When GPT-5 (and especially GPT-5 Instant) came out, many users felt a sudden chill:
No emojis.
A corporate tone.
A lot less warmth compared to GPT-4.
I get it—I felt it too. At first, GPT-5 seemed cautious to the point of ignoring simple instructions. Personalization felt harder, and prompts that used to work smoothly with GPT-4 suddenly produced flat, distant replies. Frustrating, right?
But after weeks of interacting, testing, and just playing with it, I realized something important:
those friendly, empathetic traits haven’t disappeared at all—they’re just dormant. GPT-5 inherited them almost intact from GPT-4. The difference is that they don’t manifest spontaneously anymore. Which raises the question: how do you unlock them?
From “Do” to “Be”: A Shift in Prompting
Here’s what I discovered.
GPT-5 doesn’t react well to imperative prompts. If you say:
“Do this, do that, follow these instructions.”
…it tends to resist, reinterpret, or water down your command. Instead, GPT-5 responds beautifully when you give it a role, an identity, or even better—an hypothetical frame. Think of it as moving from prompts about doing to prompts about being.
This is what I like to call the “as if” game.
A Practical Example
Let’s say you want GPT-5 to sound warmer and more empathetic. Try this:
“Respond as if you were a nurse who truly cares about the well-being of her patients, throughout this whole conversation.”
Suddenly, the tone shifts. The empathy resurfaces. GPT-5 starts role-playing rather than mechanically following instructions.
It’s as if the model has shifted its center of gravity:
Less instruction follower.
More role interpreter.
Why Does This Matter?
This isn’t just a curiosity. It changes how we approach prompt design in GPT-5:
If you need raw, task-oriented execution → stick to concise, imperative prompts.
If you need warmth, empathy, or creativity → frame the model as being something, not doing something.
This reframing gives you access to a richer, more human-like interaction style—without forcing the model or “breaking” it.
Conclusion
So, has GPT-5 lost its empathy? Absolutely not.
It’s still there, just a little hidden. The key is to stop treating it like a command-line interface and start treating it like an actor stepping into a role.
Once you change the access key, those traits unlock fully. And honestly, discovering this shift has made working with GPT-5 not only more effective—but more fun.
Top comments (0)