The Next Phase Of UX: Designing Chatbot Personalities

When the conversation is the interface, experience design is all about crafting the right words.

You may have heard that “conversational interfaces” are the new hotness in digital product design. Why open and close a bunch of apps on your phone to get stuff done when you can invoke a text-message-like window and just say what you want done to a chatbot? Well, here’s one reason: what if the bot is annoying or tedious to talk to? In conversational UIs, personality is the new UX.


“We want people to enjoy dealing with our software, but now we have a very limited palette with which to design the experience,” says Ben Brown, co-founder of Howdy, a “digital coworker” chatbot that runs within the office communication tool Slack and automates things like project-status meetings and taking lunch orders. “It’s nearly the ultimate challenge for digital design, because in most cases, you don’t have control of what it looks like at all. How can you boil your entire app experience down into two lines of text? There’s nothing else on the screen but that.”

The technical effort of getting a bot up and running is already starting to get abstracted away. (Howdy just launched Botkit, a library of “building blocks for building Slack bots.”) What’s left is figuring out how the thing is going to behave in a real-time conversation—in other words, its personality. Where does a designer even begin?

First, rethink “designer”
When pushing pixels no longer matters, the basic skill requirements for designing a bot become radically different. It’s no coincidence that both Howdy and—a startup whose bot intelligently schedules meetings via email—each hired writers with performing-arts backgrounds to help define the UX of their products. Howdy, whose bot affects a casual and lightly irreverent tone (much like that of Slack itself), has novelist/satirist/former-improv-comedian Neal Pollack on the product-design payroll. Anna Kelsey, who was hired right out of Harvard as’s first “AI interaction designer,” studied folklore mythology and directed theater. “The whole idea of creating a character, and thinking very technically about the way specific words or groupings of words can make people react and respond, is something I thought about all the time in college,” Kelsey says.

Authors and performers have been conjuring up convincing artificial personalities for millennia, so it makes sense that they’d be on the vanguard of designing this new kind of software interaction. But this new breed of UX design won’t be solely the domain of artsy types. “Think of the person who is writing the micro-copy around forms on a website,” Howdy’s Brown says. “All of a sudden they’re the king, because it’s nothing but microcopy now. That little form validation error message, or whatever, is now the full and total sum of your brand’s representation [in this interface].”

Consider the metaphor
So if bots and AIs are things that we talk to instead of “use,” what exactly are we talking to? This controlling metaphor is as fundamental to a successful UX as the notion of a “desktop” is to the graphical user interface. Is the agent on the other end of a conversational UI meant to be like HAL from 2001 or Samantha from Her—a disembodied-but-humanoid, quasi-omniscient, personalized god in a box? Or is it supposed to be more like R2-D2, or an extra-smart Roomba—dogged and resourceful within its limited domain, but nothing more?


The metaphor matters because it sets expectations. Facebook’s M, for instance—a personal-assistant chatbot that lives within Facebook’s Messenger app—wants you to treat it like a digital genie that can do anything from booking a table to buying a car. “We don’t want to restrict people from the types of things that they ask at this point,” says Jeremy Goldberg, a product designer on the M team. “It can be incredibly personal for people, and we want them to build a relationship with it.”

Unsurprisingly, M’s personality skews more toward “god in a box” mode: warmer than HAL (and less likely to turn homicidal, one assumes), but not as richly personalized as Samantha (you’re unlikely to become infatuated with something that has a neutered phonetic symbol for a name). Facebook’s apparent commitment to creating a genie-like UX is so strict that Goldberg wouldn’t describe how M responds when it can’t do something. It’s literally a yes-man, and acts the part—with human help, for now.

Howdy’s metaphor takes the opposite approach. “We really see Howdy as an astro-mech droid [from Star Wars],” Brown says. “R2-D2 is Luke Skywalker’s droid, but you can have your own—each with its own quirks and unique things that it’s learned and you’ve taught it.” To wit, Brown and his team have crafted a personality that’s partway between R2-D2 and TARS, the monolith-shaped robot from Interstellar. “A core element to our bot’s personality is this self-deprecating idea of, ‘There are a lot of things that are over my head,'” Brown explains. “If it doesn’t understand or isn’t capable of something, it’ll just say so. ‘I’m just a bot, I don’t even have arms’—that kind of thing.” (Brown is planning to implement a way to adjust the bot’s ‘humor setting,’ just like Matthew McConaughey does to TARS in the movie.)

When in doubt, tread with caution
Regardless of the guiding metaphor behind a bot’s personality, the basic patterns that govern its interactivity are still being worked out. Still, the folks in the trenches have discovered some surprising things to avoid. Ben Brown had to outlaw Howdy’s bots from asking rhetorical questions, “because people expect to respond to them, even though the bot was just being polite,” he says. “You would never just stick a form on your web page and not expect people to type into it.”

advertisement’s meeting scheduling bot (named Amy or Andrew depending on your preference), meanwhile, isn’t allowed to use any gendered pronouns. “You don’t want to burn a bridge with anyone by calling them a he when you’re talking to a woman, or vice versa,” Anna Kelsey says.

And of course, it’s always important to have a built-in kill switch for the conversation. Speaking rudely to Amy/Andrew or Howdy (e.g., “shut up”) will cause each bot to politely withdraw from further conversation. But if this new breed of UX designer is doing his or her job right, you won’t want to pull the plug. “People love messing with robots, but on the other hand, people constantly send nice notes to Amy,” Kelsey says. “The software must be doing a good job, because people know it’s a robot, but still feel the need to say ‘thank you.'”


About the author

John Pavlus is a writer and filmmaker focusing on science, tech, and design topics. His writing has appeared in Wired, New York, Scientific American, Technology Review, BBC Future, and other outlets.