by the ChatterMate team. We build an open-source AI support agent that answers on nine channels from one inbox.
What the 2025 and 2026 surveys actually found
Start with the cleanest dataset we found. In March 2025 YouGov published US figures from its Profiles panel, which is weighted to be nationally representative of the online population. It asked two separate things: which channels people use to contact businesses, and which one they prefer.
| Channel | Use it | Prefer it |
|---|---|---|
| Phone | nearly 70% | 35% |
| 63% | 23% | |
| In person | 35% | 8% |
| Website forms | 34% | 5% |
| Live chat | 31% | 10% |
| Chatbots | 18% | 1% |
| Social media | not given | 4% |
| Mobile apps | not given | 2% |
Source: YouGov, How Americans prefer to contact businesses for customer service, March 2025.
Look at the chatbot row. 18% use them. 1% prefer them. Proportionally, that is the worst ratio of use to preference on the whole table, and it is the row that anyone selling AI support (us included) should stare at for a while.
The age split matters too. YouGov found 52% of Baby Boomers prefer the phone, against 30% of Millennials and 25% of Gen Z. Gen Z leans more on email (28%) and social media (8%) than older groups. So "customers prefer phone" is mostly a statement about who your customers are.
Now the second dataset. In April 2026, CX Dive reported on a ServiceNow survey of more than 27,000 consumers. It found that 87% rank the phone as their preferred channel for customer service. In the same survey, three-quarters said they would rather try self-service before calling, about a third prefer dealing with a human over a machine, and nearly half said modern chatbots fail to understand their questions. Half named lack of empathy as their top frustration with the technology.
Then there is the business side. Gartner surveyed 265 customer service leaders in April and May 2025 and found they expect self-service and live chat to overtake phone and email as their most valuable service technologies by 2027.
So customers say phone. Leaders say chat and self-service. And the same customers who say phone also say they want to try self-service first.
Why the numbers disagree
None of these surveys is wrong. They are answering different questions, and the question shapes the answer far more than most headlines admit.
"Prefer" versus "rank as preferred." YouGov asked people to pick one favourite. The ServiceNow figure, as reported, is people ranking the phone as preferred, and we couldn't find the exact question wording in the public coverage. A ranking question lets a channel be someone's favourite for some situations without being their only answer. A single-choice question forces a trade-off. Those produce very different percentages, so we would not compare the 35% and the 87% as if they measured the same thing. They don't.
Preference in general versus preference for a task. Ask "how do you like to contact companies?" and people picture the problem that annoyed them most recently. That is usually something complicated: a billing dispute, a broken delivery, a cancellation that went wrong. For those, a human voice feels safest. Ask about checking an order status and the answer flips. The ServiceNow coverage makes this explicit: self-service is preferred for simple tasks like purchases, payments and loyalty offers.
Customers versus leaders. Gartner's 265 respondents run support teams. They are answering "which technology will be most valuable to my organisation," which folds in cost, staffing and scale. A channel can be valuable to the business and still be nobody's favourite. Email has lived in that spot for twenty years.
Who answered. The YouGov age gap shows how much the sample matters. A survey weighted towards older respondents will lean towards phone. A panel of app-native shoppers will lean the other way. Your customer base is its own sample, and it probably doesn't match any national average.
Here is the short version, laid out:
| Survey | Who | What was asked | Headline |
|---|---|---|---|
| YouGov, Mar 2025 | US online adults | Use vs single preferred channel | Phone 35% preferred; chatbots 1% |
| ServiceNow via CX Dive, Apr 2026 | 27,000+ consumers | Ranked channel preference, attitudes to AI | 87% rank phone preferred; 75% want self-service first |
| Gartner, Aug 2025 | 265 service leaders | Most valuable service technologies | Self-service and live chat to pass phone and email by 2027 |
If someone quotes you one of these numbers to justify a channel decision, ask which row it came from.
The finding we take most seriously: try first, then escalate
Strip away the headline percentages and one pattern holds across the data. People want to sort out simple things themselves, fast, and they want a person when it gets hard or emotional.
That is not a contradiction. It is a sequence.
Three-quarters of ServiceNow's respondents want to try self-service before calling. A third prefer a human outright. Half say chatbots lack empathy. Put those together and you get a customer who will happily type "where's my order" into a chat box, but who will be furious if the same box traps them when they're trying to dispute a charge.
And YouGov's chatbot row starts to make sense. 18% use chatbots and 1% prefer them. We'd guess a lot of that 17-point gap is people who used a bot because it was the only door open, and found it couldn't do what they needed. We wrote about that pattern in why customers get frustrated with AI customer service. The bot wasn't the problem on its own. The missing exit was.
Gartner has a related figure that CX Dive cites in its coverage of the leaders' survey: nearly 9 in 10 customer journeys that begin in self-service end up being resolved through multiple channels. Most people don't pick one channel. They start in one and finish in another. So the real design question isn't "which channel do customers prefer?" It is "what happens when they move between channels, and do they have to start over?"
A better question than "which channel do customers prefer"
Here is the framework we use with teams setting up ChatterMate. Instead of picking channels from a survey, sort your incoming contacts by two things: how complex the problem is, and how upset the person is likely to be. Then choose a channel for each quadrant.
| Low emotion | High emotion | |
|---|---|---|
| Simple (order status, opening hours, password reset, "do you ship to X?") | Self-service or AI chat, any channel the customer is already on | AI chat that can answer fast and hand off the second the tone shifts |
| Complex (setup problems, configuration, multi-step troubleshooting) | AI chat grounded in your docs, with a human one click away; email for anything that needs back-and-forth over days | A person, on the customer's channel of choice, with full history in front of them |
Two things fall out of this.
First, channel and handler are separate decisions. "WhatsApp" doesn't tell you whether a bot or a person answers. A billing complaint can arrive on WhatsApp and still need a human. A shipping question can arrive by phone and be better answered by a link to the tracking page sent by text.
Second, the top-right box is the one most teams get wrong. Simple problem, angry customer. "I've been charged twice." It is easy to answer, so it gets routed to the bot, but the customer is in no mood to be processed. That is where a clean human handover earns its keep: the AI recognises the frustration, passes the chat over with everything attached, and the person who picks it up doesn't open with "how can I help you today?"
Worked example: choosing channels for a 1,500-contact month
Numbers make this concrete. These are illustrative figures, not from a real customer, but the shape is typical of what we see from online retailers.
Say a store gets 1,500 support contacts a month. You tag a month of them and find:
- 600 are order status or delivery questions.
- 300 are pre-sale product questions ("does this fit a 2019 model?").
- 250 are returns and exchanges.
- 200 are account and payment problems.
- 150 are complaints, damaged goods or disputes.
Now apply the quadrants.
Order status and pre-sale questions (900 contacts, 60%) are simple. Most are low emotion. These go to an AI agent on whatever channel the customer is already using: the website widget, WhatsApp, Instagram DMs. The point is to meet the "try it myself first" instinct where it happens. If your docs and order data are good, a large share of these can be answered without a person, and you should measure that share honestly rather than assume it. Our post on reducing first response time covers how to check whether those fast answers are actually resolving anything.
Returns (250) are moderately complex and sometimes tense. The AI can explain the policy and collect the order number, then hand over when an exception is needed.
Account, payment and dispute contacts (350, about 23%) are where the phone-preferring customer lives. Many will want a person. Some genuinely need a call. If you don't run a phone line, be clear about it and make the human route in chat obvious and quick. If you do, the chat should be able to offer a callback rather than pretending it can fix everything.
What you've done is turn a survey question ("do our customers prefer phone or chat?") into an operational one ("for which 23% of contacts do we need a human fast, and are we staffed for it?"). That's answerable from your own data.
What this means if you add AI to your channels
We build an AI support agent, so here is our honest read of what the data says about AI specifically.
The 1% figure should humble anyone in this market. Customers do not wake up wanting to talk to a chatbot. They want the problem gone. An AI agent wins when it is the fastest way to make that happen, and loses the moment it becomes an obstacle between the customer and a person.
That shapes three choices we'd recommend regardless of which tool you use.
Put the AI on the channels customers already use. Don't make people come to your website widget if they message you on WhatsApp. The try-it-first instinct works best when there's nothing new to learn. ChatterMate connects one agent to nine channels, including the website, WhatsApp, Instagram, Messenger, Telegram, LINE, Slack, SMS and email, with replies from the AI and the team landing in one shared inbox. We'll be straight about a limit: on messaging channels the core works (grounded answers, handover, the shared inbox), but richer features such as workflows, in-app actions and end-of-chat ratings currently run only in the website widget.
Ground answers in your docs and admit gaps. "Chatbots fail to understand my question" was cited by nearly half of ServiceNow's respondents. A bot that answers only from your help content and says "I don't know, let me get someone" when it can't find an answer is far less irritating than one that guesses. We've written about why grounding matters more than model choice.
Make the exit visible. The "about a third prefer a human" crowd isn't going away. Give them a way out in the first message, not after three failed attempts. That costs you a few deflections. It saves you the customer.
How to measure customer service channel preferences in your own data
National surveys are a starting point. Your own data beats them. Four checks we'd run before changing anything:
Channel by contact reason. Tag a month of contacts by reason and by channel. If complaints cluster on email and order questions cluster on chat, your customers are already sorting themselves. Build around that.
Channel switching. Count how often the same customer contacts you on a second channel within 48 hours about the same issue. That is the multi-channel journey Gartner describes, and every switch usually means the first channel failed.
Handover timing. In AI chats, look at how many messages pass before a customer asks for a person. If it's often the first or second message, they've learned not to trust bots, and your opener needs to show it can actually help.
Who is asking. If your customers skew older, weight the YouGov age data. A B2B audience of developers will behave nothing like a retail audience of first-time buyers.
None of this needs a research budget. A spreadsheet and a month of tagged conversations will tell you more about your customer service channel preferences than any headline figure in this post.
So, phone or chat?
Both, probably, but not for the same things.
The surveys agree on more than their headlines suggest. People reach for the phone when the problem is serious and they want a human who can make a decision. They reach for self-service and chat when the problem is simple and they want it done without waiting on hold. The disagreement between 35% and 87% is mostly about how the question was asked.
So stop asking which channel your customers prefer. Ask which problems need a person, make sure those reach one quickly on whatever channel they arrive, and let a well-grounded AI handle the rest where your customers already are.
If you want to try that setup, ChatterMate is open source and free to start: the Free plan includes 50 messages a month, new accounts get 14 days of the Growth plan with no card, and you can self-host it under Apache-2.0 if you'd rather run it yourself.



