The biggest risk with AI at work is not that it refuses to help. It is that it hands you something polished and slightly wrong, and you send it. The term for confident-sounding errors is hallucination, and every major assistant does it sometimes.
You do not have to stop using AI. You need a short routine for checking it.
Match the checking to the stakes
A brainstorm for weekend plans needs a glance. A message to a client needs a careful read. Anything involving money, health, legal matters, or safety needs a real source and often a professional. Decide how much checking a task deserves before you start.
The five-minute routine
1. Read it out loud. Awkward phrasing and claims you would never make yourself stand out when you hear them.
2. Underline every fact. Names, numbers, dates, places, and quotes are where errors hide. Make a quick list.
3. Verify the list. Check each item against an original source: the document, the website, the person. Do not ask the same AI to confirm its own claim and call it verified.
4. Ask it what it assumed. A good follow-up is: "What did you assume that I didn't tell you, and what might be wrong here?" This often surfaces gaps.
5. Compare to what you know. If something surprises you, slow down. Your own expertise is the best check you have.
Red flags
- A statistic with no source
- A quote you cannot find anywhere else
- Links or references you cannot open
- Perfectly confident answers about very recent events
- Advice that sounds specific but does not match your situation
Make it harder to go wrong
Give the AI the source material and ask it to use only that. "Use only the text below. If the answer isn't there, say so." This reduces invented details a lot, though it does not remove the need to check.
Own the result
When your name is on it, you are responsible, whatever tool helped. That is not a reason to avoid AI. It is the reason to stay in charge of it.
The free foundations course builds this habit from lesson one.