Don't rely on AI for medical emergencies, medication decisions, mental health crises, legal filings, taxes, password security, genuine emotional moments, breaking news, investment predictions, or anything a licensed professional must sign off on. In each case, the risk of a wrong AI answer causing irreversible harm is too high. A real professional or verified resource is always the right call in these situations.
AI is genuinely useful for a long list of everyday tasks — drafting a letter, understanding a confusing bill, looking up a recipe, brainstorming ideas. But "AI can help with most things" is not the same as "AI can help with anything." There are specific situations where turning to a chatbot instead of the right resource leads to real harm. This guide is not about AI's abstract limitations — it's a concrete list of situations you are likely to encounter, with the right alternative for each one.
1. A Medical Emergency
Do not use AI. In any situation where someone may be having a heart attack, stroke, severe allergic reaction, or serious injury, every second counts. Typing a question into a chatbot and waiting for a response wastes time that the right action — calling for emergency help — does not.
Do this instead: Call 911 in the US (999 in the UK, 112 across Europe and internationally) immediately. If you are unsure whether something is an emergency, call anyway — dispatchers are trained to help you evaluate the situation while help is already on the way.
2. Medication Dosing or Drug Interactions
Do not use AI. A 2023 expert review found that 74% of ChatGPT-generated drug information was incomplete or incorrect. AI chatbots are trained on text, not on current clinical guidelines or your personal health history — and a wrong answer about dosing or a dangerous drug combination can cause serious harm.
Do this instead: Ask a pharmacist. Pharmacists are licensed drug experts who are almost always reachable by phone and free to consult without an appointment. For checking interactions between specific medications, Drugs.com and the FDA's Drug Interactions Checker are well-maintained, free public resources.
3. A Mental Health Crisis
Do not use AI. If you or someone you know is experiencing thoughts of self-harm, suicidal ideation, or an acute mental health emergency, a chatbot is not a safe substitute for a trained crisis counselor. AI cannot verify what it is being told, cannot follow up on a person's wellbeing, and cannot dispatch help if needed.
Do this instead: In the US, call or text 988 to reach the Suicide and Crisis Lifeline — free, confidential, and available around the clock. In the UK, call Samaritans at 116 123. For other countries, findahelpline.com lists verified crisis lines worldwide.
4. Legal Filings with Hard Deadlines
Do not use AI to draft court documents or calculate filing deadlines. Missing a court deadline because of incorrect AI-generated information can result in losing a case by default — a legal consequence that is very difficult to undo. AI often presents information confidently even when jurisdiction-specific rules are wrong or outdated.
Do this instead: Contact a licensed attorney or your local legal aid office. Many areas have free legal aid services for people who cannot afford private counsel. Your state bar association's website (or the equivalent in your country) typically has a lawyer referral service with free initial consultations.
5. Your Taxes — Specific Numbers and Deadlines
Do not use AI to calculate your tax liability or confirm filing deadlines. Tax rules change yearly, vary by state or region, and depend on personal circumstances a chatbot cannot reliably account for. If the AI gives you a wrong number or a missed deadline, you — not the AI — face the consequences.
Do this instead: Use your official tax authority's website (irs.gov in the US, hmrc.gov.uk in the UK). In the US, IRS Free File is available if you qualify, AARP Tax-Aide offers free preparation help for people 50 and older, and the Volunteer Income Tax Assistance program helps those below a certain income level.
6. Managing Passwords and Account Security
Do not ask AI to generate or store your passwords — and do not type your existing passwords into any chat interface. Some chatbots retain conversation history, and anything you type into them may be stored or logged. A chatbot cannot actually secure your accounts.
Do this instead: Use a dedicated password manager. Bitwarden is free, open-source, and widely trusted. Apple Keychain (built into iPhone and Mac) and Google Password Manager (built into Android and Chrome) work well for basic use. For two-factor authentication, use an authenticator app like Google Authenticator or Authy.
7. Writing a Grief Message or Navigating a Serious Personal Conflict
Do not outsource genuinely emotional moments to AI. When a close friend loses someone they love, or when you need to repair something that matters in an important relationship, what counts most is that the words came from you. An AI-written message can feel hollow — and in the wrong moment it can do more damage than saying nothing at all.
Do this instead: Write something yourself, even if it's short and imperfect. "I don't have the right words, but I'm thinking of you" from a real person means more than a polished paragraph that wasn't. For ongoing or serious conflict, a licensed therapist or mediator can offer structured support that AI cannot replicate.
8. Verifying Breaking News During an Emergency
Do not rely on AI when local conditions are changing by the minute. Most AI chatbots — including the widely used ones — are not connected to live news feeds or real-time data. In a fast-moving situation like a wildfire evacuation, a storm warning, or an industrial incident, AI may confidently describe conditions that are hours out of date.
Do this instead: Go directly to official sources: your local emergency management agency's website, the National Weather Service at weather.gov (in the US), local radio and TV broadcasts, or your government's emergency alert system. In the US, FEMA's website and app are reliable during declared disasters.
9. Investment Decisions Based on AI "Predictions"
Do not let AI-generated forecasts drive significant financial decisions. AI cannot reliably predict market movements. Language models generate plausible-sounding text — which means they can produce detailed, confident financial forecasts that have no factual basis. There are documented cases of chatbots inventing historical stock prices and market events that never occurred.
Do this instead: For significant financial decisions, consult a licensed financial advisor or a fee-only fiduciary — someone legally required to act in your interest. For free education, the SEC's investor.gov and FINRA's investor education resources are trustworthy starting points.
10. Anything That Needs a Licensed Professional's Signature
Do not substitute AI for licensed expertise in high-stakes domains. When a structural engineer stamps a building plan, they accept legal responsibility for it. When a doctor writes a prescription, they have examined you and taken professional liability. AI carries none of that responsibility — and "an AI told me" is not recognized as a defense in any professional or legal context.
Do this instead: Identify the licensed professional whose field covers your need and consult them. Many professions offer free initial consultations. Professional associations in law, medicine, and financial planning frequently have pro bono programs for people with limited resources.
What to Watch Out For
The most dangerous thing about AI in these situations is not that it sounds uncertain — it is that it often sounds completely confident. As we explain in Can You Trust ChatGPT?, AI systems are designed to produce fluent, authoritative-sounding text regardless of whether the underlying information is accurate. That quality works well for low-stakes creative tasks and becomes a serious liability in high-stakes ones.
Research by AARP has found that misinformation is the single biggest AI concern among older adults — and that concern is justified. The best protection is knowing in advance which situations call for a licensed human expert, not a chatbot.
What to Try Next
For a plain-English explanation of why AI gets things wrong even when it sounds certain, Why AI Makes Things Up — and How to Catch It covers the mechanics behind the problem. If you use AI for health-related questions at all, How to Use AI for Health Symptoms Without Getting Burned has the specific limits worth knowing.
