Lesson 4 of 5, 11 min with the exercise
Check the output before you use it
Spend your checking time where mistakes are likely and expensive.
Reading everything closely wastes the time you saved, and skimming misses the errors that sound right. Check by type instead.
- Names, numbers, dates and quotes: compare each one with the source.
- Links and references: open them. Invented ones often look real.
- Sums: redo them in a spreadsheet.
- What's missing: ask yourself what you'd expect to see that isn't there. A summary can be accurate and still leave out the main risk.
How hard you check depends on what happens if it's wrong. A draft email to a colleague needs a read. A number going to a client or a regulator needs tracing to its source. If you sign it, you own it.
See it in practice
In 2024, a Canadian tribunal ordered Air Canada to pay a customer about C$812 after its website chatbot gave him wrong information about bereavement fares (Moffatt v. Air Canada). The airline argued that the chatbot was responsible for its own words, and the tribunal rejected that. The sum was small, but the principle matters: the company answers for what its AI tells people.
Common mistake: Spending the same effort on everything. Set your checking by the cost of being wrong, or you'll either burn the time you saved or miss the mistake that matters.
Try it
Take one piece of AI output you were about to use and go through the four checks. Note what you found. After a few rounds you'll know where your tool tends to slip.
You're done when
- You've traced every number and name to a source
- You've asked what's missing as well as what's wrong
Sources: Moffatt v. Air Canada, 2024 BCCRT 149, the tribunal's decision