As the popularity of chatbots grew, the service of designing them developed. Amazon Alexa is one of the leading platforms open to creating chatbots and publishing them in the voicebot ecosystem. That's where the guideline for developers, "Conversation design principles checklist," was developed.
We've mentioned the topic of conversational interfaces and bots in a couple of articles: "What is a Conversational User Interface (CUI)?," "What is a Voicebot? Voicebot and Voice Assistant design," "Chatbot - what is it? Chatbot design." We highly recommend taking a look at them.
In this article, however, besides mentioning a few basic definitions, we'll focus on conversation design principles formulated for designers and Alexa developers. These guidelines will help you design all types of chatbots.
Conversational User Interface (CUI)
Conversational Interfaces allow a user to interact with software through text or voice while maintaining the appearance of natural conversation through technologies like, for example, natural language understanding or human speech recognition.
The role of a conversation designer
The goal of a conversation designer is to imitate natural human conversation between a user and a virtual assistant. Simultaneously, the replies of a bot need to be clear, simple, and user-friendly.
The development of AI technology and the spread of text-based interfaces has provided conversation designers with good benchmarks. That's why ChatGPT shouldn't be seen only as a virtual assistant but also as a great example of how such a conversation with a chatbot should look like.
However, the interface of a chatbot is still secondary to a graphical interface, and it needs to be truly well-designed to stand a chance of capturing user interest (especially if customers are using the chatbot to contact customer service).
Conversation design (or conversational design) concentrates on mimicking a human conversation. However, in contrast to a normal conversation, a conversation with a chatbot always has a goal (intent). Therefore, it is a process that involves gathering data from a user (slots) necessary for the chatbot to complete it.
Example. Chatbot for ordering pizza accomplishes the goal of collecting user requirements regarding the order: what kind of pizza it is, whether it has any extra toppings, the delivery address, and user confirmation regarding its price.
That's why conversation design encompasses various areas of expertise, including copywriting, business analysis, interaction design, programming, natural language processing, and user research (UX research). Experience in these fields significantly facilitates chatbot design because it allows designers to understand all the inner workings of a human.
It's also worth mentioning the wide range of applications of virtual assistants. Increasingly, conversation is becoming a natural interface when executing usage scenarios on websites or mobile apps. Chatbots are used in customer service in E-Commerce. However, if you want to make using a chatbot enjoyable, it should be based on solid foundations resulting from practice — Conversation design principles.
Principles of Conversation Design
Implementing a chatbot requires knowledge of five principles of conversational design developed by Amazon, the creator of Alexa. These key principles are universal rules that you can use to design any conversational user interface.
Five principles of conversation design include the following:
- Be natural.
- Be brief.
- Be contextual.
- Be multimodal.
- Be trustworthy.
We've emphasized a few times that the conversation with a virtual assistant should mimic everyday human interactions as closely as possible and, therefore, be natural. And it's quite a challenge, so building a chatbot can be time-consuming.
To achieve this effect, you should:
- Use leading questions to let the user know what information the virtual assistant needs to complete their request and to indicate that it's the user's turn to speak (turn-taking principle).
- Avoid using multiple questions in the same request/question/prompt.
- Avoid telling the user what they should say.
- Refrain from explaining how to use global controls such as "help" or "repeat."
- Ensure the support of various types of utterances.
- Listen to the recorded scripts to make sure they don't sound awkward or unnatural.
As a word of explanation, the "prompt" is a type of guide, a clue that tells the user what kind of information the virtual assistant requires or what to do next.
Remember that, unlike human-to-human interaction, the user can't rely on social cues such as body language or eye contact to know when to speak. Therefore, all types of prompts at the end of sentences will help them recognize their turn.
According to Amazon, users and customers expect the interactions with the assistant to be brief, quick, easy, and with as few frictions as possible. In the end, everybody wants conversations to be efficient.
To accomplish a given task with few steps, Amazon recommends using simple sentences and concise language to avoid, among others, high cognitive load. Don't present users with an essay about weather conditions when they only want to find out the temperature outside.
Therefore, if you want to write brief conversations and foster successful interactions, you should:
- Maintain a minimal amount of steps to complete a task.
- Use active voice
- Create responses that a human can say relatively quickly, and avoid monotone monologues.
- Avoid redundant information.
- Focus on providing one option or a few of them depending on the complexity of a task.
- If necessary, send the details of the response to the user in the form of a text message.
These tips will help you keep the interest of your users and result in a successful interaction. You should also keep in mind that analyzing interactions helps inform solutions that will assist you with improving the overall user experience. If you notice that users lose interest during an interaction, it will benefit you to look at where it happens and then try to infer why it does.
In short, you should offer users relevant information, options, and contextual cues appropriate for the context of a task or request.
Recommendations in this regard include the following:
- Adapt your messages in a way that will help users understand how a specific function works.
- Limit the length and number of messages along with the increasing experience of the user.
- Order the presented options according to the criterion of relevancy.
- Don't offer irrelevant options.
- Don't offer options that will result in an error.
When the user starts to learn how to navigate and communicate with your assistant, it's essential to ensure they understand how each function works. Over time, when they gain experience, you don't have to repeatedly inform them what to do, how a particular option functions, and what the bot can do for them.
Only present options that are context-aware; otherwise, they will feel out of place, confuse the user, and disrupt the conversation flow. Display only those options that result from the course of the conversation and will push the conversation forward.
The essence of this principle is to allow the potential user to use the virtual assistant in various contexts and conditions. They should be able to communicate with it whether they have access to a screen or not.
In a nutshell, if the user has access to a screen, supplementing a message with visual aids would be a nice touch. However, if the screen is unavailable, then audio instructions should suffice to accomplish a task.
If you want to create a good multimodal experience, then you should:
- Allow the user to use core functions without relying on a screen.
- Make sure that each target on a screen has an available voice command.
- Allow for interactions through touch if circumstances call for it.
- Ensure that touch interactions accompany the main voice functions.
- Display shortcuts on a screen so that the user can reach the most often-used functions.
- Enrich important moments that are relevant to a specific task by adding special interactions or interactive effects.
- Clearly display the contextually relevant information.
- Take advantage of user history and seasonality to offer various graphics and images that will diversify the static screen that the user will see most often.
Protecting your users' personal data is essential, and all the more so when they're increasingly more vigilant and more aware of how companies use their personal information.
Therefore, if you wish to design voice assistants worthy of users' trust and equipped with a truthful conversational interface that won't hide anything from them, you should keep a few guidelines in mind.
Guidelines for building trust:
- If your software offers the existence of multiple accounts, make sure that the user knows whose account they are using to avoid revealing sensitive information to an unauthorized recipient.
- Adhere to security requirements, such as the option to set up PIN codes.
- Display information that is relevant to the user.
- Don't interrupt the course of interaction with unwanted ads or other irrelevant offers.
- Express clear value propositions related to the actions that the user has to perform.
- Require users to confirm actions that can result in errors.
- Apply contextual and implicit confirmations of users' actions.
- Avoid displaying sensitive information on the screen.
- Ensure that the information displayed on the screen is relevant to the context of verbal responses.
- Perform tests of interactions to make sure that they run smoothly in all contexts: with the use of a screen, without it, and when the user is on the go.
- The name of an interaction/skill should reflect and inform about its functions, and this also applies to its description.
- Avoid promising any future functionalities.
- Don't require the user to confirm the command to exit or stop the interaction unless the action will result in the loss of progress or data.
Being trustworthy isn't limited to protecting your users' personal data. The virtual assistant should also be trusted to deliver reliable answers. If the software is caught providing false or inaccurate information, the user's desire to use it will significantly fall. That's why you should particularly care for the up-to-dateness of the information your voice or chatbot possesses; don't let it provide incorrect information. Keep in mind that the trust, once lost, is extremely difficult to recover.
As you can see, Amazon delivers a robust arsenal of recommendations and guidelines that you can utilize in your conversational design to build virtual assistants that will accomplish your business goals.
Before we conclude our article, one more principle deserves an honorary mention, as it should be the basis for designing conversation in general. Namely, we're talking about the cooperative principle.
Paul Grice, a British philosopher of language, is the author of this principle. In short, the principle emphasizes the need for cooperation between the speakers and listeners to reach an understanding.
The cooperative principle shouldn't be ignored during the design of virtual assistants as it's an essential component in human conversations. Thanks to it, system interactions will feel more intuitive and natural because the assistant and the user will work together to achieve a goal or complete a task.
Conversation Design Principles. Summary
Chatbots and voicebots will allow you to, among others, improve customer service or organize your customers' or your own daily tasks.
Understanding the conversation design principles is vital for creating an efficient and user-friendly virtual assistant that can carry out a successful conversation. That is why the principles presented by, among others, Amazon constitute an excellent set of guidelines and a checklist that can help you write natural-sounding dialogue and prevent you from creating overly automated messages that will be stiff and not human-like.
Don't underestimate the role of the conversation designer; they will be crucial to your design, and thanks to their comprehensive knowledge, the virtual assistant will be able to realize the above assumptions. With them, you can develop a conversational user interface that will be clear and easy to use.
That being the case, if you wish to achieve a holistic design process, ensure that your design is natural, brief, contextual, multimodal, and trustworthy, as well as that it follows the cooperative principle that will lead to mutual understanding between the user and the machine.
And last but not least, don't be afraid of testing. If you want your assistant to be effective, whether it will be text- or speech-based, you need to see how it actually performs in the confrontation with a user.
This will allow you to gain confidence in the capabilities of your software and, what's also important, help you determine whether the design fulfills user needs and expectations. User research and user testing are your friends; don't forget about them.