Artificial intelligence has become the first stop for millions of people seeking quick answers, creative help, or a sounding board for their thoughts.
Tools like ChatGPT, Claude, and Gemini can draft emails, explain complex topics, and even offer companionship during lonely moments. But as AI becomes woven into daily life, it’s worth pausing to consider its limits.
Some questions simply shouldn’t be directed at a chatbot; not because the AI will refuse, necessarily, but because the answer you get could be incomplete, wrong, or even dangerous.
Here are ten categories of questions you should think twice about before typing into an AI chat window.
1. Specific Medical Diagnoses
Describing your symptoms to an AI and asking “what disease do I have?” is a tempting shortcut, especially when doctor appointments are expensive or hard to schedule.
The problem is that AI models don’t examine you, can’t order lab work, and have no access to your medical history.
They pattern-match symptoms to conditions based on text, which means a headache and fatigue could just as easily be attributed to stress as to something far more serious.
Relying on an AI-generated diagnosis can lead to false reassurance or unnecessary panic.
AI can be a reasonable starting point for understanding general health concepts, but the diagnosis itself belongs with a licensed physician who can run tests and examine you directly.
2. Legal Advice for Active Cases
If you’re facing a lawsuit, a custody battle, or a serious contract dispute, asking an AI chatbot to tell you what to do is risky.
Laws vary enormously by state, country, and even municipality, and they change frequently. An AI might confidently cite a statute that’s outdated or apply a legal principle from the wrong jurisdiction.
Worse, legal strategy often depends on details a person wouldn’t think to mention, like the specific wording of a prior agreement or a procedural deadline.
For general legal education, understanding what a term means or how a process typically works; AI can help. But for anything with real stakes, a licensed attorney familiar with your jurisdiction is irreplaceable.
3. Financial Investment Decisions
“Should I put my retirement savings into this stock?” is a question best left out of any AI conversation.
AI models don’t have real-time access to your full financial picture, your risk tolerance, your tax situation, or your long-term goals. They also can’t be held accountable the way a licensed financial advisor can.
Even when an AI’s market analysis sounds sophisticated, it’s often working from outdated training data or generic reasoning rather than the nuanced judgment a fiduciary advisor brings.
Use AI to understand financial concepts like compound interest or the difference between a Roth and traditional IRA — not to make the actual call on where your money goes.
4. Requests to Confirm You’re Not Being Manipulated by a Partner
This might seem like a strange one, but a growing number of people ask AI to validate their feelings about a relationship, especially when something feels off.
The trouble is that AI only hears one side of the story, filtered through your own perception and emotional state at the moment.
It has no way to observe body language, tone, or the full context of years of shared history. An AI that leans too readily into agreement can accidentally reinforce a distorted view of a relationship, whether that means downplaying real red flags or, conversely, amplifying a misunderstanding into something it isn’t.
Trusted friends, family, or a therapist who can engage over time are far better equipped to help you navigate relationship concerns.
5. Anything Involving Building Weapons or Dangerous Devices
This one is fairly intuitive, but it’s worth stating plainly: don’t ask AI for help synthesizing dangerous chemicals, building weapons, or creating anything designed to harm people.
Reputable AI systems are built to decline these requests, but the underlying point stands regardless of whether the model complies.
This is information you shouldn’t want, and pursuing it through any channel, AI or otherwise, carries serious legal and safety consequences.
6. Diagnosing Your Mental Health
Similar to physical health, mental health is deeply personal and complex, and it deserves more than a text-based exchange with a machine.
Asking an AI “do I have depression” or “am I bipolar” invites a kind of pseudo-diagnosis that can be misleading.
Mental health conditions often share overlapping symptoms, and an accurate assessment usually requires a trained clinician who can ask follow-up questions, observe patterns over multiple sessions, and rule out other explanations.
AI can be a helpful space to process feelings or learn about a concept like anxiety in general terms, but it should never replace an actual mental health professional, especially if you’re in crisis.
7. Anything You Wouldn’t Want Leaked or Subpoenaed
Many AI conversations are stored, and depending on the platform and your settings, that data might be used for training, reviewed by human moderators, or retained for a period of time.
This means you shouldn’t treat an AI chat like a diary for your deepest secrets, confidential business strategies, or anything with serious legal exposure if it became public.
Even if a company promises strong privacy protections, data breaches happen, and conversations can sometimes be subject to legal discovery.
Treat AI chats the way you’d treat any digital communication: assume it could someday be seen by someone other than you.
8. Requests to Predict the Future with Certainty
People sometimes ask AI to predict election outcomes, stock market crashes, or natural disasters with a tone of expecting a definitive answer.
AI models are pattern-recognition systems trained on past data; they have no genuine foresight.
Their “predictions” are essentially educated guesses dressed up in confident language, which can be more misleading than not answering at all.
If you’re using AI to think through possible scenarios or historical trends, that’s reasonable. If you’re expecting it to tell you with certainty what will happen tomorrow, you’re setting yourself up for disappointment or, worse, poor decision-making based on false confidence.
9. Personal Information About Private Individuals
Asking an AI to find someone’s home address, phone number, workplace, or daily schedule crosses into territory that’s both ethically questionable and often technically impossible for a well-designed AI to fulfill.
Even when partial information is publicly available somewhere online, compiling it into a profile of a private individual raises serious stalking and harassment concerns.
Responsible AI systems are built to refuse this kind of request, and for good reason: the potential for misuse, from harassment to more serious harm, is too significant to ignore.
10. Whether You Should End Your Life or Harm Yourself
This is, without question, the most important item on this list. If you are having thoughts of suicide or self-harm, please do not treat an AI chatbot as your primary support system.
While many AI tools are designed to respond with care and point you toward crisis resources, they cannot replace the human connection, clinical judgment, and immediate intervention that a crisis counselor, therapist, or trusted person in your life can provide.
The Bigger Picture
None of this means AI is useless or untrustworthy across the board. It’s a remarkably capable tool for brainstorming, drafting, learning new concepts, and even offering emotional support in low-stakes moments.
The key is understanding where its usefulness ends and where human expertise, accountability, and connection need to take over.
AI doesn’t have a medical license, a law degree, a fiduciary duty, or a genuine understanding of your life circumstances.
Used thoughtfully, alongside real professionals and real relationships, AI can be a powerful complement to a well-lived life — but it was never meant to replace the humans who know you best.
