NOTES FROM THE TEAM
Multilingual answering service for healthcare: what to verify before launch
We piloted Pro multilingual at a community clinic in San Diego last fall. The practice manager thought maybe 20% of her calls were Spanish-speaking. We routed Spanish callers to AI multilingual answering for two weeks. Her actual Spanish call volume turned out to be 38%. Her booking rate from those calls went up by close to 40%. Same patients, same offer — they just couldn’t navigate voicemail in English.
The US has roughly 25 million LEP (limited English proficiency) residents. For most US medical practices at least 5% of patients speak something other than English at home — often a lot more. Practices that ignore this leak appointment bookings, refill requests, and after-hours triage to whichever competitor answers in the patient’s language.
What follows is the practical checklist I’d run through if I were buying a multilingual answering service for a medical, dental, or veterinary practice in 2026. Most of what’s on the marketing pages is true. Some of it isn’t. Here’s how to tell.
Related: DeskMD pricing and Physician answering service.
Step 1: Survey your actual patient language mix
Before evaluating any vendor, pull a year of intake forms and run a quick frequency count of preferred-language entries. Most practices are surprised. The dominant non-English language is often not Spanish (it is for many regions, but plenty of practices it’s Vietnamese, Mandarin, Korean, Russian, Tagalog, Arabic, Haitian Creole, or Punjabi).
If your state doesn’t have intake-language fields, look at zip-code-level Census ACS data for ‘language spoken at home’ in your service area. The ACS table B16001 gives a good directional answer.
Match your top 3 to 5 non-English languages against the vendor’s supported list. If the dominant language in your area isn’t on the vendor’s native-quality list, that vendor’s the wrong fit.
Step 2: Distinguish ‘supported’ from ‘native quality’
A lot of vendors claim ‘100+ languages supported’ because their underlying voice model has some training data in each. The honest distinction is between languages where the model holds a natural conversation (native quality) and languages where it understands but stumbles (best-effort).
DeskMD Pro supports 20+ languages at native quality with additional best-effort coverage. We deliberately don’t use the ‘100+’ marketing number because most of those languages are best-effort, not production-grade.
Ask any vendor for a sample call recording in your target language. If they can’t produce one in 24 hours, that language is best-effort regardless of what the website says.
Step 3: Verify English translation in the intake card
A multilingual answering service that captures the call in the patient’s language is only half the win. The morning queue still needs to be triageable by staff who don’t necessarily speak that language.
Confirm that the intake card includes the English translation alongside the original-language transcript. The team should be able to triage the queue in English; the original is preserved for accuracy and audit.
For practices with bilingual front-desk staff, also confirm that the original-language transcript is exposed (not just the translation). Bilingual staff often catch nuances that translation drops — and patients can tell when those nuances are missing.
Step 4: Test emergency-language flagging across languages
Urgency flagging is harder in non-English languages. The phrases for ‘I’m having chest pain’ or ‘my child can’t breathe’ don’t translate word-for-word into the model’s English emergency-language list.
Verify with the vendor that urgency triage is tuned per supported language. Run a few test calls with translated emergency phrases. Confirm the calls land in the urgent queue with the same dispatch behavior.
A multilingual service that flags emergencies in English but misses them in Spanish isn’t a multilingual service for healthcare. It’s an English service with a translation layer, and the translation layer’s worst gap is exactly the moment that matters most.
Step 5: Confirm HIPAA + BAA coverage in all languages
Some vendors use a different translation provider as a subprocessor. That provider needs its own BAA. Ask whether translation is in-model (same LLM provider) or routed through an external translation service. If external, request the subprocessor’s BAA status.
PHI redaction has to work in non-English languages too. Confirm the vendor’s redaction tooling can identify medications, conditions, and identifiers in the supported languages, not just English. Most can’t, by the way. They redact based on English regex patterns.
DeskMD multilingual answering uses the same voice model that handles English calls; no separate translation subprocessor is added for the supported native-quality languages.
Step 6: Pilot with real patient calls (synthetic if needed)
Before going live, run a two-week pilot with non-English call volume routed to the multilingual answering service. If your practice doesn’t yet have measurable non-English call volume, simulate it: ask staff or volunteer patients to make sample calls in the target languages.
Score each call on intake-field completeness, urgency tagging, transcript accuracy, English translation quality. A vendor that gets 4/5 on these dimensions is production-ready. 3/5 is borderline. Below 3 isn’t ready.
Document the pilot. Patient-language coverage is the kind of thing that gets audited. Practice managers want a written record of due diligence, especially in California, Texas, and other states with active language-access enforcement.
Common questions
Questions practices ask first
How many languages does DeskMD support?
Pro supports 20+ languages at native quality with additional best-effort coverage. Standard is English-only. Ask for the current native-quality list before you sign — we update it as the underlying voice models improve.
Are transcripts in English or the original language?
Both. The intake card includes the English translation for triage and the original-language transcript for accuracy and audit.
Is there an extra charge for non-English calls?
On Pro, no. Multilingual is included in the per-provider rate. Some traditional services charge per language or per minute for non-English calls — read the rate card carefully.
Can the AI handle code-switching?
Modern voice models handle Spanglish and similar code-switching reasonably well. Test it with a sample call before relying on it for a specific patient population — performance varies by language pair.
What about ASL and deaf patients?
Voice answering services don’t handle ASL. Deaf patients should be supported through a relay service or video-call workflow at the practice. This is a known gap across the industry, including ours.
What to do next
Verify multilingual coverage for your practice
Stop missing calls. Start sleeping at night.
Give patients a real answer after hours and give your team a clean record in the morning.