The most satisfying bots are ones with clear uses, such as the Whole Foods one on Messenger that prompts users for an ingredient then offers recipes, or the Sephora bot on Kik that offers beauty tips based on skin type.
One of users’ big problems with bots is knowing what they can do. Google Allo solves that with messages that introduce features and capabilities over time. If a query is too much for the bot, it will say the feature isn’t available yet, and suggest ones that are.
CNN’s Messenger bot, which was originally designed to answer general questions about the news, often failed to grasp what users were asking. Without any conversational perimeters, Microsoft’s Tay bot on Twitter went quickly off the rails, mimicking trolls’ racist language. Bots are better when their interactions are constrained to a few choices and quick exchanges with guided prompts.
Because the personality of a bot molds the user experience, some of the best, such as the Poncho weather bot for Messenger, take special care to communicate in chummy language—the equivalent of a brightly colored and inviting interface.
A version of this article appeared in the November 2016 issue of Fast Company