Those fearing for our robot overlords might worry that AI is too smart and powerful for their own good. But anyone who’s ever talked to a chatbot knows that they’re not perfect. Chatbot error messages are inevitable, especially when a user talks to a bot for the first time.
But how a bot handles error messages has a big effect on chatbot UX. When we talk to a chatbot, we want it to understand us; if a bot simply repeats the same error message over and over without any hint on how to set things right, it’s easy for a user to simply walk away. Think about a friend who clearly doesn’t hear what you’ve said and responds with a non-answer — you’re probably not going to invite them out to your next brunch.
So what are the chatbot guidelines for error messages? With these chatbot best practices, you’ll be able to craft the perfect error messages that steer the user exactly where they need to go.
Chatbot Error Messages Must Clarify the Issue
It sounds obvious, but chatbot error messages should, well, describe the error to users. How rare this happens might surprise you. The worst offender? Chatbot who ask the user to repeat themselves without providing any information on how or why it should rephrase the query. Let’s say your chatbot is asking for a city, but the user responds with something else. Rather than say “I didn’t get that. Please try again,” chatbot error handling should reiterate that it needs a location.
In less specific instances — if the chatbot simply became confused by a user trying to push its limits — the chatbot might remind the user of what it can do. User expectations should be made clear while onboarding, but it’s possible they might need a refresher when they try to perform a task that the bot simply isn’t prepared for.
Give Users an Out
But what if the user did provide information the chatbot asked for, but still ran into an error? In that case, we’re back at square one, stuck in a vicious cycle of error after error — but there’s a solution. Chatbot UX guidelines generally suggest providing options — contextual or persistent — to steer the conversation.
An easy way to follow this chatbot error handling tip is to direct users to a help option: “If you need help, just say ‘help.’” Another great chatbot UX tip is to provide options that fit the context of the conversation step. This gives the user a little more guidance than simply asking them to repeat themselves. Failure to provide a “safe route” might prompt users to exit the conversation altogether.
Don’t Blame the User
Perhaps one of the most important of the chatbot guidelines for error messages is not to not play the blame game. Miscommunications happen, and often no one is at fault. Imagine if a barista accidentally misunderstood something — a choice in milk or sweetener, perhaps — but reacted with a hint of passive-aggressiveness when you brought it up. This is no way to treat a customer, and no way to treat a user either.
If the Hypebeast bot on Facebook Messenger encounters an issue, it will open its error message with “We didn’t quite catch that…” before suggesting search terms and queries. While this isn’t a flat-out blame, the phrase does carry an aloofness that suggests the user did something wrong. While the tone might fit the brand voice, be careful about how you word your chatbot error handling. A good alternative example might be: “Sorry! While I’m skilled at [use case], I’m a bit daft on other things. Here’s what I can help you with:” followed by a menu of suggested queries.
Have Variation for Better Chatbot UX
A common example of chatbot best practices is to offer variations of all messages: greetings, small-talk responses and (you guessed it!) error messages. No one wants to encounter the same error repeatedly. This is disengaging and robotic — two of a chatbot’s biggest sins. If the bot suddenly appears robotic when struggling to understand the user, it decreases user confidence in the bot. With the chatbot best practices above, then, you should compose several error messages. You may then randomize them when the user runs into a general error.