Online AI Chatbot Conversational Design Principles

The most important aspects in designing an engaging AI chatbot is to use effective language and design an effective flow that helps the end customer get the answers they’re looking for or drive them to the goal you have for the chatbot. Existing constraints, such as real estate and the graphics that can be incorporated in the design, are some of the challenges with existing online chatbots.

The primary design tool you have to craft effective chatbot conversations with the customer is—not surprisingly—written language.

Your writing represents your organization to millions of potential customers. It must accurately present the style and culture of your organization and, as such, requires you to write with crisp, clear language and have the ability to understand the flow your customer would expect.

Start by understanding the customer


Effective chatbot communication starts by understanding the customer and what they’re trying to accomplish. By segmenting customers and understanding the demographics and expectations of your customer base, the structure and flow of the conversation can be prioritized. The style of the language will also depend on the organization you represent; the language and style of a government tax authority chatbot would be very different than a chatbot intended for teenagers.

Ideally, data is available from existing legacy sources such as email, CRM systems or existing live chat systems to help you identify and prioritize problems that your customer is looking to resolve. Common requests should be easy to get to, while those which are rarely asked may initially be omitted from the chatbot and dealt with either offline or through a live agent. If you try to handle too much in the chatbot, it can result in a higher percentage of incorrect answers.

AI chatbot or live chat solutions typically result in 2-300 times higher response rates than traditional Contact Us Now pages. To be successful after launch, the user questions need to be analyzed to identify shortcomings or missing answers.

Designing the flow


After prioritizing and guiding the customer down the fastest and easiest path to the answers they need, it’s important to consider the “what’s next” once they find those answers. This path should be aligned with your larger goals as well.

For example, if a user asked the chatbot about an application deadline for school and the AI provides the answer, you would likely then provide buttons for Apply Now, Application Steps or another relevant next stage to guide them along in their quest for answers, while always driving them toward your key objective.

Key conversational design principles

Writing an online chatbot conversation is very different than writing articles or FAQ, as it should come across as if it’s written in real time.

Below we have listed some of the most important elements of conversational design.

1. Short, concise and natural language

Even though a chatbot should never trick the customer in to believing they are a real person, the words used should. The language should be crisp and simple. One message should not be too long as it will come across as a canned message (which it actually is)

2. Pause between messages

If you have too much content for a single chatbot message, go ahead and set up short pauses between the messages.

On the platform (with link), we set up short pauses (in seconds) between longer messages. While we wait, we send the “writing in process” message to the user.


This will make the user feel engaged while they wait and give them time to read the messages they’ve already received. The longer the message, the longer pause.


Without pauses, the chatbot will push the messages too quickly, requiring the user to scroll back up or, worse, simply ignore the message and move on.

3. User should never get stuck

For every answer your chatbot provides, there should be an exit with logical next steps for the user to take. This includes having mechanisms to take you back to the main menu, efficient use of buttons, a mechanism to deal with sudden silence from the user and as escalation mechanisms.

In other words, the chatbot should never really have an end point.

4. Design answers and wording to make sense for a variety of questions

Customers will ask their questions in a variety of ways and the answers should be worded to fit as many of these ways as possible. This will also help reduce the number of intents which the AI has to be trained for.

Imagine, for example, that you sell blue, green and red bicycles. The answers related to colors should simply state the colors you offer: “We have blue, green and red bicycles.”.

This would then fit for all the questions related to color:

Which colors you have?

Do you have blue bicycles?

I want a pink bicycle.?

Yellow bikes?

Do you have the green bike?

You: Do you have a pink bicycle?

Bot: We have blue, green and red bicycles.

Obviously for the question “I want a pink bicycle,” the better answer would be, “Sorry, we don’t have pink, but we do have...” but this would exponentially increase the number of answers required and increase the maintenance and complexity of the NLP process maintenance of your answers.

5. Create a personality

Talk like a real person, but do no’t trick the user into thinking that you are one. Personality is key for engaging the user and, if done right, leaves the customer with the desire to come back. The online personality obviously depends on your organization, the type of problems your customer is dealing with and your audience.

The chatbot should have a friendly and warm personality, but don’t overdo the cuteness of the AI. If you’re a funeral agency or targeting teenagers, the answers and personality need to be designed appropriately.

6. Think outside the message!

Though real estate is limited, there are other means that can be used to make it easier for the user to understand your message, such as cards, pictures, forms or video.

Platforms like Facebook Messenger offers Webviews, which allows completely customized forms within the chat window itself. Webviews can be very effective and is less error-prone when capturing multiple interdependent values from the user.


Doing this through pure messaging can be more cumbersome than it is to simply selecting from a predefined set of values. This will also reduce the mistakes and friction the end user might experience, as there is less room for error.


7. Declare what you can do up front

When the chatbot is launched, it needs to inform the user how it can help. Don’t expect the customer to know what questions to ask. If you do,nt it’s likely to be the last time you connect. One-way communication is not fun.


8. Effective fallback and escalation procedures

The chatbot will not be able to handle all questions and is going to fail to understand your user. A key aspect of conversational design is avoiding irritating customers in how you deal with this failure.


Escalation can be dealt with in multiple ways, including transferring to a live agent, offline follow- up or capturing a phone number to call the customer separately. The key is to include the logic to trigger that escalation. We recommend the following considerations:

  • Maximum 3 default fallbacks in a row
  • Fallback followed by silence from customer
  • Negative sentiment / curse words
  • Requests to talk to agent
  • Maximum 3 times of the same intent triggered in a row

9. Internationalize your chatbot

You will have better response rates by supporting the languages your customers speak. With an increasing number of languages supported by different AI NLP engines such as Wit.AI or Google’s Dialogflow, it’s becoming easier to create a true multilingual chatbot.


Using our own chatbot platform (, we’ have added the flexibility to manage the content in as many languages as needed while controlling which AI NLP platforms can process the different languages.

10. Personalize messages

Depending on the channel before you even start the conversation, you actually can obtain some information regarding the user to help identify language, location and additional information which can be used to personalize the messaging.

Here’ is an example of what information can be collected from your browser when connecting through web chat:

When a user connects to your approved chatbot through Facebook Messenger, you can obtain the following information


This information can be used to default language, customize messages and localize the content to an extent, making the answers as relevant as possible.

Throughout the chat sessions, you will also collect additional information about the user. This information can be used to further customize messages and information to improve the user experience.

11. Be prepared for common questions asked to a chatbot

Some people will ask silly questions to the chatbot, so the chatbot should be ready to deal with these and get them going to what its AI is built to handle.

  1. Are you real?
  2. What’s your name?
  3. Where you from?
  4. How old are you?
  5. Tell me a joke!
  6. Rude words or phrases such as “I love you”
  7. How are you?
  8. Hi
  9. Emoji’s and stickers
  10. Photos or URLs

Though you should not spend too much time on these types of questions, you need to have a mechanism to deal with it just as your user expects a person to.

12. Vary responses

If the same intent gets triggered multiple times, it’ is a good practice to have two or three variants of the answer, often with more detailed answers every time.

You: I forgot my password?

Bot: Sure, I can help you with that! Just click “forgot password” and we will reset it for you

You: I can’t reset the password

Bot: If you go to the login page click “forgot password”

Bot: You then enter your email and submit. We will send a link to your email to reset your password.

After a repeated intent, consider adding additional options like “forgot email address,” as it may not be as simple as initially expected.

13. Information collected should be remembered

Many chatbots follow a flow to collect necessary information. For example:

You: I want to order a pizza

Bot: Sure, what size do you want?

You: Large

Bot: What kind of toppings…

If you instead said:

You: I want to order a large pepperoni pizza

The AI should capture that you already said “large” and also mentioned “pepperoni” toppings to avoid requiring the customer to repeat what they already said.

14. Remind the user if they stop interacting

If the user suddenly stops responding to the chatbot, it should have a timeout process (just like a human would) to re-ask the question or, alternatively, send the user through an escalation process.

15.Effective use of buttons or Quick Replies

The chatbot should solve a problem or question for the user. Don’t always force the user to the goal you have for the chatbot (sell something, get a lead, etc.) as this can irritate or possibly come across as pushy


Go after the top 10 typical questions/problems the users have and then carefully walk through how you can solve it for them.

Once your chatbot is launched, you need to continue improvements by leveraging chatbot analytics (link to chatbot analytics blog post), performing A/B testing, adding/modifying answers, training the AI NLP and finding drop-off points to maximize engagement and minimize failure.

What we do?

Symprio helps customers create engaging and cost effective chatbots to automate customer service, support and more

Conversational Design

By understanding end user behaviors and needs we design a conversatonal flow to engage end users and answer their questions.

Bot Platforms & AI Engines

We can design and implement your chatbot on a variety of platforms and AI/NLP solutions coupled with custom intergrations dependent on your needs.

Let us know your chatbot requirements here