I Wrote the Conversation Before I Wrote the Code, and It Saved Me a Rewrite

Last year I built a small chat feature for a side project. I started the way most of us do: pick a library, drop a chat bubble on the page, wire up some responses, and figure out the “conversation” part later.

Later never went well. The UI worked fine, but the conversations felt awkward. Users asked things I hadn’t planned for, my replies were either too long or too vague, and I had no idea how to handle someone who just typed “help.”

So I scrapped it and did it the other way round. I wrote the conversations first, on paper, before touching any code. Here’s what I learned.

1. Write three real dialogues first

Not flow charts, just dialogues. Pick a user, pick their goal, and write both sides like a script. It sounds silly, but you’ll find gaps within five minutes. Things like “wait, what does the bot say if they don’t know their order number?”

1. Plan the “I don’t get it” moment

Every conversational interface fails sometimes. What matters is how it fails. A good fallback does three things: admits it didn’t understand, offers two or three concrete options, and gives a way to reach a human. A bad one just repeats “Sorry, I didn’t get that” in a loop.

1. Keep replies short enough to read on a phone

If a reply needs scrolling, it’s probably a help article pretending to be a message. I try to keep each message to a couple of sentences and put the long stuff behind a link.

1. Don’t pretend to be human

People forgive a bot that’s upfront about being a bot. They don’t forgive one that pretends and then gets caught. A friendly tone is great, fake personality isn’t.

1. Test with someone who didn’t build it

I asked a friend to try it with no instructions. Within two minutes they typed something I’d never considered. That one test found more problems than a week of my own clicking around.

What I’d do differently

I’d start even smaller. One conversation, one goal, shipped and tested, before adding a second. I also wish I’d logged the messages people sent that I couldn’t handle, since that list turned into my to-do list.

If you’re curious about conversation-focused projects, I’ve been following what the team at Talkative Labs is working on. It’s a good place to see how others think about this space.

What’s the weirdest thing a user has typed into something you built? I’d love to hear it in the comments.