It never sounds unsure. That's the problem. Five checks that take under a minute each.
The thing that trips people up isn't that these tools make mistakes. All tools make mistakes. It's that they make mistakes in the same calm, fluent, well-organised voice they use when they're right.
A person who isn't sure hesitates. They say "I think" or "you'd want to check that." AI gives you a wrong date in the same tone it gives you a right one.
So you need your own checks. Here are the ones I teach.
1. Anything specific gets verified
Names, dates, figures, quotes, legal references, statistics, citations, prices, people's job titles. If it's a specific claim about the real world, check it before you use it.
This sounds like a lot of work. In practice it's a thirty-second search for the two or three specifics in a piece, and it saves you the meeting where someone says "where did that number come from?"
2. Ask it to show its working
"Which parts of that are you confident about, and which parts should I verify?" is a question worth asking every time.
It won't be perfect at self-assessment. But it will usually flag the shakier claims, and that tells you where to point your attention.
3. Watch for the too-neat answer
Real situations are messy. If you asked about something genuinely complicated and got back a tidy list of five points that all agree with each other, be suspicious.
Push back. "What's the strongest argument against this?" "What would someone who disagrees say?" "What are you leaving out?" The answer to that third question is often the most useful thing in the whole exchange.
4. Notice when it agrees with you too quickly
These tools are built to be agreeable. Tell it your plan is good and it will find reasons why. Tell it the same plan is bad and it will find reasons for that too.
If you want a real opinion, don't tell it yours first. Say "here is a plan, I didn't write it, tell me what's weak about it." You'll get something far more useful.
5. Trust it least where you know least
This is the uncomfortable one. In your own field you can smell a wrong answer immediately. Outside your field you can't, and that's precisely where you're most tempted to rely on it.
So the rule is inverted from what feels natural. Use it freely inside your expertise, where you're a good editor. Use it carefully outside your expertise, where you aren't.
What good use looks like
None of this is a reason to keep away from these tools. It's how you use anything powerful and imperfect. You don't stop driving because cars can crash. You wear a seatbelt and you look before you pull out.
The professionals I see getting the most out of AI aren't the most trusting. They're the most sceptical. They use it constantly and they check everything that matters, and those two things are not in conflict.
Your judgement is the thing that makes this safe. It's also the thing no tool is going to hand you. You built it over years, and right now it's worth more than it's ever been.
Where next
If this was useful, the live sessions are the same thing with your own work in front of you and someone to ask when it goes sideways.