In plain words

This is a transparency page. It tells you which parts of LINC are done by AI and which parts are done by people, what the AI is based on, and what LINC does not do with your information. The goal is that you understand the tool well enough to decide how much to rely on it.

You can use a tool more wisely when you know how it works. This page is a plain, honest account of how LINC uses AI, where people are involved, what we do not do with your information, and where the AI has real limits. Nothing here is hidden in fine print.

At a glance
  • The chatbot answers, the short guide summaries, and first-pass translation are AI; the guides and the translation review are done by people.
  • The chatbot is grounded in the guides on this site, so it stays close to reviewed material, but it can still be wrong.
  • LINC has no accounts, does not store your chats, does not train AI on your messages, and shares nothing with immigration authorities.

What parts of LINC are AI

Some parts of LINC are powered by AI.

  • The chatbot answers. When you use Ask Chatbot, an AI writes the reply. It draws on the guides published on this site to answer your question.
  • The short guide summaries. The plain-language summary near the top of many guides is generated with AI to give you the main point quickly.
  • First-pass translation. LINC’s guides are first translated into other languages by machine, and each translated guide is marked as machine-translated until a person reviews it.

What parts of LINC are human

The foundation of LINC is written and checked by people.

  • The guide content. The guides on this site are written by people, for people trying to reach mental health care in the United States. The AI does not invent the guidance; it works from what people wrote.
  • Community reviewers. People from the communities LINC serves review the translated guides so the language fits how people actually speak and understand.

The design puts people at the base and uses AI on top of that base to make the material faster to reach and easier to search.

What the AI is grounded in

The chatbot does not answer from the open internet. It is grounded in the guides on this site, which means it builds its answers from material that people wrote and reviewed for your situation. This keeps its answers closer to reliable ground than a general chatbot would be. It also means the chatbot is only as current and complete as the guides behind it, so it can still miss things or get details wrong.

Here is how grounding works in plain words. When you ask a question, the chatbot first looks through the guides for the parts that match, then writes its answer from those parts. It is more like an open-book test than a memory test. Researchers who reviewed dozens of health AI studies found that grounding an AI in trusted source material lowered its made-up answers by roughly 30 to 50 percent, and that adding a human reviewer lowered them much further (Basu & Huynh, 2026). Even the best case is not zero, so grounding lowers the chance of a wrong answer without removing it. That is why LINC still keeps people in the loop and why the habits further down still matter.

You aska question in the chat
It findsthe matching guide parts
It answersfrom those parts
Grounding works like an open-book test, so answers stay close to guides people wrote and reviewed.

What LINC does not do

Some of the most important facts about a tool are the things it does not do.

  • No accounts. LINC does not ask you to sign in or give your name.
  • Conversations are not stored. LINC does not keep your chat history. When you send a message, it goes to an AI provider to write the reply, and that provider does not use it to train its models.
  • Your messages are not used to train AI.
  • Nothing is shared with immigration authorities. LINC does not collect your identity and does not share information with immigration enforcement.

Even with these protections, a good habit is to describe your situation in general terms and avoid typing identifiers such as your immigration status, case number, or ID numbers into any AI tool. The Federal Trade Commission gives this same advice for any chatbot, because whatever you type can be collected or stored by the company that runs it (Federal Trade Commission, 2024). For more, see Using AI safely for your health.

Try it now. Open Ask Chatbot and ask it one thing: "What are you based on, and what can you not help me with?" Read how it describes its own limits. Knowing what a tool says it cannot do is a fast way to calibrate how much to lean on it.

The AI’s known limits

We would rather you know these limits than discover them at a bad moment.

Watch out

The chatbot can give a confident answer that is wrong, especially about phone numbers, addresses, costs, and eligibility rules. A sure-sounding tone is not proof that an answer is right.

  • Translation quality varies by language. Researchers have measured that AI answers are often weaker in languages other than English (Jin et al., 2024), with the largest drops in languages that have less text online. When one team tested Google Translate on emergency room instructions, it kept the correct meaning about 94 percent of the time in Spanish but only 55 percent of the time in Armenian, and they concluded it should not be relied on for patient instructions by itself (Taira et al., 2021). This is why a person reviews each LINC translation. See AI may answer worse in your language.
  • The chatbot can be wrong. Like all AI chatbots, it can produce a confident answer that is incorrect, especially about phone numbers, addresses, costs, and eligibility rules. In one study, physicians reviewed answers from four well-known AI chatbots and found that 5 to 13 percent were unsafe enough to risk real harm (Draelos et al., 2026). A sure-sounding tone is not proof that an answer is right. See AI gets things wrong.
  • It refuses clinical roles. The chatbot will not diagnose you, recommend a specific medication, or act as your therapist. Those belong to licensed people. When the chatbot is unsure, it is designed to say so, and research shows that an AI expressing its uncertainty helps people rely on its answers more appropriately (Kim et al., 2024).

Why we tell you all this

Being open about how the tool works is part of using it safely. Research on AI assistants finds that clear signals about what a tool does and does not know help people rely on it more appropriately (Kim et al., 2024). This fits guidance from the World Health Organization, whose 2024 advice on AI in health warns that these models can give false, inaccurate, biased, or incomplete answers and calls for human oversight instead of blind trust (World Health Organization, 2024). If you can say back, in your own words, what LINC’s AI is for and where it falls short, you are in a strong position to use the useful parts and bring the high-stakes parts to a person. That is exactly how LINC is meant to be used.

How to use LINC well

A few simple habits let you get the good out of LINC while staying safe.

The one habit

Treat any phone number, address, cost, or eligibility rule as a draft to confirm. Check it against the official source before you act.

  • Treat any phone number, address, cost, or eligibility rule as a draft to confirm. Check it against the official source before you act: call or text 988 in a crisis, look up a clinic at findtreatment.gov or a community health center at findahealthcenter.hrsa.gov, and confirm coverage with your health plan or your state Medicaid office.
  • Bring high-stakes questions to a person. For anything about your immigration situation, a legal problem, or a medical decision, ask a community health worker or a licensed professional. The Federal Trade Commission advises the same for any chatbot: verify its answers with a trusted source, and do not rely on it alone for medical, legal, or financial advice (Federal Trade Commission, 2024). LINC is a starting point, not the final word.
  • Keep your details general. You do not need to type your name, immigration status, case number, or ID numbers to get useful help.
  • When an answer feels off, check it. Ask the chatbot which guide it used, open that guide yourself, or weigh the information against a trusted source (MedlinePlus).

These habits match what researchers find: people rely on an AI tool more appropriately when they understand its limits and confirm its high-stakes claims (Kim et al., 2024).

Common questions

Is a real person reading my messages? No. When you use Ask Chatbot, an AI writes the reply right then, and no staff member is watching your chat. People wrote and checked the guides the AI draws from, but no person reads what you type.

Can the chatbot call 988 or reach someone for me? No. The chatbot can only write words back to you. It cannot place a call, send a message, book an appointment, or contact a crisis line on your behalf. If you are in crisis, call or text 988 yourself, or 911 for immediate danger. What the chatbot can do is help you find the right number and think through what to say.

Will anyone find out I used LINC? There is no account and no saved chat history tied to you, because LINC never asks for your name and does not make you sign in. Your messages are not shared with immigration authorities. With any AI tool, a smart habit is to keep personal details to yourself, since the company running a chatbot can collect what you type (Federal Trade Commission, 2024). You can get real help from LINC while keeping your name, status, and case numbers private.

How do I know if the chatbot made something up? Treat every exact detail as something to confirm. Phone numbers, addresses, prices, dates, and eligibility rules are the parts AI gets wrong most often. Ask the chatbot which guide its answer came from and open that guide yourself. For anything that matters, check it against an official source before you act, such as findtreatment.gov or MedlinePlus. This is the same step the Federal Trade Commission recommends for any chatbot answer (Federal Trade Commission, 2024).

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.