Users Need to Know When They’re Doing Factual Research
I’d like to propose a UX improvement that focuses less on making ChatGPT “more accurate” and more on helping users understand what kind of interaction they are currently having.
This idea came from a real conversation with ChatGPT today.
I asked about a recent political event.
ChatGPT first gave a correct answer. It correctly identified the event, quoted the relevant social media post, and explained the surrounding discussion.
A few minutes later, in the same conversation, it confidently mixed this event with a completely different one.
I pointed out the inconsistency.
Later in the same chat, something similar happened twice again. ChatGPT confidently drew conclusions about an OpenAI community forum and about my account status that turned out not to be correct.
None of these mistakes were malicious or absurd.
They were plausible.
And that’s exactly why I think they deserve attention.
The issue is not simply that ChatGPT can make mistakes.
The issue is that users often don’t realize when they have moved from creative assistance into factual research—a situation where plausible mistakes can easily become misinformation if they are accepted without verification.
The Current Disclaimer Doesn’t Describe the Real Risk
The current disclaimer:
“ChatGPT can make mistakes. Check important information.”
is technically correct.
But I don’t think it describes the actual risk.
It suggests that mistakes are isolated errors.
What I experienced today was different.
The model repeatedly filled gaps with plausible assumptions and presented them with confidence.
As someone who already knew the topic, I noticed.
Many users wouldn’t.
That is a fundamentally different kind of risk.
My Proposal
Rather than trying to label individual statements as “facts” or “not facts,” I would propose something simpler.
ChatGPT could indicate the current interaction mode.
For example:
Creative brainstorming
Writing assistance
Factual research
Analysis & interpretation
Learning & explanation
The mode could change automatically as the conversation evolves.
Why I Am Not Suggesting Fact Labels
I am not asking ChatGPT to certify individual statements as true.
That would create another layer of potential mistakes.
Instead, I think the interface should communicate what kind of collaboration is currently taking place.
For example:
Interaction Mode: Factual Research
This conversation currently contains factual claims and interpretations. Important information should be treated as something that can—and where appropriate should—be verified.
The goal isn’t to reduce trust.
The goal is to encourage appropriate trust.
Why This Matters
Most users don’t consciously distinguish between different ways of using ChatGPT.
They simply ask questions.
Sometimes they are brainstorming.
Sometimes they are writing.
Sometimes they are researching.
Those situations deserve different expectations.
Helping users recognize the difference could:
- improve transparency,
- encourage appropriate verification,
- strengthen media literacy,
- reduce the risk of confidently presented misinformation being accepted,
- and improve trust through better expectation management rather than stronger warnings.
A Personal Observation
During the conversation that inspired this proposal, ChatGPT itself eventually recognized the underlying pattern.
The mistakes were not primarily caused by missing knowledge.
They were caused by reasonable-sounding assumptions being presented as established facts.
That made me realize something:
The biggest challenge may not be making AI incapable of making mistakes.
It may be helping users understand what kind of interaction they are currently engaged in, so they know when a response is primarily creative and when it should be treated as factual information that deserves verification.
I believe making the nature of the interaction more transparent could be just as important as making the answers themselves more accurate.