Two questions land in the same place. A practice manager wants to know whether staff can paste discharge instructions into Google Translate. A buyer wants to know whether an AI interpreter is allowed at all.
The honest answer involves separating them, because the 2024 rule treats written text and live speech under different provisions, and most content on this subject blurs the two.
Is it legal to use Google Translate in a medical setting?
Not on its own for material that matters. Under 45 CFR 92.201(c)(3), when a covered entity uses machine translation, the translation must be reviewed by a qualified human translator in three situations: where the underlying text is critical to the rights, benefits or meaningful access of an individual with limited English proficiency, where accuracy is essential, or where the source documents contain complex, non-literal or technical language.
The rule lets the tool stand, but not alone, where the stakes are real.
What machine translation means in the rule

The definition matters because it is narrower than people assume. 45 CFR 92.4 defines machine translation as automated translation, without the assistance or intervention of a human translator, that is used to translate text into another language.
Text. Not speech. That single word does a lot of work, and it is the reason the machine translation provision is the wrong place to look for the rules on a phone call.
So what falls inside it? Discharge instructions, consent forms, appointment letters, patient portal messages, financial responsibility notices, translated web pages. Anything written that you push through translation software. If a piece of that material carries a right, an instruction that affects safety, or dense clinical or insurance language, a qualified human translator has to review the output before it reaches the patient.
Two practical notes.
A disclaimer does not discharge the obligation. Adding a line saying the translation is automated does not turn a critical document into a non-critical one.
The review has to be by a qualified translator. 45 CFR 92.4 defines a qualified translator separately from a qualified interpreter: proficient in writing and reading both languages, able to translate effectively, accurately and impartially, with the necessary specialized vocabulary. Somebody bilingual on the billing team skimming the Spanish is not a review.
Can we use AI interpreters under Section 1557?
The rule does not name AI interpretation and does not ban it. What it does is set a standard for the interpretation itself. Under 45 CFR 92.201(c)(1), when interpretation is required, the covered entity must offer a qualified interpreter, and 92.201(e)(4) says a covered entity may not rely on staff other than qualified interpreters, qualified translators or qualified bilingual staff to communicate with patients who have limited English proficiency.
Read plainly, that means an AI interpreter is not itself a qualified interpreter. The definition describes a person who has demonstrated proficiency, interprets accurately and impartially, and adheres to interpreter ethics.
This is where most AI vendors overclaim, and it is worth being direct about instead. Where the law calls for a qualified interpreter, use one. Where the interaction is a routine, high-volume front-desk exchange, a language assistance tool that lets your own employee handle the call is a reasonable and useful thing to have, and it does not pretend to be the interpreter.
Traditional interpreter companies add a third person to the phone call. AI companies replace the employee entirely. Interpreterly does neither. The employee stays on the call and we carry the language, which is the shape that keeps a qualified employee inside the conversation rather than substituting for one.
The provision nobody in this industry writes about
If interpretation reaches your patient over audio, there is a section of the rule aimed squarely at the quality of that audio. 45 CFR 92.201(g) sets requirements for audio remote interpreting services, and almost no vendor page mentions it.
The rule requires real-time, full-duplex audio transmission, delivered over a dedicated high-speed, wide-bandwidth connection or system that produces high-quality audio without lags, choppy or blurry transmission, static or decreased audio quality. It requires a clear, audible transmission of voices. And it requires adequate training for users and other involved individuals so they can quickly set up and operate the service.

Two things follow for anyone buying phone interpretation.
First, the quality of your telephony is not somebody else’s problem. If your interpretation arrives choppy, drops, or requires patients to repeat themselves, the rule speaks to that.
Second, training is a named requirement, not a nice-to-have. A service nobody at the front desk knows how to start does not deliver meaningful access, and 92.201(g) makes that explicit.
For a product delivered by phone, this is the provision worth designing against. Telephony-tuned noise cancelling, low-latency turn-taking and read-back confirmation on names, numbers and IDs are exactly the properties this section describes in regulatory language. If you want to see the mechanics on a real call, here is how AI phone interpretation works.
Where the line actually falls in a clinic
The rule does not give you a list of critical and non-critical conversations, so here is a working split most compliance teams recognize. The right answer for your practice is the one your counsel signs off on.
Qualified human interpreter, every time. Informed consent. Diagnosis and prognosis discussions. Medication instructions and changes. Behavioral health assessment. Anything where a misunderstanding changes a clinical decision. Anything a patient would need to understand to exercise a right.
Documents that need a qualified translator to review any machine output. Consent forms, discharge instructions, financial responsibility notices, notices of privacy practices, appeal rights, and the notice of availability itself under 45 CFR 92.11.
Routine operational calls, where a language assistance tool earns its place. Appointment booking, reminders and confirmations, directions to the office, insurance verification, pre-visit paperwork prompts, billing questions that do not involve appeal rights.
That third category is where most call volume lives and where most practices are currently failing patients because the alternative is a hold queue. The fuller regulatory picture is in our guide to Section 1557 language access requirements, and if the option you are weighing is staffing rather than software, the true cost of relying on bilingual staff covers that trade-off.
What to change this quarter
- Inventory every patient-facing document that has been machine translated. Flag anything critical, anything where accuracy is essential, and anything technical.
- Route those through a qualified translator for review, and record who reviewed what and when.
- Stop staff pasting patient-facing text into consumer translation tools. Say so in the language access procedure, with examples of what is and is not acceptable.
- Write down which conversation types require a qualified human interpreter at your practice, and make it specific enough for a receptionist to apply.
- Test your audio. Place interpreted calls from the handsets your staff actually use, on the lines they actually use, and listen for the failures 92.201(g) names.
- Train the people who answer the phone on how to start the service, and document the training.
Anything touching HIPAA, the BAA or data handling sits in our compliance approach, and if you need to confirm which languages are live before you plan around them, ask us about your languages.
Frequently asked questions
Can we use Google Translate for patient documents? Not on its own where the text is critical to a patient’s rights, benefits or meaningful access, where accuracy is essential, or where the source material is complex, non-literal or technical. In those situations 45 CFR 92.201(c)(3) requires review by a qualified human translator.
Does the machine translation rule apply to phone interpretation? No. The rule defines machine translation as automated translation of text. Live interpretation is governed by the qualified interpreter requirement at 92.201(c)(1) and, where the service is delivered over audio, the quality requirements at 92.201(g).
Is an AI interpreter a qualified interpreter under Section 1557? No. The definition of a qualified interpreter describes a person who has demonstrated proficiency, interprets accurately and impartially, and follows interpreter ethics. AI language assistance can support routine communication, and qualified human interpreters remain necessary where the law requires them.
What are the audio requirements for phone interpreting? 45 CFR 92.201(g) requires real-time, full-duplex audio over a dedicated high-speed, wide-bandwidth connection, without lags, choppy transmission, static or degraded quality, clear audible transmission of voices, and adequate training for users.
Who counts as a qualified translator for reviewing machine translation? Under 45 CFR 92.4, a translator who has demonstrated proficiency in writing and reading English and the other language, translates effectively, accurately and impartially, and uses the necessary specialized vocabulary. A bilingual employee without that demonstrated proficiency does not meet the definition.








