How to Upload Your Knowledge Base to Ensoras

This guide walks through getting your knowledge base into Ensoras and verifying the AI can find it. The mechanics are simple: a list of items with title and content. The hard part isn't uploading; it's having the right content to upload. We'll cover both.

Before you start

You'll need:

• An active Ensoras project
• Your existing documentation (help center articles, policy pages, FAQ, etc.), either as text or as live URLs
• About 1–2 hours for the initial pass on a typical brand's KB; 1–2 weeks if you're doing serious cleanup along the way

Step 1: Open the Knowledge Base page

In your Ensoras admin panel, go to Automation > Knowledge Base. You'll see a list of your current items (empty if this is a new project) with a search bar at the top and a button to add a new item.

Step 2: Decide between Text and URL items

When you create a new knowledge base item, the Type dropdown gives you two options.

Text items: you paste the content directly into the form. Use this for content you want fully controlled, content the AI sees exactly as you wrote it. Good for: policies you want to phrase specifically for the AI, internal procedures customers don't see in your public help center, content that doesn't have a public URL.

URL items: you paste a public URL and Ensoras fetches the page, extracts the readable content, and indexes it. Use this for content that already lives on your site. Good for: existing help center articles, policy pages, product info pages.

Most brands use both. URL items get you fast coverage of your existing docs; Text items let you fill gaps without publishing new pages.

Step 3: Add your first item

Click New item (or whatever the create-item action is labeled in your admin).

For a Text item:

• Item Title, be specific. "Refund policy" beats "Refunds." "International shipping rates" beats "Shipping."
• Type, Text
• Content, paste the full text. Plain text is fine; the AI doesn't need formatting.
• Active, leave on so the item is included in retrieval immediately.

For a URL item:

• Item Title, same advice; specific titles win.
• Type, URL
• URL, the full public address, including https://.
• Active, leave on.

Save. For URL items, Ensoras starts fetching and indexing in the background. For Text items, indexing happens immediately. Either way, the item is searchable within a minute or two.

Step 4: Cover your top 20 ticket topics

Once you've added one item, the temptation is to dump everything in. Resist it for a moment and do this audit instead:

Pull a list of your last 30 days of tickets. Categorize them. Identify the top 20 topics by volume. Make sure each one has at least one knowledge base item that directly answers questions in that category.

Common gaps that catch teams off guard:

• "What's your return policy?" gets a generic page-load answer when customers actually want specific timeframes for specific products.
• "How do I cancel my subscription?" exists as a help article but doesn't cover what happens to the customer's outstanding orders.
• "Do you ship to [country]?" is missing because the team thought their shipping policy page already covered it.

The teams that build a KB first and ask "what would a customer ask?" second usually under-cover the actual question patterns. The reverse order is more reliable.

Step 5: Write content the way customers ask

Knowledge base items work best when they're written in the question-and-answer voice the AI will use to retrieve them. Two examples for the same content:

Less retrievable: "Our return policy is 30 days from the date of delivery for unworn items in original packaging. Restocking fees may apply. Refunds are processed within 5-7 business days of receiving your return."

More retrievable: "Question: How long do I have to return something? Answer: You have 30 days from the day your order arrives. Question: Does the item need to be in the original packaging? Answer: Yes, items need to be unworn and in original packaging for a full refund. Question: Is there a restocking fee? Answer: Restocking fees may apply on certain product categories, we'll let you know during the return process. Question: When will I get my money back? Answer: Refunds typically appear in 5–7 business days after we receive your return."

The second version retrieves more accurately because it mirrors how customers phrase questions. The information is the same.

Step 6: Refresh URL items when content changes

URL items show their last scrape timestamp on the item's edit page. When the underlying page changes (you update your shipping rates, your return window, etc.), click Rescrape on the item to refresh the indexed content.

A common rhythm: after any policy change, immediately rescrape the URL items that point to the affected pages. Don't wait, your AI will keep answering with the old policy until the index updates.

Step 7: Test that the AI is finding what you uploaded

The fastest way to verify retrieval is to send a test message through your chat widget or email asking a question that one of your KB items should answer. Then check the conversation in your Ensoras admin panel.

The AI's reply should reflect the content you uploaded. If you uploaded "30 days for returns" and the AI says "60 days," the retrieval found a different document (or the AI hallucinated; rare but possible if the item isn't well-titled).

If the AI escalates instead of answering, the most common causes:

• The KB item exists but isn't marked Active.
• The KB item's title and content don't match the way the customer phrased the question. Add the customer's phrasing to the item's content (the question-and-answer technique above helps).
• The relevant workflow doesn't have any tools or instructions that direct the AI to use the knowledge base. The AI uses KB context automatically, but it needs to be running inside an active workflow to engage at all.

Common issues

The AI cites the wrong page. When two KB items cover overlapping topics, the AI sometimes pulls from the less-specific one. Audit for overlap; consolidate or delete the weaker version.

Long pages don't retrieve well. Ensoras splits large documents into smaller chunks for retrieval. If your shipping policy is a single 5,000-word page covering ten distinct sub-topics, split it into ten focused items. Smaller items with sharper titles consistently retrieve more accurately.

International or product-specific policies get mixed up. The AI doesn't know which policy applies to a US customer vs an EU customer unless your KB items make that explicit. Title items like "EU return policy" and "US return policy" rather than just "Return policy."

Outdated information lingers after you update your site. URL items don't auto-refresh on a schedule. Build a habit of rescraping after any policy change.

What this unlocks

Once your KB is in good shape, the AI's accuracy on policy questions jumps meaningfully. Categories that previously escalated (shipping inquiries, return-window questions, sizing questions) start resolving autonomously. Customers get faster, more consistent answers; your team stops typing the same five replies eighty times a week.

The Air Canada chatbot case from 2024 is the canonical reminder of what happens when a customer-facing AI gives wrong information without an authoritative knowledge base behind it. The chatbot promised a bereavement refund the airline didn't actually offer; the British Columbia tribunal held Air Canada liable. Documentation quality is not optional, and the highest-leverage hour you can spend on AI customer support is usually inside your knowledge base, not your AI vendor's settings.

What to do after this

If your KB is in good shape, the natural next step is configuring how the AI uses it: write your first workflow rule. Workflows tie the knowledge base together with tools and actions; the KB is what the AI knows, the workflow is when and how it acts on what it knows.

If you're early in your Ensoras setup, also see connect Shopify to Ensoras and the chat widget setup guide to round out the foundation.

Sources

• CBC News, Air Canada found liable for chatbot's bad advice, 2024 case showing the cost of running customer-facing AI without a clean knowledge base behind it.
• Forrester, Predictions 2026: AI Gets Real For Customer Service, But It's Not Glamorous Work, analyst view on why knowledge-base cleanup is the foundational work that determines AI deployment success.