As conversation designers, it’s important to acknowledge our limitations. There’s pretty much no way we can make chatbots as responsive and intuitive as real human beings. In fact, many users expect the exact opposite from virtual chat experiences. When I describe my job to family, friends, or acquaintances, they often share how frustrating and useless they find chatbots to be. They see IVAs as rigid, unhelpful, and largely unable to answer the specific questions that they pose. 

Solid natural language processing capabilities are key to addressing these problems. However, there are small steps that conversation designers can take to reduce the friction and frustration that chatbot users often feel. Even if a bot is unable to respond to questions with the accuracy or flexibility of a real person, simple language tweaks can make a conversational experience much more fluid, gentle, and pleasant.

With this in mind, consider following these guidelines while designing your next IVA.

1. Don’t Overpromise

While designing conversations for a chatbot, it is important to be upfront about what an IVA can and cannot do. Especially in situations where a user’s mental or physical well-being may be at stake, it’s crucial not to promise support a bot cannot give (and then, of course, to provide higher-touch alternatives for users who are really struggling.) Even in lower-stakes use cases, it’s important not to promise more than you can deliver – in other words, to make users believe that the bot can understand or assist them in ways that it actually can’t. 

One key way to avoid this is to set expectations from the very first message. When your bot introduces itself, make sure to mention its name (if any), a few points about what it can do, and then a direct question about what the user wants to do next – preferably one that guides them down a fully fleshed-out conversational path. And if your bot slips up, don’t hesitate to remind the user that it’s not human – it faces limitations, but is continually learning. Especially in moments where users might feel disappointed – like bot breaks or instances of negative feedback – it’s better to admit shortcomings than to try (and fail) to maintain the illusion of human intelligence. 

Example #1: Imagine you’re crafting an initial message for a bot that teaches people how to use cleaning products.  

Example #2: Imagine you’re crafting bot breaks for the same bot (a “hygiene advisor” that provides information on a specific line of cleaning products.) 

2. Craft Natural Quick Replies 

Quick replies are a handy way to guide users down conversational paths that your bot is capable of handling. However, if quick replies aren’t carefully worded, these pre-set paths can become rigid, awkward, and uninviting. To remedy this, take care to model your quick replies on natural, everyday speech. In other words: make sure they sound like something a real person would say in response to the preceding prompt or question. Stick to replies that are speech-like and direct, rather than long, technical intents that likely wouldn’t show up in a real-life conversation. This can make a guided, pre-set conversational path feel natural to the user, even within highly structured journeys like FAQ bots.

Example #1: Imagine that the bot asks the user, “What would you like to learn about this product?” 

Example #2: you can also rephrase the initial question in order to make quick replies seem more conversational. 

Example #3: Re-phrasing the initial question can also help you avoid long, technical quick replies, which can be tough for users to parse at first glance (and also hard to fit within character limits.)

3. Avoid Repetition

Sometimes, development constraints make repetition unavoidable. Conversational paths often need to loop back to previous nodes, which can’t always be altered for the sake of variety. However, if possible, consider altering the language in your design so that the same words and phrases are not repeated over and over again. This can be achieved by creating follow-up messages that are differently phrased than initial messages, or by implementing variants. Essentially, variants are slightly different versions of the same message that get randomly served to users as they encounter them again and again. They are especially useful when added to messages or flows that users may see repeatedly, such as feedback nodes or main FAQ menus. This, along with scanning through your design to catch repetitive words or phrases, will do wonders to make the experience feel less robotic and more natural. Real people vary their phrasing in subtle ways as conversations proceed, so bots should too, where possible. 

Example: Let’s say the user encounters an initial main menu-type message that reads “What would you like to learn about today?” The answer selects “Canceling a reservation,” reads the answer, and then indicates that they have understood the information. 

4. Give Users an Out 

The average chatbot user doesn’t have time or energy for a long, drawn-out interaction. But if they want or need to leave a conversation ASAP, the way out isn’t always clear. Many users end up saying goodbye or closing out of the chat window before the conversation reaches a clear end, which can feel abrupt and dissatisfying. To prevent this, consider giving users opportunities to exit the conversation at natural stopping points. This is a decidedly more natural, pleasant experience – like saying a quick goodbye to a friend instead of abruptly hanging up. Of course, if you do decide to add elective stopping points into your conversation, it’s important to make sure that they segue smoothly into whatever pre-set outro is built into the bot, whether that’s a feedback node or a final question, like “Is there anything else I can help you with?” 

Example: Let’s consider the same example as above– an initial main menu message that reads “What would you like to learn about today?” with 4 quick reply options (“Canceling a reservation,” “Booking a reservation,” “Purchasing a gift card,” “Hosting an event.”) Imagine that a user ONLY wants to cancel their reservation, and isn’t interested in any of the other topics. 

5. Don’t Mind-Read

A key step towards designing effective, enjoyable conversations is to imagine yourself in the user’s shoes. However, don’t assume that you know how they will feel at each stage of the conversation. If your bot makes a statement that assumes a certain emotion or reaction on the part of the user, it may be wrong – which, of course, is a key giveaway that you’re speaking to a bot, rather than a human. That’s why it’s best to stick to broad, neutral responses when you can’t be totally sure of a user’s feelings or attitude. This is especially important when a user has entered a free-text statement versus a quick reply – you never know what they might say, so take care not to assume anything in your response. 

Example: imagine that you’re working on a bot designed to help frequent travellers manage flying anxiety. The bot has guided the user through a breathing exercise and is now wrapping up the conversation. 

6. Put some heart into your bot breaks 

If you’re like me, you often add bot breaks into your design at the very end, when time and patience are running low. It’s tempting to write these types of messages as quickly and efficiently as possible, but this can yield curt, overly brief copy that will probably make frustrated users feel even worse. It’s not easy or fun, but if you can, try to imagine that you’re speaking to someone you really care about, and that you actually feel bad that you’re unable to help them. Do your best to write bot break messages that are genuine, sincere, and helpful (of course, without rambling or becoming overly apologetic.) Receiving a message like this instead of a cold, cookie-cutter response just might make a difference in the user’s overall experience, even if they didn’t get the answers they needed. 

This is especially important if you’re writing a bot for a client that has specifically requested a formal and/or friendly tone. A bot’s personality is often most evident when it falls short of user expectations, so it’s important to make sure that your bot breaks align with an overall attitude or persona.

Example: Imagine that you’re working on a bot for a hotel that will transfer users to a concierge if they’re unable to understand a request. The client has requested a bot personality that’s both formal and helpful, in line with their employee training guidelines.