Your AI Knowledge Base is the foundation for everything your AI assistant says and does. It gives you full control over the content your AI can access, allowing you to provide fast, accurate, and brand-aligned customer service across chat, voice, email, and agent assist.
In this guide, you’ll learn how to build a high-performing knowledge base from the ground up, keep it up to date, and unlock more advanced tools as you grow. Whether you’re starting with a few pages or setting up something more complex, this guide will walk you through it step by step.
💡 Your AI Knowledge Base can be edited, grown and enhanced at any time so don’t worry too much about getting it perfect and complete on your first try. Often, through testing and previewing you will identify any gaps or improvement opportunities!
🛠️ How to create an AI Knowledge Base [10 minutes]
This section will cover how to build an AI Knowledge Base for your chatbot.
Step 1: AI Knowledge Bases
Head to ‘AI Knowledge Bases’ in the left-hand menu, then click on the blue 'Create AI Knowledge Base' button.
🌟 All new accounts will have an AI Knowledge Base pre-built in their account. Check to see if yours exists already and you can edit and enhance it
Step 2: Adding URLs
a. Name your Knowledge Base
Add a clear and descriptive label so you can easily identify this KB later.
b. Scan your website or sitemap for URLs (optional)
To quickly add multiple URLs, use either the “Scan website for URLs” or “Scan your sitemap.xml for URLs” buttons. This will automatically search for publicly accessible pages. Follow the on-screen instructions in the modal to complete the process.
c. Manually add or manage URLs
You can manually enter individual URLs in the input box. If you’ve already bulk-added URLs, they’ll appear in this list. Use the trash icon to remove any that aren’t relevant.
d. Add another URL
Click “Add a URL” to insert a new input field for additional manual entries.
e. Set a refresh schedule
Use the dropdown to choose how often Talkative should re-scan your URLs to keep your Knowledge Base up to date. We recommend selecting weekly.
Step 3: Adding File Data
You can upload files to enrich your Knowledge Base. This is useful for providing internal documentation, detailed processes, or context that may not exist on your public website.
a. Upload your files
Click the upload area or drag and drop your file into the box. Supported formats include PDF, JSON, TXT, and CSV. Files are limited to 20MB per file.
b. Manage uploaded files
Once uploaded, your files will appear in a list. You can remove any file that’s no longer relevant by clicking the trash icon next to it.
Talkative extracts text content from uploaded files. If your document contains key information embedded in images (like scanned PDFs or infographics), make sure that content is also available as selectable text, or update the document accordingly.
Step 4: Adding Free Text Data
Use this section to manually add any standalone text you want the AI to have access to. It’s perfect for: temporary or seasonal updates, notes you haven’t yet formatted into a document, quick additions that need to be searchable by the AI right away.
Step 5: Adding A Custom Prompt
a. Write your prompt instructions
Add clear, specific behaviours you want your AI to follow.
- Think of this as guiding your AI’s personality and behaviour, the clearer you are, the better the results.
- You can include as many instructions as you like to shape how the AI responds.
b. Click Save
Don’t forget to click Save to create the Knowledge Base.
Need help crafting the perfect prompt? Check out our AI Knowledge Base: Custom Prompts guide for best practices, real examples, and tips to shape your AI’s tone and behaviour.
⚠️ Troubleshooting Knowledge Base Errors
Once your AI Knowledge Base finishes processing, you’ll receive an email notification letting you know whether it was successful or if there were any errors. If errors do occur, you’ll also be able to view them directly in the URL section of your Knowledge Base. Just click ‘Show Errors’ next to any failed URL to get more information about what went wrong.
Here are some common errors and what they mean:
Out of Memory – Knowledge Base Too Large
If your Knowledge Base hits its memory limit during processing, reach out to support to check your licence restrictions. You can also try reducing the number of URLs or removing any particularly large or repetitive pages.
404 – Page Not Found
The page no longer exists or the URL was entered incorrectly. Double-check for typos or make sure the page is still live.
403 – Forbidden URL
The page exists, but it’s restricted. This often means the content is behind a login or permission wall that Talkative can’t access or the page is blocking our scraper tool. Try replacing the link with a public-facing version.
Unknown Error
This usually indicates the website couldn’t be scraped, often because it’s not indexed by Google or Bing or doesn’t allow scraping. Try using a different version of the page, or manually upload the content as a PDF or text file.
⚙️ Understanding the Knowledge Base Advanced Options
Once your AI Knowledge Base has been successfully processed, you’ll see an Advanced Options section appear at the bottom of the page. While most users won’t need to change anything here, it’s helpful to understand what this section shows, especially when testing or troubleshooting.
Download Knowledge Base Contents
This allows you to download a snapshot of what the AI sees in your Knowledge Base. This is especially useful when:
- You’re testing and want to verify that all intended content was pulled in.
- You want to double-check if something has been chunked or interpreted correctly.
- You’re reviewing for accuracy or gaps before launch.
Keep in mind: the downloaded file is read-only, if you need to make changes, update the source (URLs, files, or text).
Chunking Data
We use a technique called chunking to break down your KB content into smaller, digestible units called chunks. These chunks help the AI retrieve and understand the most relevant information when responding to a customer query.
- The Expected Chunk Count is an estimate based on your content.
- The Actual Chunk Count reflects what the system has successfully processed.
These numbers should be fairly close. If there’s a large difference, contact Talkative Support, it may indicate that something went wrong in processing.
Chunking Configuration (Advanced Use Only)
This section allows you to modify:
- Chunk Size – how much content is stored in each chunk
- Chunk Overlap – how much content is repeated between chunks to ensure context
We strongly recommend leaving these settings at their default values. Changing them may affect how accurately the AI retrieves and assembles answers. If you’re unsure, speak to your CSM or support team before making changes.
❓Frequently Asked Questions: Creating Your AI Knowledge Base
How do I know if my Knowledge Base has processed successfully?
You’ll receive an email confirming success or failure. You can also check for errors directly in the URL section by clicking “Show Errors” on any failed entries.
I scanned my website for URLs but some are missing?
Sometimes, your website may block our scraper tool on certain pages. If this happens we recommend uploading your sitemap.xml to ensure all pages are captured
My file upload is erroring: AI Knowledge base Encoding.
Files which are uploaded should be encoded using the UTF8 encoding.
When we refresh a knowledgebase, we generate embeddings from your data for the search tool to use. Embeddings must be valid UTF8 encoded content and will fail if any non-compliant characters are detected.
If you have a UTF8 encoded file and you are receiving an error message indicating there is an encoding error, this will most likely be due to a special character which was pasted into the text document.
When you generate your text files to include in the knowledgebase, you might copy text from word documents and other websites. The encoding of these files may differ and introduce “badly encoded characters”. If a file contains any badly encoded characters, the generation of embeddings will fail, and the knowledge base will not refresh.
To avoid this, you should remove any formatting from any content you paste into a text file. You can do this by using the Ctrl + Shift + V shortcut to paste, which will strip out any formatting. Additionally, iconography like bullet points might cause issues, so please replace these with standard dash characters (-).
An additional method to remove any poorly encoded characters would be to copy the text to the text file, and then open a new blank text file and copying the content to this new text file. The act of doing this will strip out any poorly encoded characters.