I learned something surprising after building an AI progress bar for ChatGPT, Gemini, and Claude.

When I started building AI Status Monitor, I thought the hardest part would be estimating progress.

It wasn't.

The hardest part was realizing that people don't actually care about the percentage.

They care about uncertainty.

When ChatGPT is thinking for 30 seconds, users start asking themselves:

Is it frozen?
Is it still processing?
Should I refresh?
Did something fail?

The interesting thing is that none of the major AI providers expose their internal progress.

So I couldn't build a "real" progress bar.

Instead, I had to infer what was happening from visible UI signals and estimate:

Current stage
Remaining time
Confidence of the estimate

It's obviously imperfect.

But I realized something while testing it:

People don't necessarily need perfect accuracy.
They mostly want reassurance that something is happening.

That completely changed how I thought about waiting UX.

I'm curious whether others building AI products have noticed the same thing.

Do users actually want accurate progress?

Or do they mostly want feedback that the system is alive?