Dora the Explorer is a better conversationalist than most AI.

Not because she’s deep. Not because she has better answers.

Because she waits.

She asks a question, pauses, and gives you actual space to respond. The whole show is built around the idea that an answer needs time to become an answer.

That sounds like a low bar. It isn’t.

I recently talked to Professor Hansun Zhang Waring, a conversation analyst at Teachers College, Columbia University, who studies what makes human conversation actually work. She and her colleague have been researching AI conversation specifically, trying to figure out what it’s missing. Their answer came down to three things.

What AI Can’t Do

The first is emergence. Human conversation is built in real time, bit by bit. Each thing you say shapes what comes next. AI conversation doesn’t work like that. It arrives pre-assembled, like an Amazon package. The box is already sealed before you open your mouth.

The second is recipient design. When a person talks to you, they’re talking to you specifically, in this moment, in this context. AI can personalize, but it doesn’t know you the way someone in an actual conversation knows you.

The third is indexicality. Every exchange depends on context: what just happened, what both people already understand, what doesn’t need to be said out loud. AI doesn’t have that shared background. It’s talking to everyone and no one at the same time.

You don’t notice what makes a conversation human until something almost-human gets it wrong. We’ve been having human conversations our whole lives without ever having to name what makes them human. AI forced us to figure it out.

What Actually Makes a Conversation Work

Dr. Waring also studies what makes conversation productive in any context. One framework she works with is called FAB.

F is for fostering. Creating space where people actually feel free to contribute. Not through constant praise, but through small things: eye contact, timing, the way you respond in the moment.

A is for attending. Actually listening to what someone just said instead of preparing your next point. Most people aren’t doing this. They’re already thinking about what they want to say next.

B is for balancing multiple demands. Knowing how to hold what you want to say alongside what the other person needs. Being fully present with one person without shutting everyone else out.

the ‘very good’ problem

Saying “very good” in a conversation, the thing we all assume signals encouragement, actually shuts it down.

A student gives an answer. The teacher says “very good.” The answer has been accepted. The moment is finished. There’s almost no room left to ask more, add something, or keep thinking out loud.

You see it everywhere once you notice it.

• A teacher says “very good.”
• An AI gives a complete, polished answer.
• A friend nods while already waiting for their turn to speak.

The exchange technically worked. But nothing new had room to happen.

We’ve been using “very good” as an invitation. It’s actually a period.

The Bigger Point

We built AI to sound like us. In doing that, we had to define what “like us” even means.

Emergence. Recipient design. Indexicality.

Three things we do in milliseconds, without thinking about them once.

Dora somehow understood the most basic version of this before most AI did: ask, wait, leave space.

But the more interesting question is not just whether AI can learn to have better conversations.

It is how often we are actually having them ourselves.

The next time you ask someone a question, notice the moment right after.

Are you really waiting for the answer?

Or are you already preparing what you want to say next?