By Sagar Shankaran, Founder of CallSphere
Live speech-to-speech translation in 70+ languages moves language access out of the exam room and onto the reminder, prep, refill and billing calls around it.
Key takeaways
How many of your patients get a worse version of your practice than the rest, purely because of what language they speak at home?
Not the exam. The exam usually gets an interpreter, because you know the rules and because the physician is in the room and will not proceed without one. It is everything around the exam that quietly degrades. The reminder call the day before. The colonoscopy prep instructions. The refill request. The lab result that is normal and needs one sentence of explanation. The billing question about a January deductible. Those interactions get handled by a front desk that speaks English, a phone hold, a bilingual medical assistant pulled off rooming patients, or a family member who happens to be home — and sometimes they simply do not get handled at all.
Here is the sequence in a practice with a Haitian Creole and Spanish-speaking panel. The patient calls at 9:15. The scheduler hears a language she does not speak, says "one moment," and dials the telephonic interpreter line. She enters the account number, picks the language, and waits. Spanish comes up in under a minute. Haitian Creole, Vietnamese, Nepali, Mam, Karen, Somali — those can take four to nine minutes, and sometimes the line comes back with nobody available.
Meanwhile the patient is on hold, the other four lines are ringing, and the scheduler has two people at the check-in window. So a rational front desk starts making rational decisions: for anything short — confirming an appointment, taking a refill request, saying the results are normal — skip the interpreter and get through it with gestures in words. That shortcut is where the harm lives, and it does not show up in any log you keep.
Live translation on a practice phone line means the caller speaks their language, your scheduler speaks English, and each hears the other in their own language within about a second, with no third party dialed in and no hold.
Two things, both in 2026. Gemini 3.5 Live Translate detects and translates across more than 70 languages, speech to speech, close enough to real time that a conversation keeps its rhythm. Gemini 3.1 Flash Live, which arrived in March, handles audio to audio across more than 90 languages and is good at doing something in the middle of the conversation — checking a schedule, writing a request into a queue — rather than just talking.
Hear it before you finish reading
Talk to a live CallSphere AI voice agent for healthcare in your browser — 60 seconds, no signup.
The distinction from 2024 matters. Older translation was a scheduled service: you booked it, or you dialed it, or you typed something in and read a translation back. The turn-taking was awkward enough that staff avoided it for short interactions, which is exactly where volume lives. What changed is that translation stopped being a thing you arrange and became a property of the call itself. Nobody waits. Nobody dials. It is simply on.
flowchart TD
A["Patient calls, speaks Haitian Creole"] --> B["Line detects language, translation on"]
B --> C{"What is the call about?"}
C -->|Appointment or reminder| D["Book or confirm in the schedule"]
C -->|Refill or forms| E["Request into the nurse queue, both languages logged"]
C -->|Prep or after-visit instructions| F["Read back instructions, patient repeats them"]
C -->|Symptoms or bad news| G["Warm transfer to a qualified human interpreter"]
D --> H["Confirmation text sent in the patient's language"]
E --> H
F --> H
Appointment reminders and confirmations. A reminder call that the patient does not understand is a reminder that did not happen. If your no-show rate for limited-English-proficiency patients runs meaningfully above your panel average — pull the report, most practices have never split it that way — this is where you will see the change first, and it shows up in filled slots within a month.
Procedure preparation. Bowel prep for a screening colonoscopy is the sharpest example in primary care. A prep failure means an aborted or repeated procedure, a wasted referral, a patient who is now scared of the whole thing, and a colorectal screening measure that stays open for another year. The prep instruction call is five minutes and it is the single highest-value translated conversation in the practice. Fasting labs and pre-visit medication holds are the same shape.
Refill and forms requests. These are short, high-volume, and low-risk, and they are precisely the calls a busy front desk currently improvises through. Getting the medication name, strength, and pharmacy correct on the first call — rather than through three rounds of callbacks — is worth more staff time than owners expect.
Billing and coverage questions, especially in January when deductibles reset and every patient who thought a visit was free is calling about a statement. That conversation is confusing in your first language. In your second, with no interpreter, it produces an unpaid balance and a patient who stops coming.
Stated assumptions, an illustration only — put your own vendor invoice and your own no-show report in place of these. Suppose 22% of your panel is more comfortable in a language other than English, you currently place about 70 interpreter-assisted phone interactions a month averaging nine minutes, and your telephonic vendor bills $1.35 a minute.
| Line | Today | After |
|---|---|---|
| Interpreter phone minutes per month | 630 | 210 |
| Vendor cost at $1.35/minute | $851 | $284 |
| Front-desk hold and hand-off time per month | about 9 hours | about 2 hours |
| Staff cost of that time at $23/hour | $207 | $46 |
| Translated calls the line now handles | — | about 300/month |
| Cost of translated call handling | — | about $120 |
| Monthly running total | $1,058 | $450 |
Now the part that dwarfs it. If those 300 newly-workable calls include reminder and prep calls that convert four additional kept appointments a month, and your average visit reimburses $118, that is $472 a month in slots that were previously no-shows — about $5,700 a year, before you count a single colonoscopy prep that did not have to be repeated. Note that the 300 calls are not calls you were paying for before. They are calls that were being handled badly for free.
This is not a replacement for interpreter services in the exam room, and any vendor telling an owner otherwise is selling them a compliance problem. Section 1557 language-access obligations, your payer contracts, and plain clinical judgment all point the same direction on the encounters that matter most.
Still reading? Stop comparing — try CallSphere live.
See the healthcare AI agent handle a real call — complete, industry-specific, and live in your browser. No signup.
Keep a qualified human interpreter for: informed consent of any kind; delivering a serious diagnosis or a result that changes a life; any behavioral health or substance use conversation; anything involving a suspected safety issue at home; end-of-life and advance directive discussions; and any conversation where the patient asks for a human. Also keep one for languages with limited coverage and for patients whose speech is affected by hearing loss, stroke, or advanced dementia, where recognition quality drops sharply and confidently.
Two operational rules worth writing down. First, the patient is told at the start of the call that translation is being handled by a machine and can ask for a person at any point — that sentence, in their language, costs you nothing and settles most objections. Second, never use a minor child as an interpreter, which was already true and does not become less true because a faster option now exists.
Treat it as improving access, not as satisfying a requirement. The obligation is about meaningful access with qualified interpreters for the encounters that require them, and a machine on a scheduling call does not change what the exam room needs. What it does change is the enormous volume of interactions that were previously getting nothing at all — and going from nothing to good is the real gain, legally and clinically.
Coverage in 2026 runs past 70 languages for live speech-to-speech and past 90 for audio conversation, which comfortably covers Spanish, Haitian Creole, Vietnamese, Arabic, Mandarin, Russian, Portuguese, Somali and most of what a US primary care panel produces. Quality is not uniform. Test the three languages your own panel actually speaks, with your own staff listening, before you trust any list.
This is the real risk and the reason for the read-back rule. Any call that carries a medication, a dose, a date, or a time ends with the patient repeating it back, and the staff member confirms it against the screen. That is standard practice for English calls in a good office too; here it is not optional.
Some will, and they should be able to say so and get a person. In practice the comparison is not machine versus a warm bilingual staff member — it is machine versus nine minutes of hold music followed by a stranger on a speakerphone. Most patients rate the first one higher.
Split one report: no-show rate and same-day cancellation rate by preferred language. Most practice management systems store preferred language and almost no practice reports on it. If the gap is under two points, this is a service-quality project and you should still do it. If the gap is eight points, you have found a revenue problem that has been sitting in your schedule for years wearing a different name.
Since almost all of this happens on the phone: CallSphere builds AI voice and chat agents that answer practice phone lines and web chat, book appointments, and capture requests around the clock — including handling the caller in their own language rather than putting them on hold. The exam room still needs a qualified interpreter. The 300 calls a month around it do not have to go unanswered.
Written by
Sagar Shankaran· Founder, CallSphere
LinkedInSagar Shankaran is the founder of CallSphere, where he builds production AI voice and chat agents deployed across healthcare, hospitality, real estate, and home services. He writes about agentic AI, LLM engineering, and shipping voice agents that handle real calls in production.
See how AI voice agents work for your industry. Live demo available -- no signup required.
How rural primary care clinics in Crescent City, California can answer every patient call 24/7 with AI — after-hours triage, booking, and referral follow-up.
Estes Park, Colorado family clinics face summer call surges and quiet winters. An AI receptionist scales with Rocky Mountain National Park crowds year-round.
Harrisburg, Pennsylvania family practices miss commuter calls at 8 a.m. and 5 p.m. An AI receptionist answers around the clock and books every state worker.
Vietnamese, Spanish and Mandarin-speaking dental front offices order the minimum and ask nothing. Live translation in 2026 changes the lunch-hour call.
Why non-English statement calls age into bad debt, and how 2026 live translation keeps the patient on the line long enough to take a card or set a plan.
Doxycycline before IPL, fish oil before filler, blanching after lip work. In 2026 an assistant answers only from your approved sheets and names the source.
© 2026 CallSphere Inc. All rights reserved.
Made within San Francisco
Watch how CallSphere handles real customer calls, schedules appointments, and processes payments — live.
Try Live DemoBook a DemoCalculate Your ROI