Start building your own chatbot now >

If making a chatbot is quite easy, finding a great one is much more complicated due to poor user experience (UX).

Where web pages and apps have graphic elements to guide the user, bots are living inside chats and aren’t always easy to understand in the blink of an eye. Worse, as bots are still a relatively recent apparition, people don’t have a clear mental model of what chatbots are supposed to do and how to use them.

These days, more and more design rules are emerging. You will find here a few guidelines we have discovered at Recast.AI while crafting bots. Let’s start by an overview of the strengths of chatbots, and how sometimes they are not up to our expectations.

The good, the bad and the ugly

Chatbots, when well designed, have some serious strengths!

First and foremost, they scale incredibly well, and can handle thousands of conversations without too much sweat… Can you even imagine yourself having five separate conversations with five different people ? At the same time ? That makes them the perfect candidates for tasks with predictable and well-defined conversation flow.

They also shine by their versatility: you can find chatbots in a lot of different fields and areas, from medicine to hospitality, not to mention fitness  or support. Applications are endless! How convenient they are, ready to answer any queries at any time! And they’re as close to us as they can be: messaging apps are really trending with teens, who spend more time on them than on social websites. Bots are definitely finding their place in this texting world.


But sometimes, they aren’t up to our expectations. This can happen for various reasons, one of which is limited conversational capabilities.

Users are not always kind to their bots, and an inability to properly understand sentences tends to generate a lot of frustration, especially if there are no fallbacks to lower the friction. Some bots end up failing because they try to be everything for everyoneThe time for Jarvis from ‘IronMan’, or for Samantha from ‘Her’ has not come yet.

For now, it’s hard for a bot designed to manage pizza orders  to handle other tasks as effectively, like, let’s say, organising your calendar. It’s better to build specialised bots targeting a specific goal.

A good conversation goes a long way

Designing the conversational flow of a bot is absolutely crucial. It doesn’t matter how friendly a bot is, if the conversation is a labyrinth that gets the user lost or buried under a load a questions, the bot will end up talking to no-one but himself. Which, you might agree, is quite sad.


Chatbots have to be built around one or a few specific tasks: ordering a bunch of flowers or packing your suitcase according to next week’s weather. Those tasks must be at the center of your attention when you are imagining the conversation. It should be quick for someone speaking with your bot to get to that goalThis can be achieved by programming your bot in a way that at each step, it will gently and proactively guide the user on the right path. It will result in a smoother conversation, with a more natural feeling.

It’s almost mandatory, especially in business use, to make human intervention possible in the conversation. There could be various triggers:

– if the bot doesn’t understand what the user says twice in a row

– if the user asks to speak to a human interlocutor directly

– where a human decision is required

Nothing is worse than a frustrated user facing a dumb-looking bot that repeats the same sentence again and again.

Make it pretty, make it smart!

Now it’s time to focus on the bot himself. A bot has to clearly explain what he can do at the very beginning of the conversation.

As I mentioned earlier, a bot has no graphic elements to catch the eye, the initial presentation has to be crystal clear. The last thing you want is for the user to end up asking for something your bot is not designed to handle, or worse, freezing and staring at the screen not knowing how to start.

As a chatbot lives within a … chat, the introduction has to be brief and simple, a few lines max. If you’ve given your bot some memory, it can be a good idea to reduce the amount of help provided as you go. A user who has already ordered thirty pizzas might get annoyed if the bot keeps on starting with the same old sermon.


When something goes wrong, like if your bot doesn’t get the meaning of a sentence, or if the user’s response is not what was expected in the conversational flow, your bot has to be helpful, and provide some clues on what’s going on and on how to get back in the right track.

What about channels ?

A lot of different channels are available today, from Kik to Discord, through SMS and Messenger. Each have specificities, like the addition of buttons and carousel, which complement the conversational interface well, and bring an interactive feeling. There’s no good reason not to use them: some platforms provide a very diverse toolbox, and sometimes, a simple button works better than a whole line of text.

But wait, what about the future?

So what’s next ?

Creating a bot these days is like going on an adventure. There are a lot of experiments to be done, and the ecosystem is evolving rapidly.

Something I will definitely dive into is the chatbot natural evolution: the voice-controlled chatbot. It’s a big step, a new journey, with new challenges. Imagine a bot plugged in a channel full of people: how can you understand that someone is speaking to you, and how can you respond knowing that there is a lot of background noise? Should you choose a male, a female, or a robot voice ? There’s so much to research, so much to try out, it’s only the beginning. To be continued… 😏


Want to build your own conversational bot? Get started with Recast.AI !

Subscribe to our newsletter