The Knowledge Base is where your AI learns about your business. It acts as a single source of truth, giving your AI the information it needs to answer questions clearly and confidently.
By adding helpful, high-quality content, like key details, policies, FAQs, and links, you provide the context your AI needs to deliver accurate, relevant, and on-brand responses to your customers.
When set up well, the Knowledge Base helps your AI:
- Give consistent answers
- Reduce repetitive support questions
- Respond more accurately
Think of it as teaching your AI about your business: the clearer and more complete the information, the better it performs.
| The content you add to your AI Knowledge Base is used to power features like AI Replies. |
In this guide, you’ll learn how to add information to your Knowledge Base, what kind of content works best, current limitations to keep in mind, and how to troubleshoot common issues.
| ⚠️ Note: Manychat AI is currently in beta. Some capabilities are limited as we continue learning, improving, and developing the next iteration of the feature. |
What can you add to the AI Knowledge Base?
At the moment, the Knowledge Base supports two types of content: text and public links.
Text
Add the key information your customers usually ask about, along with the details you already have about your courses or programs. This can include your course content, different levels or packages, pricing, payment options, how to join, access instructions, FAQs, or policies. The more helpful information you include here, the better your AI can answer questions and guide customers.
Make sure the content is in plain, readable text. You can also include links in the text to your scheduling pages, other online sources, etc. (For example, “you can book a call with me here: [link]”).
| Important: Each Knowledge entry supports up to 250,000 characters, so you have plenty of space to include everything your AI needs to know. |
Public links
Add URLs that contain readable text your AI can scan and learn from.
| Note: Each Knowledge Base entry supports up to 250,000 characters. If a webpage exceeds this limit, content near the bottom of the page may not be processed. So, if you're adding a very long webpage, review the extracted content and upload any missing information as a separate entry to ensure your AI has the full context. |
To make sure the content in your AI Knowldege Base works correctly, keep these limitations in mind:
- Your AI processes only extracted readable text, it cannot interpret HTML structure, visual design, videos, audio, embedded files, or images.
- Extracted content is kept close to the original page and is not rewritten or summarized.
- Adding a homepage does not automatically include subpages, each relevant URL must be added individually.
Adding text to your AI Knowledge Base
- Open your Manychat workspace and select Manychat AI from the sidebar.
- Switch to the Knowledge tab.
- Click the Add text button.
- Add the text that you want to keep in your AI Knowledge Base and click Submit.
Adding links to your AI Knowledge Base
- Open your Manychat workspace and select Manychat AI from the sidebar.
- Switch to the Knowledge tab.
- Click the Add link button.
- In the window that opens, add the link and click Scan Now.
- Once the page is scanned, you can review the results and add the entry by clicking the Add Source button.
- Click Discard if you don't want to save the link and its content in the AI Knowledge Base.
Tips and best practices
To make sure your AI can properly scan and learn from your links, follow these guidelines.
Add one page at a time
The system scans individual pages, not entire websites.
For example, instead of adding: yourwebsite.com/blog add each article separately:
-
yourwebsite.com/blog/article-1 -
yourwebsite.com/blog/article-2 yourwebsite.com/blog/article-3
This ensures all important information is captured.
Make sure the page is public
Your AI can only scan content that is accessible without login permissions. Restricted content may include:
- Private Google Docs
- Password-protected pages
- Membership-only resources
If a page isn’t public, update the sharing settings (for example, set it to Anyone with the link can view) and then re-add the URL.
Ensure the page contains readable text
The system extracts visible, readable text only. It does not interpret:
- Images
- Videos
- Design layouts
- Embedded files
- Interactive widgets
- Forms
If important information is stored inside PDFs, images, downloadable files, or visual-only sections, your AI may not be able to access it. Make sure to place essential information directly on the webpage as plain, readable text.
Unsupported links
Some platforms cannot currently be scanned. Avoid adding links from:
- Facebook (facebook.com, fb.com, facebook.net)
- Instagram (instagram.com)
- YouTube (youtube.com, youtu.be)
- WhatsApp (whatsapp.com, whatsapp.net, wa.me)
- LinkedIn (linkedin.com, lnkd.in)
- Google Docs or Sheets (docs.google.com)
- Direct Amazon S3 file links (s3.amazonaws.com)
These sources cannot be processed.
Special cases to be aware of
- Canva links: only navigation text may be extracted; designs and images cannot be read
- Google Docs/Sheets: may load but return unusable table-style text
- Storefront pages (e.g., Amazon shops): may include extra navigation content
If a link doesn’t work, try adding a different page from the same website or paste the text manually instead.
| We’re actively expanding Knowledge Base capabilities. Let us know which knowledge sources you think we should make available for Manychat AI! |
Troubleshooting
If something isn’t working as expected, here are the most common issues and how to resolve them.
Restricted content
This usually happens because the page is not publicly accessible and requires login permissions or restricted access.
How to fix it:
- Open the document or webpage
- Update the sharing permissions so it’s publicly accessible
- Re-add the link to your Knowledge Base
Page scanned, but AI gives incomplete answers
This can happen if important information is stored inside a file (such as a PDF or image) instead of directly on the webpage. It may also occur if the page exceeds the character extraction limit, contains very little readable text, or relies mostly on visual elements rather than written content.
How to fix it:
- Move essential information into plain, readable text
- Split very long pages into multiple entries
- Expand or clarify written content
- Add additional supporting pages or text entries
Link added successfully, but AI cannot use the information
This usually means the link comes from an unsupported platform or the content on the page is not readable by the system.
How to fix it:
Replace the link with a supported public webpage, or copy and paste the content directly into a text entry in your Knowledge Base.