← Back to Blog
GuideKnowledge BaseAI SetupBest Practices

The Perfect AI Support Knowledge Base: What to Include and How to Structure It

The quality of your AI support agent is directly proportional to the quality of your knowledge base. Here's exactly what to include, how to write it, and common mistakes that tank deflection rates.

DT

Dev Team

Engineering · April 8, 2026 · 9 min read

Why Knowledge Base Quality Is Everything

We've onboarded hundreds of businesses onto Callsup. The single strongest predictor of AI deflection rate — stronger than ticket volume, industry, or product complexity — is the quality of the knowledge base.

A business with a thorough, well-structured knowledge base sees 80–90% deflection. The same business with a sparse or poorly written knowledge base sees 40–50%.

The AI doesn't invent knowledge. It retrieves and synthesizes what you've given it. If the answer isn't in your documents, the AI can't give it.

This guide is everything we've learned about building a knowledge base that makes AI support exceptional.


The 7 Documents Every Business Needs

1. Return & Refund Policy (The Most Queried Document)

This is the highest-leverage document you can upload. Return and refund questions are the most common support query for e-commerce businesses.

What to include:

  • Return window (exact number of days, from what event — delivery date, purchase date)
  • Which items are eligible (sale items, digital products, perishables, etc.)
  • Condition requirements (unopened? original packaging? tags attached?)
  • How to initiate a return (portal, email, or chat)
  • Refund timeline (processing days after receipt)
  • Exchange vs. refund options
  • Who pays return shipping
  • Good example:

    "Items may be returned within 30 days of the delivery date. Items must be unused, unwashed, and in original packaging with tags attached. Sale items marked 'Final Sale' are not eligible for return. To initiate a return, use our return portal at [link] or ask our chat agent. Refunds are issued to the original payment method within 5–7 business days of receiving the returned item. We cover return shipping for defective items; customer covers return shipping for change-of-mind returns."

    Every sentence is specific. The AI can answer dozens of variations of return questions from this single paragraph.

    Bad example:

    "We offer returns on most items. Contact us for details."

    The AI has nothing to work with.

    2. Shipping & Delivery Guide

    What to include:

  • Processing time (how long between order and dispatch)
  • Carriers used
  • Domestic shipping timelines by zone or tier
  • International shipping availability and timelines
  • Expedited/express options and costs
  • How to track an order
  • What happens if a package is lost or arrives damaged
  • Cash on Delivery availability (critical for Indian e-commerce)
  • Tip: Include a table of estimated delivery times by region. The AI will pull from it accurately for location-specific questions.

    3. Product FAQ by Category

    For each product category you sell, write a FAQ covering:

  • Materials and specifications
  • Sizing guidance (include a size chart in the document, not just a link)
  • Care and maintenance instructions
  • Compatibility questions (for tech/accessories)
  • Common comparison questions ("how is this different from X?")
  • Tip: Write FAQs in Q&A format. The AI is extremely good at matching customer questions to Q&A-formatted answers. Instead of:

    "Our bags are made from genuine leather and are water-resistant."

    Write:

    Q: What material is the bag made from?
    A: Genuine full-grain leather, water-resistant. Not submersion-proof.

    >

    Q: Is it waterproof?
    A: Water-resistant — it handles light rain. We don't recommend submerging it.

    Two questions, two precise answers. The AI matches both variations perfectly.

    4. Account & Order Management Guide

    Cover the most common account questions:

  • How to create an account
  • How to reset a password
  • How to track an order (with the exact URL or steps)
  • How to cancel an order (and the cancellation window — "orders can be cancelled within 2 hours of placement")
  • How to change a shipping address after ordering
  • How to apply a discount code
  • What to do if a payment fails
  • 5. Pricing & Plans (for SaaS / Subscriptions)

  • What's included in each plan tier
  • How billing works (monthly vs. annual, proration on upgrades)
  • How to upgrade or downgrade
  • Free trial terms
  • Cancellation policy (do unused days roll over? What happens to data?)
  • How to get an invoice or receipt
  • 6. Troubleshooting Guide

    For SaaS, electronics, or any product with technical complexity:

  • The top 10 most common issues and their solutions
  • "Try this first" steps for common error messages
  • How to access logs or diagnostic info (for technical support escalations)
  • Known limitations or bugs with workarounds
  • Structure this as step-by-step numbered lists. The AI will walk customers through troubleshooting systematically.

    7. Contact & Escalation Information

    When does the AI hand off, and what should the customer expect?

  • Human support availability hours
  • Average response time for email/ticket
  • Phone number (if applicable)
  • What information to have ready for a human agent
  • Links to community forum or self-service resources

  • How to Structure Documents for Maximum AI Accuracy

    Use Clear Headings

    Headings help the AI understand which section of a document is most relevant to a query. Without headings, a 20-page document looks like one blob of text.

    Good structure:

    # Return Policy

    ## Eligibility

    ## Timeframe

    ## How to Initiate

    ## Refund Timeline

    ## International Returns

    Be Specific with Numbers

    Replace vague language with specific numbers every time.

    VagueSpecific
    "within a few days""within 3–5 business days"
    "a long time""up to 14 days"
    "most items""all items except final sale, digital, and perishable products"
    "contact us""email support@yourstore.com or use the chat widget"

    The AI quotes your documents. If your documents are vague, the AI's answers are vague.

    Q&A Format for Common Questions

    As shown above: structure your most frequently asked questions explicitly as Q&A pairs. This dramatically improves the AI's ability to match customer intent to the right answer.

    Separate Documents by Topic

    Don't put everything in one 50-page document. Separate topics into individual files:

  • return-policy.pdf
  • shipping-guide.pdf
  • product-faq-bags.pdf
  • product-faq-electronics.pdf
  • account-management.pdf
  • Shorter, topic-focused documents produce more accurate retrieval. The AI finds the right document and the right section within it.


    Common Mistakes That Kill Deflection Rates

    Mistake 1: Only Uploading Marketing Content

    "Our bags are crafted with love and attention to detail using the finest materials sourced from ethical suppliers."

    This is useless for support. Upload operational documents — policies, specifications, how-tos.

    Mistake 2: Linking Instead of Answering

    "For our return policy, visit [link]."

    The AI can't follow links inside documents. Write the content directly in the document.

    Mistake 3: Outdated Documents

    Uploading your 2023 return policy when it changed in 2025. The AI will give wrong answers confidently. Keep documents current and re-upload when they change.

    Mistake 4: Only One Language

    If your customers write in Hindi, Tamil, or Arabic, consider uploading key policy documents in those languages too. The AI handles multilingual queries from English documents, but having native-language documents improves accuracy for complex nuanced questions.

    Mistake 5: No Size Charts or Specification Tables

    If you sell apparel, electronics, or furniture — customers will ask "will this fit?" endlessly. Include actual size charts, dimension tables, and compatibility matrices in your documents. The AI reads tables accurately.


    The Iterative Improvement Loop

    Your knowledge base isn't a one-time setup. It's a living system. Here's how to improve it continuously:

    Week 1–2: Upload your initial documents. Go live.

    Week 2: Review conversations where the AI said "I'm not sure" or customers escalated. These are your gaps.

    Week 3–4: Add content for every gap topic identified.

    Month 2+: Review monthly. Add new product FAQs when you launch new products. Update policies when they change.

    Most businesses see deflection rate increase by 15–25 percentage points between month 1 and month 3 — purely from knowledge base iteration.


    Quick Checklist Before You Go Live

  • [ ] Return/refund policy uploaded (specific, no vague language)
  • [ ] Shipping guide with timelines and carriers
  • [ ] Product FAQ for top 3 product categories
  • [ ] Account management guide
  • [ ] Troubleshooting guide (if applicable)
  • [ ] All documents use headings
  • [ ] No links instead of content
  • [ ] All dates and numbers are current
  • [ ] Documents are separate by topic
  • [ ] Test: ask 10 common customer questions and verify answers

*Ready to build your knowledge base and go live? Start your free Callsup trial — upload your first document in minutes.*

Topics

Knowledge BaseAI SetupBest PracticesRAGDocumentation

Try Callsup Free

Automate 80%+ of customer support queries with AI voice and chat — live in under 2 hours.

Start Free Trial

More from the Blog

The Perfect AI Support Knowledge Base: What to Include and How to Structure It — Callsup Blog