How do you write successful chatbot dialogs?

The answer: You should proceed in a structured manner and take enough time.

1. preliminary exercise on the target group and chatbot personality

It pays off if you do a persona definition exercise and a chatbot personality exercise in advance. have made. Both exercises are the basis for writing successful chatbot dialogs.

Imagine your target group and, if possible, create a persona definition of the chatbot target group. You then define the chatbot personality that exactly matches your target persona. You think about it, how the chatbot personality would respond in this use case or situation. What language is used? What phrases? How much humor do you want to incorporate into the dialogues?

2. do not forget your goals

Furthermore, you must of course keep your defined goal in mind. Every dialog between chatbot and user must ultimately achieve your defined goals. For example, if your goal is a purchase, you should structure all dialogs in such a way that the user only wants to buy at the end and is also motivated to carry out the conversation to the end.

3. NLP or rule-based

Before you can start writing, you need to consider whether you have opted for a chatbot with free text and NLP or a chatbot with buttons.

3.1. Chatbot dialogs with NLP

In the first case, you now need to define the individual intents and then define example sentences for each intent that the user could say and the appropriate chatbot response.
As an example, let’s imagine the intent “Wrong product” ordered. Users express this with the following sentences, for example:

• I ordered the wrong one
• I have selected the wrong product.
• I made a mistake in the product selection.
• My son ordered the wrong thing.
• Etc.

I recommend that you define at least 15 example sentences per intent at the beginning. To define the example sentences, you can possibly use existing customer inquiries from emails or live chat or you can ask your target group or your employees who otherwise communicate with the target group.
You then define the bot’s response to each intent. This can be, for example, “I’m sorry, what is your order number”. And then you specify that the bot should then make a database comparison with the order number written by the user and what happens next.

In addition, you must write so-called case backs. What does the chatbot answer if it doesn’t understand a question? Depending on the bot personality, the answer may be more funny or more serious. In some cases, you may also define dialogs with placeholders. These are then filled in by a database during the query, as described above.

3.2. Rule-based chat dialogs

If you have opted for a simple chatbot Usecase, then write the dialogs for the rule-based chatbot now. Here you can imagine a tree diagram where the individual questions and answers are written down one after the other. However, depending on the complexity, this can quickly become confusing.

If you want to work analog, then it is best to use a very large sheet or start with small cards that you can then push together or move again if necessary.

If you prefer a digital approach, there are tools such as Collaboard or Mindmister that allow you to create digital mind maps or tree diagrams quickly and easily.

Although you have opted for the method with predefined buttons, small exceptions are still possible. Exceptions are possible if the user input is not relevant for the next chatbot response. For example, if you ask the user for their email address, the chatbot will always answer straight away. As a company, however, you have saved the user’s email address in the chatbot backend and can continue to use it if necessary.

Take some time to write the individual dialogs and read each dialog several times before you define your tree diagram as finished.

Regardless of which method you choose, you must ensure that the dialogs contain a certain consistency. You have defined the personas and personality in advance. You always have these in mind when designing the dialogs and write the dialogs in such a way that your target group will feel comfortable with them.

4. consider UX factors

4.1. No text blocks that are too long

Remember, many users use your chatbot via their smartphone. Consequently, the screen is small and the number of characters that fit on one screen is limited. Therefore, make sure that you do not overwhelm your users with too long blocks of text.
If you absolutely want to give a long and detailed answer, then at least divide it into short consecutive blocks.
As a rule of thumb, a text block should not be much longer than 2-3 lines. If you have blocks that are too long, there is a risk that users will quickly drop out of the conversation because they don’t want to read text that is too long. If you have several blocks below each other, the amount of text is the same, but it has less of a psychological effect.
Here are two examples: The first illustration shows a conversation with a very long block of text, which tends to overwhelm the reader. The second illustration shows basically the same text, but divided into several sections.

Figure: Negative example “text blocks too long” (own illustration)

Figure: Positive example “several short text blocks” (own illustration)

4.2. Find the right start

First impressions count. This also applies to chatbots. The start or welcome message is one of the most important messages of the entire chatbot. It influences whether the user continues to use the bot and what expectations they have of the chatbot.
Start with a friendly, inviting message that matches the personality of your chatbot and your target group persona.
As a rule, the chatbot is developed for a specific purpose. So make it clear right at the start what the chatbot can do and what it was developed for. The user must have the right expectations of the chatbot right from the start.
An open question such as “How can I help you?” is usually wrong. Then users think the chatbot can do everything, even though it was actually only developed for a very limited area.
You often have to mention the data protection rules at the beginning. But be sure to follow the first tip. Do not write too much text at once, but rather several small blocks. The wording of the privacy policy should also match your chatbot personality. Depending on the situation, you can also introduce them in a humorous way, such as “I know it’s annoying, but unfortunately it is… before we can chat with each other, you have to approve the privacy policy”.

4.3. Use of different media

You have already learned that chatbots can process different media. Whether text, images, videos or voice, different media formats can be linked together in a conversation with the chatbot. To ensure that the conversation with the bot does not become boring but remains vivid, it is worth integrating images or even videos in many cases.
If you want to explain something, it makes sense to insert a short how-to video.
If you want to show your products, you should definitely include pictures or even videos.
Media also belong to the dialogs and should be defined in the tree diagram or in the intent responses.
However, do not regard this point as a must. If there is no point in using images or videos, then leave them out.
It is also possible for the user to send a picture, video or file to the bot. Depending on the use case, this may even be mandatory if, for example, the user has to “show” their ID for identification purposes. Or it can be used in marketing campaigns to interact with the target group.

4.4. Consistency

Your chatbot must be consistent. This starts with the name of your bot, continues with the image and ends with the entire dialog design. No matter what content you develop for the chatbot, everything must match the story of your bot or match the personality of your defined chatbot and the personas of your target group. If the bot does not always have the same personality, your target group will be confused and this will reduce the success of your chatbot.

4.5. Planning fall-backs from the outset

This is only important if your chatbot works with free text. As soon as this is the case, the following will occur:
● Users insult your chatbot
● Users ask questions that the bot cannot answer
● The bot does not understand the user’s request

You must have defined a suitable answer for all three cases in advance.
If users insult your bot or perhaps even the exact opposite “fall in love with the bot”, then you must have defined an answer in advance as to how the bot should react here. Depending on the personality of the chatbot, the answer is either serious and direct or more fun.
Below is an example of what a humorous fall-back might look like, followed by a more serious fall-back.

Figure : Not understood “humorously expressed” (own illustration)

Figure : Not understood “expressed in serious terms” (own illustration)

If users ask a question that is definitely outside the capabilities of the chatbot, for example a chatbot for cooking recipes is asked about the weather, an answer must be defined as to how the bot should respond here. For example, “I’m sorry, but I can only provide information on recipes.”.

It is different if the bot has not understood the user correctly. Then it might make more sense for the bot to reply something along these lines “I’m sorry, I didn’t understand that, can you put it in other words”. However, the bot should only give this answer once or twice at most. The chatbot or your company must then honestly admit that the bot did not understand the request and must offer alternative options. For example, forwarding to a customer service employee or the bot asks the user for their contact details so that your employees can contact the user again later.

If you do not work with free text, but only with fixed answer buttons, you do not need fall-backs. Here, it may only make sense to integrate a button for further questions so that users can also place requests outside the bot’s area of expertise. However, the chatbot will then always respond in the same way, for example with a generic answer such as “Thank you for your question, I will pass it on…”. Ideally, your chatbot will ask for the user’s contact details directly so that you can contact them to reply. These open questions also offer opportunities to further develop your chatbot. You find out what else users would like to discuss in the chatbot and can even expand your dialogs if necessary. The following illustration shows what such an open question could look like in a chatbot

Figure: Chatbot with open question (own illustration)

If you need help writing your chatbot dialogs, I’d be happy to help. Click here to go to my contact page.

Do you already know the template “Let your Chatbot Talk”? Take a look here…