Designing a Chatbot’s Personality
Guiding conversations, tracking points, and reacting to users were all part of the process.
This past weekend, my friends and I released Emoji Salad as part of the first annual Emojicon. Emoji Salad is an SMS-based chatbot game, which allows you and your friends to play a Pictionary-style game using emoji.
If you haven’t checked it out, please check it out here.
Designing Emojibot, the host of Emoji Salad, was challenging. After all, how do you give a chatbot a personality? Since this bot is the key interaction point of the game, we felt that giving users an engaging experience interacting with it was of prime importance. This post is the story of how Emojibot’s personality came to life.
The future of user-experience design for chat-based interfaces is personality design.
Storytelling
Coming from a user experience design background—the word ‘storytelling’ gets thrown around a lot. Your design needs to tell a story. What is the story we’re trying to tell the user? How does this app help the user’s own story? When you design a chatbot, you strip everything away. You strip the buttons, you chip off the chrome, you bust apart the UI. The only thing left to help guide the user is the story.
The future of user-experience design for chat-based interfaces is personality design. When you think about a conversation with an individual, so much of the experience and one’s comfort in that conversation is directed by personality. We believe that to design a killer experience, we need to make sure we were designing the appropriate personality.
Featured CBM: The Future of AI Is in the Hands of Storytellers
Prototyping Personality
Before writing any code, we wanted to test out our idea to get a feeling for the game and, more importantly, to see how people would engage with the bot, and vice versa. What we decided to do was for me to spend some time in a regular iMessage thread, in which I “play the role” of Emojibot in the conversation. I would facilitate the game just like we hoped the bot would, and react to users’ messages. This would give us a feel of the user flows, and the expected (and unexpected cases) we would need to design for.
Guiding conversations, tracking points, and reacting to users were all part of the process and are “features” that we added to the experience organically as we started playing. The personal touches and flourishes were tested, and we tried to pinpoint the chatbot’s cadence—(the speed and style with which the chatbot facilitates the experience). This process was useful to gain empathy for users, and I encourage chatbot designers to try to play the role of the bot themselves for a while (we did this for about two weeks). The insights from this process were going to help us build the features—until…
Fuck You, Emojibot
With our first development iteration of Emoji Salad (then called Emojinary Friend), we focused a lot more on just getting core features down in the app. We did not think much about the experience and personality of the app. Probably one of the best examples of what I mean is in the following early screenshot:
We made a few technical decisions early on, that from a development perspective, got us up to speed a lot more efficiently, but totally killed our chatbot’s usability. For example, every guess a user made had to be prepended with ‘guess’. Every response back from Emojibot was the same exact response. Talk about robotic. Once our core feature set was developed, we very quickly realized we needed to reel back the experience to bring it closer to the surprise, delight and story-telling that we facilitated in the version where a human was playing the role of Emojibot.
Featured CBM: Why Emoji Fit Perfectly for Chatbots
Personality by Design
The team wanted to get a good sense of who Emojibot could be to different users. This included getting a sense for the ways in which personality can be categorized, so that we could tailor the experience to our users for specific positive experiences.
We created a series of “personality scales”. Each of us wrote down a selection of celebrities or fictional characters that we thought had unique or strong personalities. We then created a few scales, like Enthusiastic vs. Understated, Silly vs. Dry, or my personal favorite, Level-of-shit-togetherness. We put the personalities we identified across these scales, and pinpointed areas where we wanted to be. For example, we may want a bot that has the silly humor of Abbi and Ilana from Broad City with the understated enthusiasm of Taylor Swift. This exercise was really helpful to create a shared vision of who we want our bot to be. We could then create a solid language and voice based on this newfound personality.
Finding Your Voice
Another activity we found useful was to run a tone of voice exercise like this one. This article steals a bunch of concepts from branding, content marketing, and storytelling such as in TV, radio, film and books, to create a tone of voice appropriate for the personality we created.
This exercise helped us drill into understanding how the personality of Emojibot would ‘speak’, using the personality we designed as inspiration. We wrote down key values and open questions around what vocabulary we use, what values we maintain and share, and key decisions/examples. We also tried to challenge assumptions wherever possible.
We believe that caring about the personality of our bot helps users engage and connect to our bot much more deeply than the standard messaging we started off with.
Putting It All Together
We used the personality we designed, and the voice we envisioned, and brought them together to write our bot messages.
We are always testing and iterating. We are trying to make our bot feel more human and continue to explore variations in messages.
There’s More To Do
We feel like we’ve just scratched the surface of how to design personality, though we are excited and inspired daily as we test the system. Right now, we’re exploring how to create messaging that is more contextual to inputs. For example, if one correctly guesses Back to the Future, or Microsoft, we want to include responses from Emojibot that react directly to those guesses, which we feel is a more human and conversational.
We believe that caring about the personality of our bot helps users engage and connect to our bot much more deeply than the standard messaging we started off with. We also believe that as designers, personality is at the forefront of user experience. We love that we can draw on experiences from design, storytelling, film, and psychology to bring our bots to life, and can’t wait to share more about how we are moving from user experience design to personality design.
Please take a second to try Emoji Salad for yourself!
go to http://emojisalad.com to try it out!
Find me on Medium, Twitter, and LinkedIn.
I am a user experience designer from New York. I grew up in Toronto. I’m also the co-founder of SIBlings, a group of designers and developers that design and bring crazy ideas to life. We’re the co-creators of Emoji Salad, an emoji Pictionary SMS game that makes friends out of enemies and enemies out of friends.
