It can be wrong with total confidence
AI doesn't mumble when it's unsure. It states wrong things in the same steady voice as right ones. Double-check anything that matters: names, numbers, dates, laws, medical claims, prices.
Try: “How sure are you about this? List anything I should verify myself.”
Make it show its sources
For factual questions, use a tool that searches the web and cites where it looked, then actually click a link or two. If it can't point to a source, treat the claim as a lead, not a fact.
Think before you paste
Keep passwords, card numbers, and other people's private details out of the chat box. If it's work material, check what your employer allows first. Good rule: if you wouldn't email it to an outside company, don't paste it into one.
Polished isn't the same as right
AI writing sounds finished on the first pass, and that shine can lull you into skipping the read-through. If your name is going on it, read every line before you send it. You own what you sign.
Ignore the hype merchants
Anyone promising AI will make you rich quick, or selling 'secret prompts' at a markup, is selling you. The real tools are cheap or free, the real skills are learnable, and real results take real work.
Don't collect subscriptions
You don't need five AI subscriptions. Start on free plans, notice which tool you actually reach for every day, and pay for that one. Upgrade when you keep hitting the free limit, not before.
Watch the calendar
Models learn from data with a cutoff date, so 'what's the best phone right now' can come back a year stale. For anything current (prices, news, laws, products), use the web-search mode or ask it to look things up.