I’ve spent a lot of time using ChatGPT, not just for answering questions, but as part of wider digital and technology work. The more I’ve used it, the more useful it has become. But I’ve also learned that getting value from AI is not simply about asking better questions and accepting whatever comes back.
Sometimes the most important thing you can do is challenge the answer. Here are three practical lessons I now keep in mind when using ChatGPT.
1. Don’t mistake confidence for accuracy
The wider lesson is simple: confidence and accuracy are not the same thing. AI can help you move much faster, but that speed makes verification more important, not less.
2. The more customised your AI becomes, the more governance matters
The second lesson came from moving beyond general ChatGPT conversations and working with more customised AI setups. Today, plugins can bring together reusable instructions, reference material, workflows and connected tools. That makes it possible to create AI that is much more closely aligned to a particular area of work.
That could be something professional, such as a business workflow, or something much more specific. The value is obvious. Instead of explaining the same context repeatedly, you can give the AI a defined set of instructions and information to work from.
But there is an important catch. Just because you have given an AI system instructions, reference material or rules does not mean you should assume that every answer will apply them perfectly.
I’ve learned to treat governance as part of the design. That means checking whether the AI has actually used the relevant information. It means questioning an answer when something looks inconsistent. It also means asking it to confirm what it relied on before taking an important recommendation at face value.
The more capable and customised these tools become, the easier it is to place greater trust in them.
That makes good governance more important, not less. For me, the useful question is not simply: “Can AI do this?” It is also: “How do I make sure I can trust the way it is doing it?”
3. Ask AI how long the solution will actually take
This is probably one of the most practical lessons I’ve learned. ChatGPT can make a technical solution sound very simple. You describe what you are trying to achieve and the response can come back almost immediately: Do X. Connect Y. Configure Z.
It can feel like you are a few steps away from the finished solution. Sometimes you are. Sometimes you are absolutely not. I ran into this while looking at a way of connecting one of my customised AI workflows with a spreadsheet. The objective was for the system to carry out research and then write structured information back into the spreadsheet.
Initially, the suggested implementation sounded relatively straightforward. Once I looked into it properly, the route was considerably more involved and the eventual solution was different from the original recommendation. Had I simply followed the first answer, I could have spent considerably more time and money going down the wrong route.
That changed how I approach technical recommendations from AI. Now, if ChatGPT suggests an implementation, I also ask: “How long do you realistically think this will take?” Then: “Break the implementation down into its individual stages.” And, where the decision matters: “Research whether this is actually the best way to achieve it.”
This is particularly important when AI is being used to support digital, data or technology work.
A solution that sounds simple in a conversation may involve authentication, integrations, infrastructure, permissions, testing, security, data structure and several other dependencies once you start building it.
Before committing meaningful time or money, understand the likely delivery effort.
AI is most useful when you challenge it
None of these lessons make me less positive about AI. Quite the opposite. The reason these things matter is because the tools are useful enough that it is very easy to start relying on them. I use ChatGPT because it can accelerate research, challenge thinking, help structure problems and make complex work easier to explore.
But I get the most value from it when I treat the conversation as exactly that: a conversation.
Ask questions. Challenge assumptions. Request evidence. Test the proposed approach. And when an answer sounds unusually simple, ask what sits underneath it. AI can help you get to an answer much faster. Knowing when to question that answer is becoming just as important.
