Things NOT to do
~ min read
30-second summary
- Even when AI is good at something, there are things it shouldn’t decide for you: not because of limits in the technology, but because of who takes responsibility.
- Two quick tests: if it’s wrong, is the damage reversible? Would you catch it in time? If even one answer is “no”, use AI to understand, not to decide.
- Five hot areas: health, money, legal matters, decisions that affect other people, work where your signature stands for your own competence.
- In every case the pattern is the same: AI helps you understand, prepare, explore. The decision stays with you, or with a professional who signs for it.
Up to now the manual has shown you what you can do with a conversational AI: write, understand, summarize, study, work with photos and files. This lesson is the other side of the coin. It’s not about the AI’s technical errors (that’s When to trust it and when not) or about privacy (that’s What you share when you use AI). It’s about things the AI can do, but that you shouldn’t be asking it to do anyway.
The overarching rule
Section titled “The overarching rule”Before the examples, the general rule, in one sentence:
If the cost of being wrong is high and you wouldn’t notice immediately, use AI to understand, not to decide.
Two questions to ask yourself before handing something over:
- If the AI gets it wrong, is the damage reversible? An email you can rewrite is reversible. A medication taken the wrong way, a signature on a contract, an investment placed, a message that breaks a relationship: no.
- Would I notice in time? A wrong summary you catch by rereading the document. A tax tip that makes you pay double you notice six months later, when the notice arrives. Watch out for the worst case: if the topic is specialist (medicine, law, taxes) and you lack the competence to judge the answer, the test doesn’t hold. You’ll never spot the error.
If even one of the two answers is “no”, AI stays a tool for exploration, not for decisions.
Five areas where not to delegate
Section titled “Five areas where not to delegate”1. Health
Section titled “1. Health”Don’t ask:
- “What is this symptom?” expecting a diagnosis.
- “Can I stop taking this medication?”
- “Can I mix these two drugs?”
You can ask:
- “What does this lab report say in plain language, and what questions should I ask my doctor?”
- “What does HbA1c mean in my blood test?”
- “What side effects are usually reported for this active ingredient? I want to be able to mention them to my doctor.”
Why the difference? A diagnosis isn’t made from an isolated symptom without seeing the person, reading their medical history, evaluating the whole picture. A doctor can do that; the AI can’t: it can’t examine you, it can’t order tests, and it doesn’t answer for what it tells you. If it’s wrong, your body pays the cost.
2. Money
Section titled “2. Money”Don’t ask:
- “Where should I invest twenty thousand dollars?”
- “Should I buy this house?”
- “Do my tax return for me.”
You can ask:
- “Explain the difference between an ETF and a mutual fund, so I know what my bank is talking about.”
- “Does this variable-rate mortgage offer have a renegotiation clause? Explain it in plain English.”
- “What documents do I need to deduct medical expenses? I want to show up prepared at my accountant’s office.”
Why? Your situation is specific: income, family, obligations, how much risk you can bear. The AI doesn’t know it, and generic advice applied to a particular case does damage. For real decisions you need a professional who signs for them.
3. Legal matters
Section titled “3. Legal matters”Don’t ask:
- “Can I sign this contract?”
- “Am I right in this dispute with my neighbor?”
- “Write me the demand letter with the right clause for my case.”
You can ask:
- “What does the non-compete clause in this contract mean? Explain it with an example.”
- “What questions should I ask a lawyer before our first meeting on this topic?”
- “When a labor dispute goes to court, what are the usual steps? I want to understand what to expect, not to act on my own.”
Why? The law is local, changes often, and varies on the details of the single case. An answer that sounds right in the abstract can be wrong for you. The consequences of a signature or a lost case are usually irreversible.
4. Decisions that affect other people
Section titled “4. Decisions that affect other people”Don’t ask:
- “Write me the message to end this relationship.”
- “How do I fire this person on my team?” meant as full delegation of the conversation.
- “You decide what to tell my father about his diagnosis.”
You can ask:
- “I need to have a difficult conversation with someone about X. What points should I cover? How might they respond?”
- “Help me understand what the other person is probably feeling in this situation.”
The line is who holds the words. “You write the message I’ll send” is delegation. “What tone should I pick? Anticipate three possible reactions” is help thinking it through.
Why? Other people aren’t boxes to optimize. A difficult conversation is the moment a relationship is built or broken: if you delegate it to text you didn’t write, the other person feels it. The difference between an “AI-written” message and a message from you, for someone who knows you, is palpable.
5. Work where your signature stands for your own competence
Section titled “5. Work where your signature stands for your own competence”Don’t ask:
- Theses, essays, papers meant to demonstrate your ability.
- Job applications (CVs, and especially cover letters) where you’re selling yourself on a skill.
- Technical samples or professional work where someone is paying for your skill: a piece by a journalist, a translation by a translator, a draft by a consultant, code by a developer.
You can ask:
- Help getting unstuck, rephrasing a paragraph, finding examples, reading with a critical eye.
- Early brainstorming on an idea, with the text you write later.
- A second opinion on style, the way you’d ask a colleague.
Why? Your signature carries a message to the reader: “this work reflects my skills”. If the signature is yours but the work is the AI’s, you’re sending a false signal. A brilliant application that gets you the interview but that you can’t back up in the interview itself is a boomerang: the promise was strong, you weren’t.
One compass
Section titled “One compass”In all five cases the pattern is the same. Don’t ask the AI to decide. Ask it to explain, prepare, explore. The right verb is understand.
In practice, the difference is who signs. Asking “what does this clause mean?” and then deciding with a lawyer: you’re the one deciding, the AI gave you context. Asking “can I sign this contract?”: you’re asking for a decision, and if the AI is wrong, you have no way to defend yourself.
This isn’t a technical rule imposed by the software. It’s a stance of responsibility: the things that weigh, you decide, possibly with a professional who signs for them. The AI is an assistant that prepares you for the conversation, not the conversation itself.
Check what you understood
Section titled “Check what you understood”What comes next
Section titled “What comes next”This module ends here. You now know how to use AI to read the world around you (documents, photos, files), to study, to write, and you know what to keep out of the conversation. In the next module, At work, the context changes: company data, clients, colleagues, and new ground rules.