Picture a 74-year-old homeowner with a running toilet. She calls the plumber her neighbor recommended, hears a voice say it's the company's automated assistant, and nearly hangs up. Then it says: “You can ask for a person at any time. What's going on with your plumbing?”
She tells it. It asks whether water is leaking onto the floor, reads back her address, offers Thursday morning, and texts her a confirmation before she's put the phone down. Whether she'd describe that call as “talking to a robot” or “booking the plumber” depends almost entirely on choices the business made before she dialed.
What the research actually says
Opinions about AI on the phone are everywhere. Verified research is thinner, and it points one way:
- In a Gartner survey of 5,728 customers conducted in December 2023, 64% said they would prefer companies didn't use AI for customer service. The top concern was that AI would make it harder to reach a person.
- In a later Gartner survey of 3,566 B2B and B2C customers, run in February and March 2026, 50% said their interactions are easier when companies use generative AI, and 87% said it's essential for companies using it to offer a way to reach a human agent. Among customers unwilling to use AI, the most common thing that would change their minds was being able to switch to a human.
- In a June 2023 report on chatbots in banking, the Consumer Financial Protection Bureau described customers stuck in “doom loops” of repetitive, unhelpful answers with no way to reach a human representative.
These sources cover customer service broadly, not only phone calls, but the lesson carries over. In the survey fielded in December 2023, most respondents preferred that companies not use AI in customer service, and their top worry was losing access to a person; the 2026 survey found customers split evenly on whether AI makes things easier. You can't change that preference, but you can design against that worry, and that part is in your hands.
Seven design choices that decide how callers feel
- 01Say what it is, first. “I'm [Business]'s automated assistant” belongs in the opening sentence. Callers who work it out later feel tricked, and some states now require disclosure in certain situations, covered in the post on AI phone answering laws. The details differ by state; treat this as background, not legal advice.
- 02Offer the exit before anyone needs it. “You can ask for a person at any time” in the greeting, then honor it on the first ask, with no attempt to talk the caller out of it.
- 03Keep it short. One question at a time, no menu of options read aloud, no small talk. A booking should take less time than holding for the front desk would.
- 04Give up gracefully. After two misunderstandings, the assistant should stop and say so: “I'm not getting this right. I'll have someone call you back within the hour.” This is the anti-doom-loop rule.
- 05Confirm details back. Read back the name, date, time and address before booking. It's the moment callers decide whether to trust the result.
- 06Don't pretend to be human. No invented first names, fake keyboard sounds or “let me check with my manager” theater. A natural voice is fine; a fake persona isn't.
- 07Follow up in writing. A text within a minute that summarizes what was agreed gives the caller something to check, correct or forward to a spouse.
Flowchart
The comfort path on one call
- StartGreeting: automated assistant, call recorded, “ask for a person at any time”
- DecisionCaller asks for a person?YesA personTransferred on the first ask, with no attempt to keep themNo
- StepOne question at a time, in plain language
- DecisionHas the assistant misunderstood twice?YesA personSays so plainly; a person calls back within a set timeNo
- StepReads back name, time and address before booking
- DoneA text summary arrives within a minute
Before and after
A call that feels like a trap vs. one that feels like service
Feels like a trap
- 01Introduces itself as “Ashley from the front desk”
- 02Reads a five-item menu before asking anything
- 03Loops on “Sorry, I didn't catch that”
- 04Ignores “representative” and asks again
- 05Ends without confirming anything
ResultCaller hangs up annoyed and calls someone else
Feels like service
- 01Says it's automated and offers a person up front
- 02Asks “What can I help with?” and listens
- 03After two misses, offers a callback from a person
- 04Transfers on the first request
- 05Reads back the details and texts a summary
ResultA booked appointment the caller trusts
Different callers, different comfort
Who calls you changes the settings, not the principles.
- Restaurants: calls are quick and transactional. Answer after a few rings during service, get to the booking or the answer fast, and keep most calls under a minute.
- Dental practices: many callers are existing patients who know the front desk by name. Overflow-only answering keeps the familiar voice first, and the assistant catches what would have gone to voicemail.
- Plumbers and other trades: callers often have a problem happening right now. Lead with the question that matters (“Is water still leaking?”) and get to a booking or a page quickly.
- Older callers: slow the pace, skip jargon, repeat key details and make the human option especially easy. Don't assume they'll refuse; assume they're checking whether they're being heard.
Calls that should skip the assistant
Comfort also comes from knowing when not to answer with automation at all. Many business phone systems can route by caller ID or time of day, so a few groups can ring straight through to a person:
- Key accounts, such as the property manager who sends you twenty jobs a year, or a restaurant's regular catering clients.
- Anyone you promised a personal callback, for the next day or so.
- Patients or clients in the middle of something sensitive, like a treatment plan or an active complaint.
- Calls during the hours your team is fully staffed and the phone is quiet. Overflow-only answering exists for a reason.
Tell your regulars before you switch
Surprise is what turns a decent automated call into a complaint. A week before launch, add a line to your website and booking confirmations, mention it at checkout or on job sites, and send regulars a short note: “From Monday, an automated assistant will answer when we can't, including evenings. It can book and reschedule, and you can always ask for one of us.” Callers who know what's coming are far less likely to feel ambushed by it.
Measure comfort instead of guessing
You don't need to commission a survey to know whether callers are comfortable. Your call records already hold the signals:
- Hang-ups in the first 15 seconds. A rising count suggests the greeting is too long or off-putting.
- Requests for a person, by call type. A cluster shows where the assistant isn't trusted or isn't capable yet.
- Repeat calls within a day. A caller who rings back usually didn't get what they needed the first time.
- What customers tell your staff. For the first month, ask the team to jot down every comment about the phone, good or bad.
Then run a comfort audit: call your own line and score it against the seven choices above. Each “no” is a script change, not a reason to abandon the idea. The call script template shows how to write these choices into rules, and it's the same checklist every AI receptionist we build is tested against.
Sources
- Gartner: Survey Finds 64% of Customers Would Prefer That Companies Didn't Use AI for Customer Service (July 9, 2024)
- Gartner: Survey Finds 87% of Customers Say Companies Using GenAI for Customer Service Must Provide Access to a Human Agent (Aug. 4, 2026)
- Consumer Financial Protection Bureau: Chatbots in consumer finance (June 6, 2023)