Yesterday, I asked ChatGPT a simple question. But it decided to jazz up its answer.

I simply asked: What’s the best temperature to cook bacon on the stove?

I expected a straightforward answer. I intentionally went to ChatGPT because I didn’t want the classic food-blog experience where you scroll through ten pages of someone’s childhood memories describing how their grandmother’s house smelled like bacon on Sunday mornings — all while dodging ads that take over 90% of the screen. I just wanted the temperature.

But ChatGPT had other plans. It responded:

“Think of a shy actor stepping onto a warm stage. If the lights are too hot, it panics and burns. If they’re too cool, it sulks in its own fat.”

WTF?

I stared at the screen, confused, and double-checked I had actually opened ChatGPT. I had expected a straightforward, factual answer, like normal.

Instead, it felt like I requested a Waymo, but got a chatty guy in a yellow VW Bug blaring psychedelic surf-rock.

What’s actually going on

Later, I saw an announcement from Fidji Simo (CEO of Applications at OpenAI) explaining that ChatGPT now offers personality modes, like professional, quirky, nerdy, friendly, etc.

Voice assistants like Siri or Alexa, in-car navigation systems, and even ChatGPT voice mode have all offered alternate voices for years. But those were explicit, opt-in choices. Users could pick an alternate voice, but knew exactly what they were getting.

What felt so jarring here is that ChatGPT has had the same personality for years. While it’s voice has maybe gotten a bit warmer over time, it was still clearly machine-like and neutral. But then overnight, it became a theater kid.

If your spouse woke up one morning with a totally different personality, you’d have questions…

How products change personality

But this raises an interesting question: How can you shift a product’s personality without unsettling the people who use it?

Back in 2015, Facebook encountered a version of this. Brand sentiment was dropping and many users didn’t like or trust the product. Facebook had spent years shaping itself to be a neutral platform. It had no editorial voice or emotional tone anywhere. But this neutrality came at a cost — the product felt inhuman. So it was easier to blame, distrust, and dislike.

I joined a product team tasked with making Facebook more human and more likeable. So we gave Facebook a warmer voice and brought this voice forward to celebrate happy moments like birthdays, holidays, and memories. We acknowledged holidays and joined in celebrations with simple messages like:

“Wishing you a happy birthday from all of us at Facebook.”

“Here’s a memory we hope you enjoy.”

“Leena and Nabil have birthdays today. Wish them the best!”

Now, while we experimented with volume, Facebook’s newfound voice didn’t show up everywhere. The adjustment wasn’t perfect, but there was no personality setting or dial for users to adjust. The product’s voice just gently shifted in moments where a human-sounding message felt natural.

That subtlety mattered. The product didn’t suddenly sprout a personality, but it did start to feel a little less stark and a little more human.

Modeling humans

Humans shift tone intuitively. We don’t speak the same way when joking with a friend about a wild night out in college vs sharing a cancer diagnosis. We each have our own stable personality and we modulate our tone based on context.

Choosing a personality from a drop down is a reasonable first step. But ChatGPT’s new personality modes operate like a fixed character, showing up the same way across all contexts without understanding social queues.

The real opportunity is to make AI products that feel natural by modeling humans — and dynamically shift tone based on the context of the question and conversation. In this world:

• Quick factual questions get crisp answers.

• Emotional conversations get warmth.

• Creative prompts get big energy.

Not because you toggled a setting, but because the AI understood you and the moment.

Just for kicks, here are a few snippets of conversations from the day ChatGPT was all about theatrical answers: