The State of Voice AI for Small Businesses 2026

Why we ran this study

Voice AI is getting cheaper and better at handling real phone calls.

The cost of running a voice agent has fallen by roughly three to five times in the past year. xAI launched its voice API in December 2025 at a flat rate well below the going price, Google released Gemini 3.1 Flash Live in March 2026 at a lower rate again, and OpenAI has cut its audio pricing repeatedly over the same period.

The models handle real conversation better too. They interrupt and recover, take turns without long pauses, and finish a task instead of just answering a question. OpenAI's Realtime API became generally available in August 2025 and can connect straight to the phone network, and xAI's latest voice model, released in July 2026, reasons while it is speaking rather than pausing first.

What nobody had measured is whether customers accept an AI on the other end, and which businesses actually need one. So we asked consumers and business owners separately, then compared their answers to what happens on real calls.

About this research. We ran two surveys in August 2026 through a third-party research panel, one of 270 US consumers and one of 262 US small and mid-sized business owners and managers. Consumer responses are weighted to US Census age, gender, and region, and business responses to the US Census Bureau's 2022 distribution of employer firms by company size. Alongside the surveys we analyzed 677,562 calls handled for 2,967 US businesses between January 2025 and August 2026. Full methodology appears at the end.

Unanswered calls move customers to competitors

We asked consumers what they did the last time they called a local business and no one answered. Only 24% left a voicemail. 34% tried again later, and 22% called a different business on the spot, without waiting for a callback. Small numbers emailed, texted, or gave up altogether.

Three quarters of callers did not leave a voicemail at all.

Across their lifetime, 78% of consumers say they have hired or bought from a competitor because a business did not answer or return their call.

Patience is limited. 22% of consumers said a single failed attempt is enough to send them elsewhere, and 65% said two or three. Fewer than one in seven will try more than three times.

The phone also remains the channel consumers choose when the stakes are highest. Asked how they would most want to handle an important matter with a local business, meaning a quote, a booking, or a problem, 50% chose a phone call. In-person came second at 21%. Email, text, web chat, and social media together accounted for 28%.

Hiring more people does not solve the missed-call problem

The businesses missing the most calls are the ones that employ people to answer them. Missed calls climb steadily with headcount and peak among businesses with 10 to 49 employees, where 91% miss at least one call in a typical week.

The climb is steep enough that it is not a quirk of the sample. A solo operator carries the phone and answers it, and 29% report missing any calls at all. A company above 50 employees usually staffs a front desk. Businesses in between take more calls than the owner can absorb and not enough to justify a full-time receptionist, and hiring has not closed the difference.

In that 10-to-49 band, 76% miss four or more calls a week and 79% have already lost a customer to a missed call. The work is worth more at that size too, with 81% saying a new customer is worth more than $500. Among solo operators the same figures are 14%, 26%, and 43%.

Across all businesses surveyed, 58% miss at least one potential-customer call in a typical week and 35% miss four or more. 49% say a new customer or job is worth $500 or more, and 23% put it above $2,000. Combining the two, 32% both miss calls in a typical week and value a new customer at $500 or more, and 42% miss calls and have already lost a customer because of one.

Among businesses that miss calls, the median new customer is worth $500 to $2,000, and the median business misses four to ten calls a week.

AI receptionists are helping businesses overcome language barriers

29% of business owners say a language barrier on the phone has already cost them a customer. Most small businesses cannot justify hiring for a second language, and the calls arrive regardless.

Our call records show how routine those calls are. Among the 1,102 businesses running an Upfirst receptionist with at least 50 calls, 90% took at least one call in Spanish. Across all 677,562 calls, 13.4% were in Spanish.

For most businesses the volume is occasional, which is exactly what makes it hard to staff for. Around a third take Spanish calls on 5% or more of their volume, one in seven on 10% or more, and one in fourteen on a quarter of their calls or more.

A business taking Spanish calls on 5% of its volume cannot justify hiring a bilingual receptionist, and cannot predict which hour of which day the call will come. Software handles the switch at no extra cost, which turns an occasional lost customer into a normal booking.

Callers are more receptive than owners expect

Business owners' most common hesitation about AI answering is a reaction most of their customers do not report having. Asked for their single biggest hesitation, 51% of owners said their callers would be annoyed or would prefer a person. Distrust of accuracy came a distant second at 17%, followed by cost at 12%.

The figure barely moves however we cut it. Weighted and unweighted, from solo operators to companies with 250 or more employees, and across all six industry categories, it stays between 48% and 54%.

Consumer comfort runs the other way. 43% of consumers are comfortable with a small business using an AI receptionist and 29% are not.

Adding the neutral group, 71% of consumers are either positive or indifferent.

Put the choice to them directly and half of consumers would rather an AI help them immediately than leave a voicemail, in an after-hours scenario where the business cannot answer live.

The 29% who are uncomfortable were specific about what would change their minds, and their answers are covered further down.

Businesses adopting AI answering already employ people to answer phones

The common assumption is that businesses adopt AI answering to avoid paying someone to pick up the phone. Our data points the other way. Solo operators, who have no receptionist to replace, are the least likely group to use an AI one at 2.4%. Businesses with 50 or more employees are the most likely, at 41.5%.

Businesses with staff adopt at 28 percentage points higher than solo operators. If the technology were substituting for front-desk labor, the pattern would run in the opposite direction.

The coverage data points the same way. Businesses using an AI receptionist are more likely to have human after-hours coverage than businesses that reject the idea, 79% against 48%, and far less likely to send after-hours calls to voicemail, 7% against 39%.

Adopters are also the businesses under the most phone pressure. 72% miss four or more calls a week, against 24% of the not-interested. 71% spend six or more hours a week answering, returning, and chasing calls, against 19%. 85% have lost a customer to a missed call, against 32%.

Cost is the leading hesitation for only 12% of owners, which is not the profile of a market buying on labor savings.

The call records fit the same picture. 16.4% of calls include an attempt to hand the caller to a human, so the AI is routing people to staff rather than standing in for them.

Interest in AI answering more than doubles once a business starts missing calls

Interest concentrates among businesses with a missed-call problem and rises steeply with the size of it, from 32% among businesses missing no calls to 77% among those missing four or more a week.

Among all businesses that miss calls, 69% are open to an AI receptionist and 25% already use one. Among those missing four or more calls a week, 77% are open and 28% already use one.

Resistance sits almost entirely with businesses that do not have the problem. 71% of solo operators report missing no calls in a typical week, which is what you would expect from one person carrying a phone.

Two further findings come from smaller groups of respondents and should be read as directional. Of the 80 owners who have used an AI receptionist, 68 still use one, a retention rate of 85%. And not one of those 68 current users picked "I don't trust it to get things right" as their biggest hesitation, against 17% of everyone else. Worry about accuracy appears to come from not having used one.

Businesses with smaller jobs are adopting at the same rate as businesses with large ones

A business whose typical new customer is worth under $100 is about as likely to use an AI receptionist as one whose typical customer is worth more than $10,000. Adoption sits between 21% and 29% across every value band, with no meaningful relationship between the two.

40% of businesses currently using an AI receptionist have a typical customer worth under $500, and 9% have one worth under $100.

Smaller jobs do not mean fewer missed calls. 61% of businesses with a typical customer worth under $500 miss at least one call in a typical week. Among that group, 39% already use an AI receptionist and 76% are open to one, both above the rate for businesses overall.

An answering service priced for a law firm was hard to justify for a business missing a handful of calls worth a couple of hundred dollars each. AI answering costs a fraction of that, and the flat line above is what you would expect once price stops deciding who can buy. This reads one survey rather than a trend, and we have no comparable data from earlier years to test it against.

Adoption stays low in one place, at 2.4% among solo operators, where the constraint looks like need rather than price.

Receptiveness is higher among higher-income households

Household income predicts comfort with AI receptionists more strongly than age does. 22% of consumers in households under $75,000 are comfortable with a small business using one, against 57% between $75,000 and $150,000 and 77% above $150,000.

Testing age, income, and gender together, income was by far the strongest predictor. Age mattered less, and gender did not matter at all. The income pattern also holds among younger and older consumers alike, so this is not simply older people having lower incomes.

Businesses serving higher-income customers can expect stronger acceptance. Businesses serving lower-income customers will have more callers in the skeptical group, and the three things that matter most to those callers are in the next section: saying up front that they have reached an AI, keeping the route to a human short, and limiting the AI to the tasks callers accept.

Disclosure builds more trust than voice realism

We asked consumers which factors affect whether they trust a small business's AI receptionist, then looked separately at consumers who are neutral or uncomfortable about AI. Among that skeptical group, being told it is an AI is the highest-ranked factor at 55%, and sounding like a natural human voice is the lowest at 19%. Completing the request and knowing the business rank first and second in both groups.

Asked directly, 80% of consumers said they would want an AI to tell them it is an AI at the start of the call. Only 13% said they do not need to be told.

Caller behavior points the same way. In earlier Upfirst research on 450,702 calls across 503 AI receptionists, receptionists that told callers they were an AI saw roughly 20% fewer hang-ups, counting a hang-up as a call under 15 seconds or one where the caller never spoke. Callers appear to drop off when they cannot tell what has answered. That analysis is observational, so it shows an association rather than a cause.

Consumers accept a specific set of tasks

Acceptance is highest for the tasks that fail when a call goes unanswered and lowest for tasks involving money, negotiation, or dissatisfaction. 70% of consumers would let an AI answer questions about hours, services, and pricing, and 62% would let it take a message. Booking or changing an appointment clears 45%. Taking a payment and handling a complaint sit at 16% and 15%.

Among consumers who described themselves as uncomfortable with AI receptionists, 67% were still comfortable letting an AI take a message, a higher share than among comfortable consumers. The job callers most readily hand to an AI is the one voicemail does worst.

What the call data shows

Across 677,562 calls handled for 2,967 US businesses between January 2025 and August 2026, 23.1% arrived outside 8am to 6pm, Monday to Friday, in the business's local time zone.

After-hours calls were more likely to be urgent than business-hours calls. 4.3% referenced an emergency against 2.5% during business hours, and after-hours calls were slightly more likely to concern a booking or a quote. Calls outside business hours are not worth less than calls during them.

The median conversation lasted 69 seconds, and 29% ended with the caller's contact details captured.

These figures describe calls routed to an answering service. Businesses forward calls to us when their own line is busy or unanswered, and some forward only outside opening hours, so the sample leans toward calls the business could not take itself. It cannot be used to calculate how often businesses miss calls, which would need the count of calls each business answered on its own.

Recommendations

Wait until you are actually missing calls. Only a third of businesses that miss no calls have any interest in an AI receptionist, and they are right not to. Missed-call rates start climbing sharply somewhere between five and ten employees.

Treat it as extra capacity, not a replacement. The businesses getting the most from AI answering already have people on the phones and use it for the calls those people cannot reach.

Tell callers they are talking to an AI. 80% of consumers want to be told, it is the single most important trust factor for the callers most likely to react badly, and it is associated with fewer hang-ups.

Do not try to pass your AI off as a person. Voice quality matters less than most people assume. Skeptical callers rank it last of the five factors we tested, well behind knowing the business and finishing the request.

Start with questions, messages, and bookings. Each clears 45% consumer comfort. Fewer than one in five consumers want an AI taking payments or handling complaints, so send those to a person.

Check that the service handles other languages. Nine in ten businesses with real call volume take Spanish calls, and 29% of owners have already lost a customer to a language barrier.

Think about who your customers are. Comfort with AI receptionists runs 22% among consumers in households under $75,000 against 77% above $150,000. If your customers are at the lower end, disclosure and an easy route to a human matter more.

Methodology

Limitations. Both surveys ran on an opt-in online panel rather than a probability sample, so neither is nationally representative and we do not describe them that way. The consumer panel skews affluent, with 51% of respondents reporting household income above $100,000 against roughly 37% of US households, and because income turned out to be the strongest predictor of comfort with AI receptionists, our comfort figures are more likely too high than too low. Comparisons between businesses that use an AI receptionist and those that do not describe who is buying, not what adoption causes.

Surveys. Two surveys fielded through a third-party research panel on 20 and 21 August 2026: 270 US consumers, and 265 US small and mid-sized business owners and managers screened for small-to-medium business ownership or management. Three business respondents were excluded because their open-text description indicated they do not operate a business, leaving 262. Ten questions each.

Consumer responses are weighted to US Census age, gender, and region. Business responses are weighted to the US Census Bureau's Statistics of U.S. Businesses 2022 distribution of employer firms by company size. That weighting matters here: our raw sample over-represents larger firms, with businesses under five employees making up 32.8% of respondents against 63.0% of US employer firms, and unweighted figures run materially higher on every measure of missed calls and phone workload. Figures broken out by company size, household income, or job value are unweighted within each band.

Margins of error are about 7 points on consumer figures and 8 points on business figures, wider within subgroups. Comparisons narrower than that are identified as ties or left out.

Call data. 677,562 completed inbound calls handled for 2,967 US businesses between 9 January 2025 and 31 August 2026, excluding test calls, spam, and blocked numbers. After-hours means outside 08:00 to 18:00 Monday to Friday in each business's local time zone. Language figures use the 1,102 businesses with at least 50 calls and report Spanish specifically, because the wider non-English figure reflects the language mix of our own customer base.

We plan to field this survey again in 2027, so the next edition can measure change rather than a single point in time.