10 step guide to preparing your knowledge base for Gen AI search
7 min read
Last edited:
If you’ve run into an obstacle when using a product, you know the drill. Your first instinct might be to ask the chatbot your question, wait for it to give you ‘helpful links’ that may solve your problem, and then read through these full-length articles to find a section that actually answers your question. Or you might go directly to the help center and find a section related to your product query or type in a keyword to find a relevant article. Either way, you need to put in a lot of effort and often read through multiple articles to find a solution. AI search agents are changing all of that.
Gen AI search agents can pull from a number of different article sources to generate the most accurate response to a customer’s question. DevRev’s search agents are powered by our AgentOS platform. With GPT-powered Search, your knowledge base articles become more discoverable to users. Instead of having to manually navigate and comb through articles, your users can now get accurate solutions at the speed of their keystrokes.
Also read: Explore DevRev’s new article creation and article analytics dashboard
Preparing your KB for Gen AI search
Your knowledge base is the fuel that feeds your AI search agent. You want to be confident that the information available in your knowledge base articles is relevant, up-to-date, and optimized for search. Here are ten steps to follow to get your knowledge base ready for AI search.
1. Centralize your article sources
Your knowledge articles might currently be spread across a number of formats and sources. For instance, you may have QnAs, help articles, PDFs, and links that feed your AI chat. If you’re using a separate tool like Docs or Notion to write new articles, then articles in progress or under review might be spread across different tools as well.
Centralizing your articles and sources is the first step in optimizing your knowledge base for AI. This is important for two reasons: one, it ensures you have control over what sources an AI search agent is pulling information from and two, it makes it easier to review and update knowledge articles.
2. Conduct a thorough content audit
This step is crucial to ensure your AI search agent isn’t offering outdated or incorrect information to users. Evaluate the current state of your knowledge articles to identify articles that need to be unlisted or updated. At this stage, it can be useful to bring subject matter experts from various departments to validate articles related to their focus area. This ensures that the articles are accurate and makes the review process faster by involving more teams.
3. Ensure AI-optimized language
You don’t have to drastically change content to make your articles ready for AI agents. DevRev’s Gen AI search agent can pull and contextualize information from different article sources to generate the best response to a user question. There are, however, a few content guidelines you should follow:
- If you have diagrams and charts, include supporting text to describe them
- Avoid using ‘yes’ and ‘no’ responses to FAQs. Instead, answer in full sentences with a description of the solution
- Ensure that each article only covers one topic
- Every paragraph should make sense when read in isolation. Avoid using transition phrases or words (such as ‘in addition’) at the start of a paragraph
This step is also a good opportunity to evaluate the quality of your articles and make sure they’re written in user-friendly language. Avoid using technical jargon as this can confuse users.
For a complete list of content guidelines, read this article on the Best practices for documentation that supports AI.
4. Identify and address content gaps
The whole objective of using Gen AI search is to help users self-serve. If you’re having the same questions repeatedly land in your support agents’ inbox, this means that there are gaps in your current knowledge base that need to be addressed. Use data from your support inbox to identify whether there are recurring questions agents have to respond to. Once you have a list, prioritize the questions in order of frequency and work on creating articles that answer them.
5. Differentiate between internal and external articles
A common concern organizations have when implementing a Gen AI search agent is that the agent may go rogue and surface confidential internal documents to external users. This is very simple to prevent. DevRev allows you to set viewing permissions for each article. You can decide what articles can be shown to customers and what should only be used internally. You can also set up segments based on certain criteria (eg. paid vs free-tier customers).
Once you have a consolidated list of KB articles, run a check of the visibility permissions on each one to ensure that confidential information is protected.
6. Organize articles with a structured folder system
Grouping your articles under categories makes it easier to manage your article collection and navigate through it. Without a classification system, your knowledge base resources can soon grow and become unwieldy.
One way to do this is to create folders for each category of your help center or for each product (if you have multiple). In DevRev, you can also attach articles to the related ‘part’ of the product they refer to. This makes it simple for internal users and help center admins to find an article without having to navigate through your whole knowledge base.
7. Collaborate with cross-functional teams for content creation
Writing knowledge articles shouldn’t have to be the responsibility of the Knowledge team alone. Product and engineering teams should be actively involved in supplying knowledge articles related to their product area. This is critical to ensure that your knowledge base is exhaustive and the search agent is able to surface accurate information to a wider range of questions. In DevRev, you can share different access levels to members within your organization to make it easier to contribute to writing articles:
- Viewer: Users can view all articles in the knowledge center
- Creator: Users can create new articles and edit existing ones
- Publisher: Users can create and publish articles
This allows multiple users to contribute to writing KB articles while still allowing KB admins to conduct a quality check before publishing.
8. Implement a process for incorporating product updates
Your knowledge base has to keep up with product updates or it’s going to be inaccurate. And an inaccurate database leads to an ineffective AI search agent. Along with involving product and engineering teams in the article creation process, it’s also a good idea to set a regular cadence of meetings to discuss product updates. At DevRev, for example, we have a monthly changelog discussion that brings together the product, engineering, product marketing, and knowledge teams. This helps keep track of new feature releases and features that have been deprecated so the article repository can be updated.
9. Track and analyze article performance metrics
So you’ve built out a refreshed help center that’s optimized for AI. But is it working the way you intended it to? Managing your knowledge base shouldn’t just end at publishing an article. DevRev’s article analytics dashboard allows you to track upvotes and downvotes for articles, time spent on each article, and most viewed articles. This makes it easier to understand what articles need improvement and whether your users are deriving value from your knowledge base.
10. Block time to periodically update your knowledge base
Knowledge teams get inundated with so many requests to create new articles that it can be difficult to hit pause and review the quality of existing articles. Make KB reviews a priority by setting up team meetings at a frequency that works for you. During this time, review old articles that may have gone stale, check if there is outdated terminology used in articles, and do a retrospective on how your article collaboration and publishing process is working.
Preparing your knowledge base for Gen AI search is an essential step to enhance user experience and streamline information retrieval. Implementing these ten steps will not only make your knowledge base more discoverable but also empowers users to find accurate solutions swiftly, transforming their interaction with your product and support system.
