If your AI is giving wrong answers, sending customers in circles, or just flat-out missing the mark, I can almost guarantee the problem isn’t your AI technology. It’s what you’re feeding it.
The number one reason your AI chatbot is failing your customers is your knowledge foundation.
This is Part 2 of my 4-part series on Implementing successful AI chatbots.
This series is going to give you the blueprints for addressing issues with failing AI chatbots and the three-part framework for turning your AI support from a liability into an asset.
In my work with SaaS companies, I’ve found that even the most advanced AI systems fail when built on knowledge bases designed solely for customers to find their own solutions or it’s stuffed so full of keywords for a richer SEO value, while they’re missing the vital pieces for AI consumption.
Today, I’m giving you 7 principles for rebuilding your knowledge foundation which is focused on both self-serve support, and unlocking AI-powered support. Your knowledge base needs to be focused on customer readability while it has the elements AI needs to provide great answers.
THE PROBLEM WITH TRADITIONAL KNOWLEDGE BASES
Let’s look at the structure of support articles that most companies have right now:
- Lacking clear answer in the title
When a customer asks your AI, “How do I reset my password?” and you have an article titled “Account Security Best Practices” with password reset buried in paragraph seven, your AI is going to struggle no matter how advanced the technology is. - Support articles contain marketing language
When Product launches a new feature, Marketing has created the story and many times this marketing language makes its way to the Support article. In this example, it talks about how “intuitive” and “powerful” the feature is at the top. - Steps to solve the issue are halfway down the page
As a customer, if I have to scroll past the marketing language to get to the answer, I might not make it to the solution. The steps need to be more accessible. - Solutions are in the media
Not all AI models can read images or videos and create solutions. At least, not yet. If your support page has an image that shows the solution, but the answer and step-by-step process isn’t written out in text, AI will not be able to decipher the correct answer to present to your customer. - Support article contains multiple related topics
Combining all best practices for security in one support article develops confusion for the correct answers. - Using internal jargon that doesn’t make sense to customers
Sometimes internal terms make their way to a support article that aren’t part of the customer’s workflow. And, the company uses these terms so much that you might think it’s an industry term. Internal jargon confuses customers with unclear answers.
These 6 issues are causing confusion with AI-powered responses in traditional knowledge bases. If this sounds like your knowledge base, don’t worry, I’m going to give you 3 actionable steps to rebuild your knowledge base for AI success.
STEP 1: CONDUCTING A COMPREHENSIVE KNOWLEDGE AUDIT
The first step is conducting a full external and internal knowledge audit.
This is the structure you need to turn a good knowledge hub into a powerful one.
What to audit
These are the places your company stores information about the product and functions.
- Customer-facing articles
- Internal support docs
- Product release notes
- Training materials
- Email templates
- Chat/ticket transcripts
Your first focus should be on the customer-facing support articles. However, this is the time to organize all of your content so it’s linked together. This will create company-wide value that I’ll cover in a bit. Let’s focus on the content of the support articles.
How to organize the audit
Use a Task Management tool your team already has. Likely something like: Linear, Asana, Trello, Airtable, even Google Sheets. Group content by section (e.g., Security) and intent (e.g., Password Reset). Link every related doc to that intent. Yes, it’s work, but it’s proactive work. Think of it as your single source of truth for updates. When the product team tweaks the password reset page, you’ll swap out screenshots in one go.
🔥 Pro Tip:
Set up an intake form or Slack command so your team can flag outdated content or product changes.
🔥🔥 Bonus Tip:
Track revisions in a separate board to show off your ROI. Imagine this: “Team X caught 15 gaps, and they were fixed in a week!”
What to look for
Delegate this audit to 10 people. Each spends 1 hour in the week and it will save you more than a full workday.
This is what you are documenting from your review of each piece of content:
- Outdated Information: Content that’s no longer accurate
- Intent Mismatches: Articles that don’t align with how customers phrase their questions
- Structure Problems: Topics that are scattered across multiple documents
- Knowledge Gaps: Questions customers ask that have no documented answer
🔥 Pro Tip:
Use a dropdown in your tool (e.g., Up to Date, Rewrite, New Content) to keep it simple.
Look at the Audit Status column in the above image.
Notes to take
For each intent, ask yourself these questions:
- Does this match what customers ask?
- Are customer keywords (e.g., “password reset”) front and center for AI and SEO?
- Priority: High (top 10%), Medium, or Low? (Hint: Check analytics—high views, low resolutions = High.)
- Who’s the expert to verify this?
This audit isn’t just cleanup—it’s your foundation for AI and self-serve success.
STEP 2: REWRITING CONTENT FOR AI CONSUMPTION
Now, let’s make your content AI-friendly without losing that self-serve or SEO value. Think of AI as a smart assistant who needs clear, short directions—not a novel.
Here are 7 principles when rewriting content for AI.
- One clear intent per article: This article is for “Password Reset”. Move “Two Factor Auth” to another article. This title, “How to Reset Your Password” beats “‘”Password Reset Tips and Tricks” for both AI and search engines.
- Categorized content: User Segments, Product Area, and Last Verified are likely in the details of the support articles. If you’re using a blogging platform, these details need to be added. (WordPress users: update the publish date when you make changes to your articles!)
- Front-load Answers: Put the solution first: ‘To reset your password, click Forgot Password on the login page.’ No digging required.
- Use customer language: Swap “authentication flow” for “password reset.” Simple, human language rules.
- Step-by-Step Instructions: Turn paragraphs into:
- Go to login.
- Click Forgot Password.
- Check your email.
- Style Formatting – Utilize style formatting, such as headers or bulleted lists. Rich formatting is a core principle to allow AI to scan the article and find the right solutions for your customers.
- Media Descriptions – When using photos or videos, add accompanying text to explain the media, which provides AI with the right context to give your customers.
STEP 3: STRUCTURING INFORMATION IN Q&A FORMAT
Traditional docs are task-oriented: “How to export data” or “Setting up a new account.”
But customers ask specific questions:
- Why didn’t I receive the reset email?
- Why did the reset link expire?
- What are the password requirements?
Here’s how you can swap your content with the way customers ask questions:
- Pull the top 100 questions from chat transcripts and tickets.
- Group them by topic (e.g., Password Issues).
- For each cluster, create an article:
- Title: “Why Didn’t I Receive My Reset Email?”
- Answer: “Check your spam folder first—emails sometimes hide there.”
- Steps:
- Log in.
- Click Forgot Password again.
- Still stuck? Contact support.
- Extras: Link to “What Are Password Requirements?”
Add your company’s tone and format
Now that you have the basic structure for AI, format the content to reflect your company’s tone so customers see this in the self-serve route, SEO reflects this when customers still google the answer, and your AI chatbot can also be trained on the way it should provide answers.
Your roadmap to measurable results
These aren’t theoretical concepts. Once you accomplish this foundational shift in your knowledge base, you are going to see an improvement in customer satisfaction, improve AI accuracy, reduce escalations to human support, and decrease time to resolution.
- Happier customers (higher CSAT)
- Smarter AI (fewer wrong answers)
- Less agent burnout (fewer escalations)
- Faster fixes (lower resolution time)
This is how you can make a pivotal shift in just one quarter.
- Week 1-2: Conduct a comprehensive audit and planning
- Week 3-7: Rewrite your highest-priority content (20% of your knowledge base)
- Week 8: Retrain your AI chatbot with the new content
- Week 9-13: Continue to iterate, retrain, measure while rewriting your medium-priority content (60% of your knowledge base)
Ask yourself: What’s it worth to fix this next quarter?
Start putting pen to paper and estimate the workload and build a proposal. I’ll say this again, it can’t be one person. This is a company-wide initiative to be the foundation of internal and external knowledge.
If you’re struggling with AI implementation, I’m offering a free 30-minute strategy session to review your current setup, identify the most pressing issues, and map out next steps to turn your AI chatbot from a customer frustration into a competitive advantage.
Loved this? Subscribe to Rush to Resolution for more CX insights.
Have a great day and a productive week.
