In plain words

AI translation is when a computer automatically converts words from one language to another without a human translator. Medical words are precise, and a computer can pick the wrong word and change the meaning of a diagnosis or instruction in a way that could cause harm. That is why, for anything important such as a consent form or a medication explanation, a trained human interpreter is still the safer choice.

AI translation tools have become much more common. They are built into phones, browsers, and search engines, and they can help you understand everyday information in your language. For many routine things, that is a real help. Not all translation is equal, though: a tool that renders “take once daily” as “take all at once” in a lower-resource language could lead someone to take a full week’s dose in a single day. Knowing when to trust a machine and when to ask for a person can protect you from errors that matter.

At a glance
  • Machine translation is often fine for everyday information, but it makes serious errors on medical and legal wording, and more errors in some languages than others.
  • For consent forms, a diagnosis, medication instructions, or insurance paperwork, ask for a free, qualified human interpreter.
  • Consumer mental health apps are often not covered by HIPAA and can share or sell your data; check the privacy policy, and in a crisis call or text 988.

An AI chatbot is not a therapist

If you are using an AI chatbot or companion app for emotional support, talking through worries, or processing difficult feelings, it is important to know this: the chatbot cannot replace a therapist, a counselor, or a human relationship. AI companions can feel responsive and supportive, but they cannot diagnose a mental health condition, provide treatment, or recognize when a conversation is moving toward a crisis. They are not bound by the ethics and training that govern licensed mental health providers.

If you find yourself turning to an AI companion regularly for emotional support, that is a signal to reach a real person. A community health worker, a counselor, or a trusted friend can offer what an AI cannot. If you are in distress right now, call or text 988 (the Suicide and Crisis Lifeline), available 24 hours a day in more than 240 languages. Decisions about your mental health care are too important to make based on a chatbot’s answer.

How to evaluate a mental health app

Before you trust an app with your personal information, ask a few questions.

  • Who made it? Look for a research institution, a hospital, or a government-funded program. An anonymous developer with no named organization behind it is a warning sign.
  • What does the privacy policy say? Find it before you register. It should state clearly what information is collected, who can access it, and whether data is shared or sold.
  • Does it connect you to a real clinician? A trustworthy app makes it easy to reach a human provider. It does not try to replace professional care.
  • What does it do in a crisis? A responsible app will immediately direct you to call or text 988. It will not try to handle a crisis through chat.

For health information you are uncertain about, go to a government source first. MedlinePlus, run by the U.S. National Library of Medicine, offers a clear guide for judging whether health information online is trustworthy.

And when what you really need is a person, not an app, you can find low-cost care in person or by video at FindTreatment.gov or a community health center at findahealthcenter.hrsa.gov. Both serve people regardless of ability to pay and can arrange an interpreter.

AI translation: what works, and where to be careful

For everyday language, machine translation is often useful. If a clinic website describes its hours, or a reminder card tells you to arrive fifteen minutes early, a translation app will usually get that right. Using one to get a general sense of a page or document is a reasonable first step.

Medical and legal language is harder. Clinical words are precise. Terms like “malignant,” “acute,” “prior authorization,” or “contraindicated” carry exact meanings that do not always cross languages cleanly. A translation can sound fluent and still be wrong in a way that changes what you understand about a diagnosis, a medication dose, or your rights. In one study, doctors ran emergency department discharge instructions through Google Translate: the Spanish and Chinese versions were mostly accurate, but a meaningful share still contained errors, and some of those errors were serious enough to potentially harm a patient (Khoong et al., 2019).

Some languages are at greater risk. AI translation tools are trained on large amounts of written text. Languages with less online content available, including many spoken by immigrants and refugees in the United States, tend to produce less reliable translations. A confident, fluent-sounding output is not a reliable sign of accuracy. Researchers have measured this drop for Spanish (Gonzalez Fiol et al., 2024), Arabic (Sallam et al., 2024), Chinese, and other languages (Jin et al., 2024); for what it means for you and how to work around it, see AI may answer worse in your language.

For anything important, ask for a qualified human interpreter. Consent forms, insurance paperwork, a diagnosis being explained by a provider, medication instructions, and any legal step in your care are situations where accuracy is not optional. Under Section 1557 of the Affordable Care Act (the health civil-rights law) and Title VI of the Civil Rights Act, most health programs that receive federal funding must provide a free, qualified interpreter. You cannot be charged, told to bring your own, or asked to use a family member or child. See Your right to a free, qualified interpreter for step-by-step instructions on how to ask.

BetterQualified human interpreterFor consent forms, a diagnosis, medication instructions, insurance paperwork, or any legal step in your care. It must be free, and accuracy is not optional.
AvoidMachine translation aloneFor medical or legal wording. It can sound fluent and still change what you understand about a diagnosis, a dose, or your rights.

Practical ways to double-check a translation. If you have a machine translation and want to verify it, try running the same text through a second tool: Google Translate and DeepL sometimes catch each other’s errors. Another quick test is to translate the result back into English and see whether it still says what the original said; if the meaning drifts, that sentence needs a person. Ask a bilingual community health worker or navigator to look it over. For any document you are about to sign, ask the clinic to arrange an interpreter before you put your name on anything. If a number is involved, such as a dose, a date, or a dollar amount, read the number itself in the original document; numbers are where a small translation slip does the most harm.

Try it now. Take one sentence from a health message you received and run it through two different translation tools. Read both results side by side and notice any place where the meaning shifts. If a word changes the instruction, that is exactly the kind of sentence to confirm with a qualified interpreter.

Mental health apps and AI chatbots: privacy and data

HIPAA often does not cover consumer apps. Many people assume that any health app must follow HIPAA (a US health privacy law) rules because it handles health information. That is not always true. HIPAA applies to health care providers, health plans, and their business partners. A consumer mental health app you download on your own is often not covered by HIPAA, which means the app may be able to share or sell your personal data in ways a doctor’s office cannot. The HHS HIPAA and Health Apps page explains how coverage works in detail.

Be careful about what you type. Before you enter personal information into any app, think about what it can see: your mood, your symptoms, your family situation, your immigration status. The FTC has warned consumers that some health apps collect sensitive data and share or sell it to third parties, including advertisers. The FTC’s consumer guide Does your health app protect your sensitive info? explains what to look for before you sign up for anything. Avoid entering immigration details or document information into any consumer app.

This is not just a worry; it has happened. In 2023 the Federal Trade Commission ordered the online therapy company BetterHelp to pay $7.8 million and banned it from sharing users’ health data for advertising. The FTC found that after promising to keep information private, the company had shared users’ email addresses, IP addresses, and their answers to sensitive mental health questions with Facebook, Snapchat, Pinterest, and an advertising company (Federal Trade Commission, 2023). Independent reviewers have found the problem is widespread. When the Mozilla Foundation studied 32 popular mental health apps, most carried a privacy warning label, and some contained large numbers of hidden trackers that begin collecting data almost the moment you open the app (Mozilla Foundation, 2023). The lesson is not that every app is bad, but that a calming design and a caring tone tell you nothing about what happens to your words behind the screen.

Watch out

Some apps collect sensitive data and share or sell it, even after promising to keep it private. Never type your immigration status, your documents, or another person's private details into a consumer app.

A few habits protect you. Sign up with your email and a new password instead of tapping “continue with Facebook” or “continue with Google,” which can link the app to your other accounts. Skip permissions the app does not truly need, such as your contacts, photos, or location. Remember that deleting an app from your phone does not always delete the data the company already stored, so if you want your information erased, look for a “delete my account and data” option or email the company to ask. And never type your immigration status, your documents, or another person’s private details into a consumer app.

In a crisis, do not use an app or the chatbot

If you are in crisis

No app or chatbot should be your first contact in an emergency. If you are in danger or thinking of hurting yourself, call or text 988, the Suicide and Crisis Lifeline, free and 24/7 in more than 240 languages. For immediate physical danger, call 911.

No app or AI should be your first contact in a mental health emergency. If you are in danger, thinking of hurting yourself, or in crisis, call or text 988 (the 988 Suicide and Crisis Lifeline). It is available 24 hours a day, 7 days a week, at no cost, and connects callers to interpreters in more than 240 languages. For immediate physical danger, call 911. The chatbot is designed to send you directly to crisis resources if you are in distress; it is not a crisis service itself.

For more on finding care online, see Telehealth and online therapy. For the bigger picture on using the chatbot safely, see Using AI safely for your health.

Sources

LINC is a research prototype, not a medical or legal service. Translation can be wrong; you have the right to a free, qualified interpreter. For anything high-stakes, do it with a community health worker or someone you trust. In a crisis, call or text 988, or 911 for immediate danger.