How to Build Your AI Knowledge Base in Voxe (Step-by-Step)

Why Your Knowledge Base Is the Most Important Thing You'll Configure

When customers ask your AI a question, it doesn't guess. It searches.

Specifically, it searches your knowledge base — the collection of documents, policies, FAQs, and guides you've uploaded — and pulls the most relevant content to form an answer.

The quality of that answer is directly tied to the quality of your knowledge base. A sparse, disorganized knowledge base produces vague, unhelpful responses. A well-structured one produces answers that feel like they came from your best support agent.

This guide covers how to build one that works.

---

What Belongs in a Knowledge Base

Before you open Voxe, spend five minutes thinking about what your customers actually ask.

Common categories to cover:

• Product or service FAQs — pricing, features, limitations, availability
• Policies — refunds, shipping, cancellation, SLAs
• Onboarding docs — getting started guides, setup instructions
• Troubleshooting — common errors, known issues, fix steps
• Account management — how to upgrade, change billing, reset passwords

You don't need to cover everything on day one. Start with the top 20 questions your support team answers most often — those will give you the highest-impact coverage immediately.

---

Step 1: Navigate to Your Knowledge Base

1. Log in to your Voxe dashboard
2. Select the Demo (chatbot) you want to configure
3. Click Knowledge Base in the left sidebar
4. You'll see your existing knowledge bases — or an empty state if this is your first time

Each Demo has its own knowledge base, so you can run different AI assistants with entirely separate knowledge sets (useful if you manage multiple brands or products).

---

Step 2: Create a Knowledge Base

1. Click + New Knowledge Base
2. Give it a clear, descriptive name — e.g., Refund Policies, Product FAQ, Onboarding Docs
3. Add an optional description to help you stay organized as you scale

Tip: Group by topic, not by document. One knowledge base per major support category is easier to maintain than one giant base with everything mixed together.

---

Step 3: Upload Your Documents

Voxe supports the following file formats:

• PDF — ideal for policy documents, manuals, and formatted guides
• DOCX — Word documents and internal wikis
• TXT — plain text, exported FAQs, or raw support scripts

To upload:

1. Inside your knowledge base, click Upload Document
2. Drag and drop your file, or browse to select it
3. Voxe will automatically:
   • Extract the text content
   • Split it into semantic chunks
   • Generate embeddings for each chunk
   • Index it for real-time retrieval

Processing typically takes a few seconds for short documents and up to a minute for long PDFs.

---

Step 4: Review What Was Indexed

After upload, click the document to see how Voxe chunked the content.

What to look for:

• Are the chunks meaningful? Each one should represent a coherent idea or policy point — not a sentence fragment.
• Are headers preserved? Well-formatted PDFs and DOCX files produce cleaner chunks than unstructured text.
• Is anything missing? Scanned PDFs (image-based) may not extract cleanly — convert them to text-based PDFs first.

If a document extracted poorly, delete it and re-upload a cleaner version. The indexing is instant, so iteration is fast.

---

Step 5: Connect the Knowledge Base to Your Demo

Uploading documents doesn't automatically activate them. You need to link each knowledge base to the Demo that should use it.

1. Go back to your Demo settings
2. Navigate to AI Configuration or Workflow Settings
3. Under Knowledge Sources, select the knowledge bases you want this chatbot to query
4. Save your configuration

You can connect multiple knowledge bases to a single Demo. Voxe will search across all of them when a customer asks a question, retrieving the most relevant chunks regardless of which base they came from.

---

Step 6: Test Your AI Against Real Questions

Before going live, test the chatbot with questions your customers actually ask.

1. Go to Simulate in your Demo settings
2. Type in a real question — e.g., "What's your refund policy if I'm unhappy with the product?"
3. Review the AI's response:
   • Is the answer accurate?
   • Is it pulling from the right document?
   • Is the tone appropriate for your brand?

If the answer is wrong or vague, it usually means one of three things:

• The relevant information isn't in the knowledge base yet — add it
• The document chunk that contains the answer is too large or poorly formatted — split the content manually and re-upload
• The system message doesn't give enough context about the company — update it in AI Configuration

---

Step 7: Write a System Message That Sets the Tone

Your knowledge base provides the facts. Your system message shapes how the AI presents them.

Think of the system message as your AI's job description. A good one includes:

• Who the AI is: "You are a support assistant for [Company Name]."
• What it should and shouldn't do: "Always recommend contacting a human agent for billing disputes."
• Tone and voice: "Be concise, friendly, and never make promises about timelines you can't confirm."
• Escalation triggers: "If the customer expresses frustration or asks for a refund over $200, escalate to a human agent."

Access the system message under Demo Settings → AI Configuration → System Prompt.

---

Maintaining Your Knowledge Base Over Time

A knowledge base is never truly finished. Policies change, products evolve, and customer questions shift.

Build a simple maintenance habit:

• Monthly: Review any AI responses that were escalated to humans. If the AI consistently fails on a particular topic, add a document covering it.
• Quarterly: Audit your documents for outdated policies, old pricing, or deprecated features.
• On every policy change: Update the relevant document immediately. Voxe re-indexes on upload, so changes take effect within seconds.

---

Knowledge Base Limits by Plan

If you're on the Starter plan and hitting limits, consider consolidating documents — combine FAQs into a single well-organized file rather than uploading each section separately.

---

What a Good Knowledge Base Looks Like in Practice

Here's a real-world example structure for a SaaS company:

Knowledge Base: Product FAQ
  └── features-overview.pdf
  └── pricing-faq.txt
  └── integrations-guide.docx

Knowledge Base: Policies
  └── refund-policy.pdf
  └── cancellation-terms.pdf
  └── sla-agreement.pdf

Knowledge Base: Onboarding
  └── getting-started-guide.pdf
  └── first-30-days.docx
  └── common-setup-errors.txt

This structure keeps retrieval fast and answers precise. The AI knows exactly where to look for each type of question, and you know exactly where to update content when policies change.

---

A well-built knowledge base is the difference between an AI that frustrates customers and one that genuinely replaces a support agent.

Start with your top 20 questions, get those right, then expand from there.

Start your free trial at voxedesk.com — no credit card required.