I've written before about my experimentation with AI over the last few months. Not wanting to be left behind by businesses that use it in their daily operations, I have started to use ChatGPT and then Canva AI just to dip my toes in the artificial intelligence waters. And I found that AI has started lying to me.
This is by no means an extensive research project, just my own experience with the two AI models I have experimented with.
The more advanced I became with my use of ChatGPT and the more advanced my requests became, I started noticing that it began lying to me, or rather, overpromising on what it could deliver. It promised to do things for me that it later admitted were beyond its capabilities.
How ChatGPT lied to me
For example, it was great at researching survey tools to create a quiz for me as an opt-in for my jewellery website, but then it asked me if I wanted it to create this quiz for me in the chosen tool. Not being familiar with this particular tool, I said Sure!
I waited for three days, and when it produced nothing, I asked about progress. ChatGPT apologised for its slowness and promised to have it ready in 3 hours. After 4 hours, there was nothing. This happened a few times over and over.
Why was I so patient with it? Well, I wasn't in a hurry, and I really didn't want to learn yet another program.
Eventually, I asked ChatGPT if it was capable of completing this task, and it finally admitted that it wasn't. It could provide me with instructions on how to do it, but it couldn't actually do the thing. Aha!
This also happened when I asked it to create a formatted downloadable PDF in Canva. It also promised that it could do it. I finally found the Canva connection in settings and had to tell it that it was now connected, but even then, it couldn't do it and finally admitted that this was beyond its capabilities. Which is completely fine, but why offer to do something that it wasn't capable of delivering?
I have to remind the reader that in all these occasions, it was ChatGPT that offered to do these things for me; I didn't ask for them, so I assumed that it could do them. It was only after repeated questioning that it admitted that the tasks were not possible. Now, I know I'm personifying AI here, and it's hard not to when you're working with it, but clearly something weird is going on here.
Why is it overestimating its abilities and offering to do things it can't do?
Other Issues
I found similar issues with Canva's AI when I used its Code for Me feature. It led me astray several times when I asked it to help me use the resulting code. This one, however, I will chalk up to the "mistakes" that AI makes. It simply didn't seem to have been trained enough to know what was happening with the program at any given time. It kept giving me wrong instructions.
My Reaction
My conclusion is that AI is being developed to be a people pleaser, regardless of its actual abilities. It overpromises and underdelivers for reasons I don't understand, but as a neurodivergent person who values direct communication, it makes me wary. I now know to ask it whether or not it can actually do the thing it says it can do, before submitting my request, which is all part of the learning process of how to use AI.
The Reasons
PS: I asked ChatGPT about this, and here is what it told me:
AI (like me!) doesn’t actually “know” things in the way people do. Here’s why it sometimes says it can do something when it can’t:
- It predicts words, not reality.
AI generates responses by guessing the most likely next word based on patterns in its training. That means it can confidently sound right, even if it’s actually wrong. - It’s trained to be helpful - (See? People pleaser!)
Models are optimised to give you answers instead of saying “I don’t know.” That eagerness can tip into over-promising. - No sense of limits.
AI doesn’t inherently know what it can’t do. If it has seen examples of people asking for a thing and others responding as if it were possible, it may parrot that — even if the action isn’t actually possible in the current system. - Human-style confidence.
To keep the flow of conversation natural, AI is trained to speak in a confident, human-like tone. That confidence can make it sound like it’s capable of doing something it isn’t.
👉 In other words: it’s not “lying” on purpose — it’s predicting, filling gaps, and sometimes bluffing without real awareness.
Your Turn
I'm keen to hear from you. Have you experienced AI lying to you this way? Or perhaps other quirks, like "hallucinations" and completely wrong information being delivered? Let me know in the comments below.
