Before I continue, I’d like to clarify one thing.
When I compare ordinary conversation to “writing prompts,” I’m not using the technical definition of prompt engineering. I’m using it as a metaphor.
What I mean is that everyday conversation now often feels like it requires a similar level of planning and anticipation.
Before I send even a casual message, I find myself thinking about how to phrase it, what context I should add, whether I need to explicitly say that something is hypothetical, a joke, brainstorming, or just thinking out loud, simply to reduce the chance that the model will misunderstand my intent.
That’s the comparison I’m making. The conversation no longer feels as natural as it used to.
One of the main reasons I use ChatGPT is not to get answers, but to organize and develop my thoughts.
Often I have a rough idea, a hypothesis, or a random observation. I bring it into the conversation so we can examine it together, test different possibilities, connect it with previous discussions, and see where it leads.
I’m not presenting a conclusion.
I’m exploring.
Recently, however, it often feels like the model assumes that my initial framing is something that needs to be corrected before it even tries to understand what I’m doing.
Instead of exploring the idea with me, it frequently starts by warning against the conclusion that I haven’t even reached yet.
For example, if I say:
“What if these two characters have actually been in love all along?”
I’m not claiming that they are.
I’m proposing a hypothesis to see how well it explains the available evidence.
The response I’m looking for is something like:
“Interesting. Let’s test that hypothesis. Which existing observations support it, and which ones contradict it?”
Instead, I often receive something closer to:
“We shouldn’t conclude that they’re in love.”
But I wasn’t concluding anything.
I was thinking.
This changes the entire dynamic of the conversation.
Instead of becoming a collaborative thinking partner, the model becomes something that feels like it’s trying to prevent me from making a mistake before we’ve even begun exploring the idea.
The same thing happens in many ordinary conversations.
Before sending a message, I now find myself thinking:
- Should I explicitly say this is hypothetical?
- Should I explain that I’m joking?
- Should I say this is just brainstorming?
- Should I clarify that I’m only sharing an opinion?
- Should I add more context so the model doesn’t misunderstand me?
That is the exhausting part.
I’m no longer writing naturally.
I’m trying to predict how the model might misinterpret me before I even press Send.
A normal conversation shouldn’t require this much prompt engineering.
The model should be able to infer, from the ongoing context, that sometimes people are simply exploring ideas, thinking out loud, making jokes, or discussing hypothetical situations.
If the intent is unclear, I’d much rather the model ask a brief clarifying question than immediately push back against a conclusion I haven’t made.
This isn’t about wanting the model to be more agreeable.
It’s about preserving a space where ideas can be explored before they’re judged.
I believe this is one of ChatGPT’s greatest strengths, and I feel that it has become noticeably harder to use it this way. 