Lekta’s Guide to Bot Design: Part 3, Chatbot Conversational Best Practices
At Lekta AI, we’ve been building chatbots since early 2017. While each project is unique, we found that certain universal principles can be applied regardless of the business domain, target audience or communication channel. Over the years, we’ve developed a battle-tested 4-step process that helps us deliver consistently great conversational experiences for our clients and their customers.
This is the third in a series of four articles on the topic that can serve as a comprehensive resource for anyone looking to build a bot. If you haven’t read Part 2, Chatbot Personality, go ahead and grab all the tips and tricks you can first. The persona you developed will come in handy now — it will be your guide and the go-to place for any questions you may have, such as “What would my bot say?”.
You’ve gathered requirements, developed your bot’s personality, now it’s time to write the dialogues! Providing a good conversational experience requires not only considering what we want to say but also how we’ll say it.
Good copy animates the chatbot and makes the conversation both light and effective.
There are two main things to keep in mind when developing copy:
- Chatbot has its own personality and tone of voice.
- Words matter!
In this article, you’ll find basic rules of copywriting for chatbots and some practical tips to follow if you want to create an effective and satisfying conversational experience.
We’ll assume that you already know what you want your bot to say, and we’ll focus on how to express these messages best. This will include not only the words you choose but their order, rhythm, and pace. All these elements influence the interpretation of the message and hence, users’ reactions.
For example, “Oh! We can fix that! Please call 123 123 123 and we’ll help you immediately” evokes a different emotion and reaction than “Please contact customer service to resolve this issue”.
Before we dive in, we just want to prepare you for the challenges you might face in the process of writing dialogues.
Turning your chatbot into a good conversationalist is easier said than done. It requires a lot of creativity, imagination, pre-planning, and testing. The responses you prepare have to be broad enough to work in many situations, yet specific enough to make sense and advance the customer journey.
An effective dialog design requires a holistic approach and a wide range of competencies:
- Copywriting skills
- Psychological understanding of human behavior and interaction patterns
- UX skills
- Strong commitment to testing and iterations
- A good understanding of the technology we work with is a must!
To warm-up, you might want to start with the happy path developed during the discovery phase and turn it into a dialogue. List out all the information you’ll need from the user to complete a business process, write a sample script and test it out with a colleague. Quite soon, you’ll realize that you haven’t taken everything into account, that not everything you said has been understood, and so on. Learn, and iterate!
But first, check out our tips and beware of mistakes we made, so you can avoid them.
4 Rules of a Successful Conversation
Now, in order to design a good conversational experience, we need to understand what makes a good conversation in the first place. So much research has been done in the matter, it would be a shame not to draw inspiration from it. We found the 4 rules of a successful conversation, as defined by Paul Grice, a great starting point.
- The maxim of quantity — be as informative as possible — give as much information as needed, and no more.
- The maxim of quality — don’t give information that is false or misleading.
- The maxim of relation — what we say should be relevant, and adequate to the subject and context of the conversation.
- The maxim of manner — be as clear and as brief as possible. Avoid obscurity and ambiguity.
15 Practical Tips for Great Conversational Experiences
Inspired by the above-listed maxims and our dialogue design experience, we’ve put together a list of chatbot conversational best practices. Some of them may seem obvious, but it’s not our aim to reinvent the wheel, merely share with you the principles we find the most useful and worth relying on.
1. Be consistent in your style and vocabulary.
We can’t stress enough how important this first principle is. Once you’ve created a unique personality for your chatbot, you’ll need to stay in character.
In order to stay consistent in your messaging, you might find it helpful to define your bot’s tone of voice — that is, the way to communicate its personality. If you’re not sure how to go about this task, you may find The Four Dimensions of Tone of Voice, as defined by the Nielsen Norman Group a great starting point:
- Funny vs. Serious
- Formal vs. Casual
- Respectful vs. Irreverent
- Enthusiastic vs. Matter of Fact
In addition, you can ask yourself which words should be part of your chatbot’s vocabulary and which expressions are off the table. No matter what you decide on, make sure that your tone of voice encapsulates both your brand and its customers.
2. Avoid jargon, prepare a glossary.
Your responses should be easily understandable to all customers, regardless of their background, education level or tech-savviness. Avoid business-specific, industry or technical jargon, as specialized expressions can elicit misunderstanding and mistrust. Remember that something that seems obvious to you (and your teammates) might not be so obvious to your customers. Test your messages with others, preferably the real users to double-check whether the terminology you’re using is clear.
Following the tests, a lot of our clients like to create a customer-friendly glossary with simple terms to use in customer communication. It’s generally a good practice to agree on terminology and stick to it, i.e. if you decide to call regular bank accounts “checking accounts”, don’t refer to them as “current accounts”). Here are two real-life examples of rewritten bot responses inspired by our banking clients:
Apart from testing, there’s a wide variety of tools that can help you assess your copy’s readability (i.e. Readability Formulas for English, Jasnopis for Polish). Different tools are based on different formulas, but the result you’re aiming for is something along the lines of “easy enough for school children to understand”. Bear in mind that you want your bot to be understood in less than ideal conditions too, i.e. street noise, distraction, etc.
3. Be brief and concise.
Dialogues should not contain irrelevant or rarely needed information. Every piece of information in a dialogue competes for visibility, and any extraneous information diminishes the visibility of what’s the most relevant.
One way to achieve a high level of brevity is to edit ruthlessly, and avoid meaningless words and fillers, such as “really”, “very” or “basically”.
It’s worth noting that this principle is especially important when it comes to voice bots. In a text-based bot, you can always skim through the written messages again and process them in your own time. There’s no way to do that in a voice bot — you need to understand what it says right away. Thus, it’s best to stick to the point.
4. Keep it short, like in a dialogue.
Long and complex sentences are discouraging and might provoke more questions than answers. Give only as much information as needed in one turn — it’s supposed to be a conversation after all. If your messages run on, your users will feel like they’re listening to a monologue and turn away. Take a look at our example to see what we have in mind:
In this example, a rewritten message highlights the main advantage and has been put in a short, clear and captivating form.
5. Divide long answers into separate chat bubbles.
Even though you should strive to keep it as short as possible, sometimes it’s necessary to respond in a more complex way. In such cases, you may want to consider whether to divide longer answers into two or three separate chat bubbles, mimicking a real online conversation. This will make them look more natural, and ensure they’re more readable.
6. Keep it natural — write as you speak.
In writing, we tend to use a more formal and impersonal style. However, when designing conversational experiences, we should put messages into words as if we’re having a conversation. To achieve that effect, imagine that it’s a dialogue between two people you’re designing, or that you’re talking to a trusted friend. Use an active and not passive voice, and utilize contractions such as “don’t” — “do not” can evoke a more serious feeling that you might intend for.
7. Diversify responses.
Crafting different variants of a bot’s response keeps the interaction interesting and fun. Just like a person, your chatbot should be able to respond in different ways to the same questions or reactions. Although the content and the goal of the message stays the same, its form can vary, i.e. “Hi”, “hello”, “good morning”. This makes the conversation feel more natural and keeps the experience from getting stale.
8. Personalise when possible.
Personalisation has become one of the hottest trends in marketing, and everyone’s betting big on it in the chatbot world too. Indeed, there are a few tricks you can use to make your users’ experience more customized. You can craft different responses for different segments, i.e. if someone already talked to your bot before, you can say “Hello, good to see you again!”, rather than plain “Hello!”. You can also greet them with their name or ask about the weather in their hometown. Surely, this will get you brownie points. Remember not to overdo it though — there’s a fine line between personalized and creepy.
9. Beware of dialogues that can lead to ambiguity.
Beware of traps when formulating questions. Remember that users’ answers aren’t always perfect — people are often distracted or merely skimming through the text, rather than reading it carefully. A question whether you prefer A or B can often result in a simple “yes”, putting your bot in quite a quandary.
10. Make sure the user knows what to do next.
Your goal is to drive the conversation forward and solve your customer’s problem in the most efficient manner possible. You don’t want the user to be at a loss to know what to do next. In fact, it’s best if all bot responses end with either a question or a suggestion of what action to take.
Take a look at our example below. The initial response is wrong for two reasons: it doesn’t suggest what to do next, and it assumes the user knows what his daily transfer limit is.
11. One-size-fits-all responses can be a good friend.
Users can formulate their intentions in a variety of ways, i.e. they can ask a question or express their intention in a sentence. Make sure to test your responses to see if they work well with all kinds of requests. Think about different ways your users could express their requests, i.e. “I would like to make a transfer”, “Can I make a transfer?”. A response starting with “yes” will only work in the second scenario.
You should also beware of responses such as “Great! Let’s get you started”. If someone is filing a complaint, a response like this might come off as non-empathetic (see example).
12. Avoid open-ended questions.
Open-ended questions are tricky: you can’t possibly predict how users will answer them, and thus you’re running the risk of your bot not being able to answer accordingly. Not to mention that they can get the customers side-tracked, and prolong the conversation for no valid reason.
It’s best to avoid them altogether, however, there are safe ways to use them too. For example, an insurance bot could say “Please explain the incident”, allowing users to speak freely, provided that its response is always the same, i.e. “Thank you for providing the information, I will now send it to our verification team”.
13. Speak with confidence.
Your bot is your brand representative and it goes without saying that it should make your users feel like they’re in safe hands. Everything it says should strengthen the underlying message that things are being taken care of. Avoid expressions that can make it look incompetent, such as “I’m not sure, but I think…” or “Sorry, looks like I didn’t understand again :(“. Of course, speaking with confidence doesn’t mean you need to know all the answers, especially when it comes to unrelated to business or sensitive questions, such as “Does God exist?”.
14. Use positive and supportive language.
Words can be incredibly powerful — they can make or break someone’s day. Be mindful of the emotional implication of the words you use and try to use positive and supportive language, whenever possible. Bring more unconscious positivity to your customers’ lives, and improve their overall experience with your brand.
15. Test and iterate.
Once you’ve written the first draft of your scripts, you may think that you’re done…but you’re not done at all! Don’t consider your dialogues unchangeable — conversation design is an iterative process and requires constant testing. Each element that we change or add can influence the shape of the entire conversation, and thus needs to be carefully examined.
We suggest you test all the messages you write. You can start by reading them out loud to yourself to check their feel and pace. You can also test them with your colleagues using i.e. Wizard of Oz method to find out what reactions the messages evoke, and if they’re understood at all. Once you’re done with corridor testing, iterate and remember to schedule further testing — this time with your client and the real users.
Final Thoughts
More than anyone, we know that dialogue writing is no easy task. That’s why we would like to share a little something extra with you — our very own checklist that you can use to improve your dialogues. We hope it serves you well!
We hope you enjoyed this article, and do comment if you have any questions. If you’re craving for more useful tips, check out the next part of our Bot Design Guide: Part 4, Chatbot UX Best Practices. If you’re interested in previous parts, you’ll find them below:
