Ever had one of those conversations with a friend where the tone changes halfway through, and by the end, you’re talking about something completely different from where you started?

That’s the magic of human conversation—it’s unpredictable, flexible, and, let’s be honest, sometimes a little messy.

Diana Deibel and Rebecca Evanhoe, in their book Conversations with Things, get it. They remind us that human conversations are full of these beautiful nuances that bots just don’t get. Bots are rigid. They’re programmed to follow a script, and when something throws them off? Well, things get awkward—fast.

Linguist and conversation designer Mary Constance Parks puts it perfectly: “Conversations are co-created. They unfold. They’re messy.” And that’s where conversational AI falls short. When a bot forgets what you said, gives you a totally weird response, or doesn’t catch a simple social cue, it’s clear—bots just aren’t keeping up with the complexity of human conversations. Ugh, so frustrating, right?

Here’s where it gets tricky. When you’re designing a chatbot, you’re not just thinking, “How do I get this thing to give the right answers?” Nope. You’re thinking, “How do I make this thing act like it understands people?” And that’s hard because, unlike us, bots don’t understand context the way we do. They don’t have that brain magic to adjust tone, remember subtle cues, or pick up on what’s left unsaid. So when a bot messes up, people really notice. Cue the sighs and eyerolls.

Another challenge? There’s no one-size-fits-all code for politeness—either for humans or bots! What’s polite in one culture might come off as cold or formal in another. So, it’s our job to figure out the tone—should the bot say “please” and “thank you”? Should it call you by your first name or go with “Mr./Ms.”? Is it using usted or tú in Spanish? These tiny decisions change how users feel about the bot—do they trust it? Is it likable?

Diana and Rebecca point out that your chatbot’s personality shows through in how it talks—its word choices, tone, and behavior. But here’s the thing: users know they’re talking to AI. So, while we’re designing these bots to be relatable, we can’t pretend they’re human. It’s a balancing act—friendly and helpful, but still clearly a bot.

And if there’s one big lesson I’ve learned? A bot is never really finished. You can’t just slap some code on it and call it a day. Conversations are living, breathing things, and your bot needs to be ready to learn and adapt—just like a real conversation with a friend.