It Will Lie to You With a Straight Face
Not on purpose — it isn't sneaky. Remember what it's doing: predicting what sounds right, not looking up what is right. So when it doesn't know something, it doesn't stop. It produces something plausible.
The industry calls that a hallucination. You should call it what it is: it made that up.
There is no tone of voice for "I'm making this up."
It all sounds equally certain.
That's the whole danger. A real statistic and an invented one arrive in the same confident sentence, with the same clean grammar, at the same speed. Nothing tips you off. This has caught out actual attorneys who filed AI-written briefs citing court cases that never existed.
When to Relax, When to Verify
You don't need to fact-check everything. You need to know which side of the line you're on.
✓ Trust it fairly freely
- Rewriting something you wrote
- Changing tone — warmer, shorter, friendlier
- Brainstorming ideas and names
- Explaining a general concept
- Fixing grammar and spelling
- First drafts you'll edit anyway
✗ Verify before you use it
- Any number, statistic or date
- Names of people, books, studies, cases
- Prices, laws, rules, deadlines
- Medical or legal specifics
- Anything going out under your name
- Anything a customer will rely on
The pattern's simple: it's excellent at shaping words you already have, and unreliable at supplying facts you don't.
Asking "are you sure?" is not verification
It'll happily tell you it's sure. It can invent a confirmation exactly as easily as it invented the original. The check has to come from outside the tool — a real source, a real website, a real person.
The Other Half: What Not to Type In
Checking its output is one habit. Guarding your input is the other. On a free plan, assume what you type may be used as training material.
Never paste this into a free chatbot
- Customer or client information — names, addresses, account numbers
- Anything medical
- Passwords, card numbers, bank details
- Social Security numbers
- Employee records or contracts you don't own
Don't put anything into a free tool
you wouldn't post in public.
Skits' Tip
You almost never need the real details anyway. Don't paste the customer's actual email — describe the situation: "a customer is unhappy about a late order, help me reply." You get the same useful answer, and nothing of theirs ever leaves your desk. If you want the full picture on data, plans and who's legally on the hook, AI: Do I Need It? covers it properly.
Quick Check
Which of these can you take at face value without checking?
That's the skill.
You can pick a tool, ask it properly, refine what comes back, and tell the difference between a good-sounding answer and a true one. That last part is what separates people who get value from AI from people who get embarrassed by it. Quick recap and the certificate's yours.