It sounds sure. Check anyway.
A contractor I know asked ChatGPT about the rules for a home renovation permit in his municipality. He got a clear, well-written answer: the square footage thresholds, when a permit was required, the general process for applying. He quoted it to a homeowner client.
The rules had changed eight months earlier.
The client found out when he submitted the application. My contractor friend found out when the client called him about it.
The answer ChatGPT gave wasn't fabricated. It was accurate at the time it was written, based on the information the program had learned from. The problem is that ChatGPT doesn't know what it doesn't know. It had no way to tell my friend that its information on this particular bylaw was out of date, because it had no way to know that it was. It answered with the same confident, complete tone it uses for everything.
That's the one thing worth understanding before you use AI regularly.
What AI can do here
AI is genuinely useful for a very wide range of questions, and most of those questions don't require pinpoint current accuracy. If you're asking about how something works, or looking for ideas, or trying to get words on a page, or thinking through a decision, the slightly-out-of-date knowledge base doesn't matter much. The risk is low and the benefit is real.
The situation changes when you're acting on specific facts: a price, a bylaw, a medical guideline, the hours a place is open right now. Anything where being wrong by a year or two has real consequences. For those, ChatGPT is a useful starting point, not a final answer.
Confidence in the writing tells you nothing about whether the claim is current.
Try this
This week, whenever you use ChatGPT for something practical, do one quick check:
- Ask yourself: does acting on this require the information to be current and accurate?
- If the answer is no (you're writing something, brainstorming, thinking through a problem), use it freely and don't worry.
- If the answer is yes (the fact, rule, price, or policy matters), take the specific claim you're about to act on and check it from a primary source. That means the government website, the official business website, or a phone call. Skip the second-hand article, which may have taken its answer from ChatGPT in the first place.
Most of what ChatGPT produces you can use as it comes. The habit to build is asking, "does this need to be current?" before you act on something specific.
There's also a different kind of error worth knowing about. Sometimes ChatGPT will invent a source. It might say something was written by a specific expert, or published in a specific report, and that source doesn't exist. This happens more with detailed factual claims than with general knowledge, but it happens. If a source sounds specific and useful, check that it's real before you cite it to anyone.
What just happened
ChatGPT learned from text written up to a certain date. It knows a great deal, more than any one person could know, but that knowledge is frozen at the moment it stopped reading. It's the road atlas in the glovebox: everything in it was right the year it was printed, and it has no way of telling you which roads have been rerouted since.
The program is doing exactly what it's built to do, which is answer from what it learned. Knowing where that holds and where it thins out is what makes using it safely routine. Training a new model costs real time and real computing power, so the knowledge cutoff is part of how the tool works, the same way a printed map has a publication date on the cover.
One thing to remember
Confidence in the writing tells you nothing about whether the claim is current. ChatGPT writes with the same confident, clear voice whether it's right, slightly out of date, or completely mistaken. Read a well-written response the same way you'd read a rough one when the claim matters.
My contractor friend now checks anything regulation-related through the municipality's website before quoting it to a client. He still uses ChatGPT, often and for a lot of things. He told me the habit took about a week to stick. He's had no problems since.
Keep using the tools. The check is now part of the job, the same way it is with anything else you'd put your name to. The companion piece, on what to do when the first answer is close but not quite right, is here, and if your team is leaning on these tools without having agreed what checking looks like, that is the conversation I have most often in training.
Making Sense of AI is a six-part series walking first-time users through the basics, one skill at a time. It starts with what AI actually is, and if you're bringing these tools into a classroom or a team, the two ways teaching with AI goes wrong is the one to read next.
The same habit scales. When AI output touches a client site, someone has to own the check, and that verification discipline is the core of my AI operations audit. If AI has gotten ahead of your review process, let's talk.