Ever set up an AI chatbot on your website, only to quickly hear from frustrated customers because of irrelevant, lengthy answers?
You're not alone.
Support interactions via self-service channels cost just $0.10 on average compared to $8 for agent-led tickets.
It's no surprise that support leaders across industries are looking to cut costs, boost efficiency, and enhance the customer experience by experimenting with Generative AI (GenAI) bots.
When configured correctly, these bots help customers get quick answers 24/7 without contacting your human agents or sifting through a ton of articles.
But simply launching a bot without equipping it with the right data will harm your customers, not help them.
Just like your human agents, AI bots require robust and well-organized resources to reference. In this article, we’ll cover the key steps you need to take to ensure that your generative AI bot can provide reliable and accurate answers.
Structuring your help center for generative AI
A clear structure with organized buckets of articles grouped by theme is what helps readers navigate your content easily. And it’s an absolute must for AI models to be able to comprehend your knowledge base content.
If the structure isn’t logical or consistent, your AI bot will struggle to understand how one piece of information fits into the bigger picture, leading to a poor customer experience with incorrect answers. Or your bot may just fail to reply at all, slowing down operations and causing an unwanted influx of support queries.
1. Speak the language your customers speak
The first step in structuring your knowledge base for generative AI is to align on the terms and language used across the help center. Consistent terminology is important, but it’s even more important to ensure that your vocabulary matches the vocabulary your customers use.
Say you sell reusable water bottles and most customers refer to the bottle’s "lid" as a "cap.” If you use "lid" throughout your help center, your chatbot may not understand that nuance. You can ensure that both terms are included in your article descriptions or as search tags, which will help the bot locate the relevant information and provide accurate answers.
But the best approach is to ensure the terms your team and your brand use match your customers default vocabulary choices.
Review customer service interactions, product reviews, and social media comments to understand which terms your customers use most often.
If your help center software allows it, reviewing no-results search queries can be particularly insightful for identifying terms used by customers that you haven’t covered in your articles yet (and if you’re using Zendesk, then Help Center Analytics can give you even more info).
Once you've identified these common terms, adjust your content accordingly or include alternative terms as tags or search keywords so your bot can understand different names for the same thing. Using functionality like Help Center Manager’s ‘find and replace’ can make it really easy to update terms across your help center, even if you have hundreds of articles.
2. Organize content logically
Once you’re aligned on terms, you need to define the scope and decide how to organize content in your help center.
Look at the common customer queries, reviews, and questions raised in live conversations again. This will give you insight into common problems and areas where customers need assistance. You should cover those items in your knowledge base to equip your AI bot to handle popular requests and questions.
Assign the identified themes into sections, categories, and sub-categories (depending on how many layers your help center tool allows) to group related content together and help AI better understand it. You can organize help center content by different themes, such as:
Related department
Target audience
Product line or features
Once you have your buckets ready, distribute any existing content into those buckets and assign your new pieces to the same categories.
The main idea is to ensure that these groups are mutually exclusive and that a single piece of content always belongs to only one group – while the entire knowledge base is exhaustive, covering everything you want to put out there for your customers and the bot to use.
Mapping your categories on a visual board can help verify the logic and ensure that groups don’t overlap and include every topic you need to cover.
3. Implement a tagging system
Most knowledge base platforms support tags in some form. For example, if you’re using Zendesk Guide, they offer labels to boost article visibility in search results and for their AI Answer bot.
Even if you have the terms your customers use in all articles, tagging your content is still a good idea to help generative AI better understand the key theme of each article and quickly locate information – with tags acting as search filters. By adding labels associated with the article content, you create a list of approved articles for the bot to pull from for different queries.
For instance, your “How do I reset a password on my user account?” article can have a tag like reset_password, especially if you have other articles talking about different system resets and password managers in your help center as well.
Develop content guidelines to help your AI bot excel
Once you have the structure set, it will serve as the foundation for your future content. And the content you add to your help center matters just as much – if not more – than the structure. To ensure success, you need to develop content guidelines and align your team on best practices for creating content that aids GenAI.
Let’s look into the key principles to write content that both your AI bot and your customers will love.
1. Stick to the “One article one message” rule
Your team members or customers would likely struggle if you wrote your help center articles as haikus.
For AI algorithms, it’s even harder, and a long read of FAQs won’t help your bot pull the best answer.
The "one article, one message" rule has been around for a long time, and with AI here, it’s even more important to stick to it. Every article should deal with one problem and one solution only. This is the key rule when writing for generative AI, as it helps your bot comprehend your help center content and provide customers with precise answers.
Wherever you can, break articles with multiple sections into separate pieces and group them together in the same help center folder if they are related. The more organized, straightforward, easy-to-navigate, and concise your content is, the better outcomes you can expect.
If you do need to write a longer article (as it’s always about a fine balance between your AI bot’s performance and help center experience for human readers), keep in mind that typically bots place greater emphasis on the first paragraph or so of the article.
That means you should always place as much contextually relevant information at the top of each article as possible.
2. Use a standard title format
To simplify knowledge base management and help your AI bot easily sift through the content, choose a consistent title format and stick to it. Ideally, each title should include three key elements:
The product name (if you support multiple products in the help center)
The feature name
The article theme
This consistent naming convention allows both humans and your bot to quickly identify which product or feature the article relates to, which is especially important if you have a large help center.
Retaining original question phrasing can also help the bot find more relevant answers.
For instance, to address inquiries about adding additional licenses, such as "I can’t find how to add a team member to my workspace?", you can create an article with the title "How do I add a team member to my workspace?" and include alternative terms your customers use for team members as tags and within the article text.
By maintaining a clear and consistent title format, you’ll enhance the efficiency of your AI bot and improve the user experience for your customers as a result.
3. Create article templates
Once your content guidelines are documented, the final step to set your help center team up for success is creating templates for different article types.
These templates help adhere to the established guidelines and ensure consistency and quality across your knowledge base. They provide a clear structure for your team to follow, which is particularly useful for maintaining uniformity in tone, style, and format. And this consistency is crucial not only for human readers but also for your AI bot, which relies on predictable patterns to locate and retrieve information accurately.
To get started, consider the various types of articles you might need, such as how-to guides, troubleshooting steps, FAQs, and feature overviews Each type can have a tailored template that highlights the specific elements required for that format.
For helpful examples, check out our blog post on four templates for creating great knowledge base articles.
Measure the effectiveness of your AI bot’s resources
Implementing a generative AI bot to help your customers isn’t a one-time task. Any bot requires constant review and training to bring the best results and avoid negative impacts on customer experience.
To understand if your AI bot is improving with changes made to the knowledge base, monitor its response, deflection, and resolution rates and see if the numbers are improving over time. If not, you might need to dig deeper into your bot’s configuration and continue to improve the documentation. You can:
Review AI-generated responses. Regularly evaluate responses provided in customer interactions for accuracy, relevance, and tone. If applicable to your workflow, try the KCS (Knowledge-Centered Service) approach and set up a system for your team to flag outdated bot answers or articles directly and improve content if they have edit permissions.
Identify areas for improvement. Note any discrepancies in the AI's responses or instances where the bot failed to deliver responses. These gaps signal opportunities to improve your documentation.
Improve the bot’s resources. Based on your findings, improve the bot’s performance by adding more tags or updating existing content in the knowledge base. When you add or update content, AI bots should re-learn and start delivering new responses based on your changes.
By improving your knowledge base for your AI bot, you’re ultimately enhancing the help center experience for customers and employees alike.
We all enjoy well-structured, easy-to-navigate, clear, and concise content much more than messy pages of duplicated and outdated texts, don’t we?
Improving your Zendesk help center
While AI is still evolving quickly, creating content that is organized, on point, and up-to-date is always a safe strategy. By investing in a robust knowledge base, you can hit two birds with one stone: providing a valuable resource for your team and customers while enabling your bot to offer helpful assistance around the clock.
And while creating and maintaining a knowledge base that works well for both human readers and AI applications may require significant effort, there are tools that can help.
Swifteq's Help Center Apps, specifically designed for Zendesk help centers, can help streamline knowledge base management and make the process more efficient. They’re the easiest way to manage articles, understand your help center’s performance, and automatically translate content into any language you can imagine.
Start your 14-day free trial today to bring your Zendesk Guide content to the next level.
Written by Maryna Paryvai Maryna is a results-driven CX executive passionate about efficient and human-centric customer support. She firmly believes that exceptional customer experiences lie at the heart of every successful business. |