In plain words

A green, yellow, and red guide sorts tasks by how much a mistake would cost you. Green means a chatbot is a good fit because you can catch any error before it matters. Yellow means the chatbot helps but you must check the important details with a person. Red means the decision is too important for AI and belongs to a qualified person every time.

An AI chatbot is a good fit for some tasks and a poor fit for others. Instead of guessing each time, you can sort what you need into zones: green, yellow, and red. The rule behind the zones is simple. The more a mistake would cost you, the more a person needs to be involved.

At a glance
  • Green tasks, like directions, preparing questions, or a first draft, are good to do with AI because a mistake is easy to catch.
  • Yellow tasks, like a summary of your own document, are useful with AI once you verify the details with a person.
  • Red tasks, like a diagnosis, a medication decision, legal advice, or anything you sign, always belong to a qualified person.
  • Crisis is always human: call or text 988, or 911 for immediate danger.
GreenGood with AI; a mistake is easy to fix
YellowUse AI, then verify the details with a person
RedAlways a person, every time
The more a mistake would cost you, the more a person needs to be involved.

Green: good to do with AI

These are tasks where an error is easy to notice and easy to fix before it matters. Use the chatbot freely here.

  • Getting directions and general steps, such as how a first therapy visit usually goes.
  • Preparing your questions for an appointment.
  • Understanding a term or a general topic, such as what “sliding scale” or “referral” means.
  • Drafting what you want to say, such as a phone script for calling a clinic or an email to a provider.
  • Practicing a hard conversation, such as how to tell a receptionist you need an interpreter, or how to ask your boss for time off, so the words feel ready before you say them.

You will get better answers if you tell the chatbot who you are and what you need. A short request like “Write three sentences I can read to a clinic to ask for a therapist and a [your language] interpreter, at a simple reading level” works better than a one-word question. In the green zone, you stay in control. You read the draft, you decide what to keep, and nothing happens until you choose to act.

Yellow: use AI, then verify the details

These tasks are useful with AI, but they contain specific details that need a second check before you rely on them.

  • Summaries of your own documents, such as a letter or a form explained in plain language. A summary is a helpful starting point, and the numbers, dates, dollar amounts, and deadlines inside it are exactly where AI readers make the most mistakes.
  • Local facts, such as which clinic takes your insurance or what a program costs right now.

The habit for the yellow zone is short. Use the chatbot to understand the general picture, then confirm the important details with an official page or a person before you act. A good way to check your own understanding is to say the summary back in your own words and ask a community health worker or interpreter, “Did I read this right?” For more on catching errors, see AI gets things wrong.

Try it now. Think of one task you were about to bring to a chatbot. Say out loud which zone it belongs in: green, yellow, or red. If it is yellow or red, name the one person you would check with. That is the whole decision, made in ten seconds.

Red: always a person

Some decisions are too important to base on a chatbot’s answer, no matter how clear it sounds. These belong to a qualified person every time.

  • A diagnosis. Only a licensed clinician who knows your full history can tell you what is going on.
  • A decision about medication. Starting, stopping, or changing a medicine is a clinician’s decision with you.
  • Legal or immigration advice. Bring these to a lawyer or an accredited legal advocate who is bound by confidentiality.
  • Anything you will sign, such as a consent form or an insurance appeal. You have the right to a free, qualified interpreter before you sign. See Your right to an interpreter.
  • A crisis. See the next section.

Crisis is always human

If you are in danger, thinking about harming yourself, or in crisis, do not turn to an AI. Go straight to a person.

If you are in crisis

Do not turn to an AI. Call, text, or chat 988, the Suicide and Crisis Lifeline, free and confidential 24 hours a day. Press 2 for Spanish, or stay on the line and ask for an interpreter. Call 911 for immediate physical danger.

  • Call, text, or chat 988, the Suicide and Crisis Lifeline. It is free, confidential, and open 24 hours a day, every day. You can call or text 988, or chat at 988lifeline.org. Press 2 for Spanish, or text the word “AYUDA” to 988. For other languages, stay on the line and ask for an interpreter; 988 connects callers with interpreters in more than 240 languages. Veterans and service members can press 1. If you use American Sign Language, you can connect by videophone (SAMHSA).
  • Call 911 for immediate physical danger.

988 is not only for thoughts of suicide. You can reach out for anxiety, depression, deep loneliness, the effects of trauma, a substance use problem, or simply needing someone to talk to. You can also call about someone else you are worried about. A trained counselor will ask a few questions to make sure you are safe, then listen without judging you (SAMHSA).

Many people worry that calling will bring the police. The Lifeline says its counselors work to help you in the least forceful way possible, and for most callers the counselor can help right there on the call, text, or chat. They involve 911 only when there is an immediate threat to someone’s physical safety (SAMHSA). You do not have to give your name, and you do not have to share your immigration status.

A chatbot’s response to a person in crisis can be uneven. When researchers compared chatbot answers with the judgments of expert clinicians, the chatbots lined up well for the lowest-risk and highest-risk questions but were least consistent for questions in the middle range of risk (McBain et al., 2025), which is exactly where a wrong response is most dangerous. This is why crisis routing goes to trained humans every time. If you are unsure, treat it as a crisis and call 988.

When you are not sure which zone

If you cannot tell whether a task is yellow or red, treat it as the more careful zone and bring it to a person. The cost of checking with a community health worker or an interpreter is small. The cost of acting on a wrong answer to a high-stakes question can be large. When in doubt, prepare with AI and decide with a person.

Is what I tell a chatbot private?

Treat a chatbot like a public place, not like a doctor’s office. A doctor, a clinic, and a therapist must keep what you tell them private under a federal law called HIPAA. A general chatbot is not your doctor and is not covered by that law, so what you type can be stored and used to improve the tool. See Your privacy: who can see your mental health records.

Good to know

Treat a chatbot like a public place, not a doctor's office. A general chatbot is not covered by HIPAA, so what you type can be stored. Leave out details that could identify you when you only need general information.

A safe habit is to leave out details that could identify you when you are only looking for general information. You can ask, “What usually happens in a first therapy visit?” without typing your full name, your address, your date of birth, or your immigration status. Save the private details for the person who is actually helping you, such as your provider, a community health worker, or an interpreter.

Where to find a real person

When a task is yellow or red, you do not have to find the right person on your own. These ways to reach one are free, and most offer help in your language.

  • Call, text, or chat 988, the Suicide and Crisis Lifeline, any time, day or night. Press 2 for Spanish, or stay on the line and ask for your language; interpreters are available in more than 240 languages.
  • Call 911 for immediate physical danger.
  • Dial 211 to reach a local helpline that connects you to counseling, food, housing, and other help near you. It is free and confidential, reaches 99 percent of the United States, and is offered in more than 180 languages (United Way, 211). You can call, and in many areas text or chat, around the clock. You do not have to give your name.
  • Find a low-cost clinic used to working with interpreters at findahealthcenter.hrsa.gov. Community health centers charge on a sliding scale based on what you can pay, and they serve you no matter your ability to pay or your immigration status.
  • Search for mental health or substance use treatment at findtreatment.gov, a free federal tool.
  • Reach the NAMI HelpLine for information and support at 1-800-950-6264, or text “NAMI” to 62640, Monday to Friday, 10 a.m. to 10 p.m. Eastern. Learn more at nami.org.

A community health worker, a caseworker, or someone you trust can sit with you while you make one of these calls. Asking for that is a normal, reasonable thing to do.

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.