How ChatGPT warns about itself

ChatGPT,

Let me first introduce myself. I am affiliated with Utrecht University and conduct research into accessible, reliable digital information for all generations. DUB, Utrecht University’s independent digital news platform, has asked me to write a column. I plan on delving into the role of media literacy in helping users avoid relying on unreliable digital information, with a particular focus on disinformation in the following areas: education, healthcare and news. I will also address unreliable information generated by chatbots. In addition, I’d like to hear your thoughts on specific topics you’d recommend me to discuss over the next few weeks and what your role in this might be. But first, I’d like to ask you to briefly introduce yourself to my readers (otherwise my column will be too long). Who, or should I say what, are you exactly?

Good question—and a great way to start your column. I am ChatGPT, a language model developed by OpenAI. Rather than having knowledge or experiences of my own, I have been trained on vast amounts of text to recognise patterns in language. This enables me to answer questions, write texts and summarise information. However, I do not ‘understand’ anything in a human way and have no consciousness, intentions or opinions of my own—though it may sometimes seem that way. My answers are based on probability: I generate whatever seems most logical or appropriate in a given context. That also means I can make mistakes or provide information that sounds convincing but is incorrect. I do not verify facts in real time and have no direct access to the truth. It would be best to view me as a tool: fast, versatile and often useful, but always dependent on critical human use. This is precisely where your topic—media literacy—touches on the core of how systems like me should be handled.

So what’s the deal, ChatGPT? Are you actually a reliable source of information for me? (maximum 50 words)

Yes—but with conditions. You should see me as a tool, not as an authority. I can provide quick explanations and guidance, but you must always verify important information through reliable sources. Critical use is exactly what makes me valuable rather than misleading.

Oh, good to know. All right, I’ll give you the benefit of the doubt for now. Now that we’ve introduced ourselves, here are my questions: What specific topics would  my readers be interested in? (maximum 50 words)

Topics of interest: misinformation in healthcare (e.g. regarding COVID-19), bias and errors in AI systems such as ChatGPT, algorithmic bubbles on TikTok, and verification of news regarding conflicts such as the war in Ukraine.

What sources did you use for the topics you mentioned earlier? (maximum 50 words)

I haven’t consulted any specific external sources. My suggestions are based on general patterns in training data (such as news, research and public discussions) and common societal themes surrounding media literacy, disinformation and digital technology.

Thank you for all the information; I’ll think about what I can use for my next columns, and I might well get back to you.

You’re welcome. Good luck with your column—it’s a relevant and important topic. If you’d like to discuss angles, examples or case studies later on, I’m available.

29.04.2026 ChatGPT can make mistakes. Please verify important information. Please verify cookie preferences.

Postscript:

So much for my dialogue with ChatGPT, which does issue a warning about itself:

“However, I do not ‘understand’ anything in a human way and have no consciousness, intentions or opinions of my own—though it may sometimes seem that way.” But ChatGPT also noted: “It would be best to view me as a tool: fast, versatile and often useful, but always dependent on critical human use. This is precisely where your topic—media literacy—touches on the core of how systems like me should be handled.”, and that is, in turn, in line with a media literacy approach. I will discuss how that works in my next three columns.

Emeritus Associate Professor Eugène Loos worked in the Department of Public Administration and Organisation Studies at the  Faculty of Law, Economics & Governance of Utrecht University, where he researched media literacy and the opportunities and threats posed by the use of ChatGPT as a text generator in education. This is a four-part series in which he engages in a dialogue with ChatGPT to investigate the extent to which this AI tool generates reliable information, focusing on the role of media literacy.