How to get started with HelpKit AI

HelpKit AI turns your Notion knowledge base into a smart, 24/7 AI assistant that provides precise and instant answers to your users. Let our AI help you resolve issues faster and focus on what matters most. In this article we’ll go over everything you need to know to get fully started and supercharge your knowledge base.

HelpKit AI is an add-on that can be added to any of our subscription plans. You can purchase the add-on it via your HelpKit dashboard by clicking on the HelpKit AI button within your settings. A popup will be presented from which you can subscribe to the add-on.

After activation, HelpKit will display an "Ask AI..." button as soon as your users start using the search bar on your site. When they click on the button, HelpKit presents an interactive chatbot interface that can answer their questions.

Depending on your selected knowledge base layout the chatbot experience will look slightly different:

Within the HelpKit embeddable widget as well as on mobile phones, HelpKit AI will show a full-screen chatbot interface to the user:

That’s all you have to do. Go ahead and feel free to test your chatbot either from the right preview section or live from your knowledge base site. After you have activated HelpKit AI we will automatically re-index the articles for your AI chatbot whenever you sync your content so that everything is always up-to-date. In case you notice that the chatbot might be lacking some help article content or is out of sync, you can always use the Train AI on Articles button on the settings page to re-index your knowledge base.

You can customize the language and behavior of your chatbot. Simply navigate to the HelpKit AI settings page and open the Chatbot Settings drawer. There you will be able to change fields such as your chatbot display name, the intro message and an error message in case the OpenAI™️ server is having some difficulties. You can tweak the chat intro and error message. Make it a bit more on-brand, let your users know they are talking to an AI chatbot or perhaps don't, if you feel like it. It's all up to you. Please make sure to hit Save Settings to see all of your changes applied on the right preview section. You can also enable to show a Experimental Badge on top of your chatbot so that users are a little bit more aware of potential flaws.

HelpKit AI offers an insights dashboard where you can learn about what your users are asking, how the AI bot is responding, and how satisfied they are with the answers. Simply navigate to the Insights section and click on HelpKit AI Chat. Here you’ll find your users chatbot conversations chronologically ordered. HelpKit AI automatically creates anonymous unique to a person Visitor IDs that help you better understand who interacted with your bot. You can sort by Visitor ID or the Start Time of the conversation. By clicking on a table row you’ll be presented with the full conversation history of your user on the right side of the screen.

After you have initially trained your knowledge base on our AI, HelpKit indexes your content whenever you sync using OpenAI™. When a user asks a question, it finds and combines information from within your knowledge base to provide an answer. It doesn’t use content from anywhere else to form its answers.

Given the nature of AI and quite frankly similar to us humans, please note that we can't assure you that every solution will be 100% accurate. We are constantly tweaking our AI model so that we can assure maximum possible answer accuracy.

We are currently using the OpenAI™ GPT-3.5-Turbo model for answer generation. Depending on our beta evaluation results we’ll consider adding GPT-4 support as well.

We use the OpenAI™ API to pass your content to OpenAI to index and process data. OpenAI does not use this content for service improvements (including model training). You can find out more about how OpenAI handles data here. Additionally we store chunks of your help article content in a vector database for ensuring the best semantic search experience.

If you found this article, you are awesome! We haven’t launched our public BETA yet but if you are part of the private beta we’d be more than happy to hear your feedback.