AI often gives wrong information. You try to correct it by calling it out, but is it actually learning to fix itself?
I’ve asked different AI apps for information and received wrong answers. In fact, if I had used some of that information in conversation — let alone in publication — I would have been the laughing stock.
The problem is simple: AI has not reached a level where the information it gives is 100% reliable.
So you have to be careful. Very careful. Especially if you’re using AI for something important — scenario planning, budgets, travel decisions, health insurance choices, or financial matters.
There is plenty of evidence online showing that AI makes mistakes. Just read this article and you might be shocked: https://joshbersin.com/2025/10/bbc-finds-that-45-of-ai-queries-produce-erroneous-answers
In February 2024, an Air Canada chatbot gave false information about bereavement fares. It told a customer they could claim a discount after booking, when the airline’s policy required the discount to be applied beforehand. A tribunal later ordered the airline to pay damages.
That shows how AI errors can have real-world consequences.
I’ve found a fun way to test AI: ask it something you already know the answer to. The response gives you a quick sense of how reliable that particular tool is.
Today’s AI systems can’t correct themselves very well. Engineers still have to go in and update or retrain them.
In the future AI systems will need to include systems that improve themselves, reduce errors over time, and become more reliable the longer they operate.
Until then caution is essential.
https://joshbersin.com/2025/10/bbc-finds-that-45-of-ai-queries-produce-erroneous-answers
