Welcome Message

Welcome to my blog about Cisco Collaboration, voice, and contact center technologies.

I’m Dmytro Benda, a Cisco instructor (CCSI#33268) and collaboration specialist. Here I share practical articles, lab notes, configuration tips, and troubleshooting ideas based on real work with Cisco Unified Communications Manager, Unified CME, gateways, CUBE, and contact center solutions.

This blog is for engineers, students, and anyone working with Cisco collaboration technologies who wants clear technical content and real-world examples. I also post updates about Cisco training, my own courses, and upcoming learning activities.

If you have a question related to a post, a technology covered here, or Cisco training, feel free to get in touch.

Friday, April 24, 2026

Webex AI Agent Website Knowledge Base: How to Use Website Content for Customer Self-Service

One of the most useful recent updates in Webex AI Agent Studio is the ability to use website pages as a knowledge source for an autonomous AI agent. Cisco announced web URLs as a knowledge source on April 9, 2026, and documents website sources as a supported source type in the Knowledge section of AI Agent Studio.

Why does that matter? Because most companies already have the answers. They are sitting on the website: help pages, shipping rules, return policies, onboarding guides, clinic instructions, and product documentation. The problem is not missing content. The problem is that customers still have to go looking for it. A website-based knowledge source turns that existing content into something the AI agent can use in conversation.

That changes the role of the bot. Instead of acting like a separate system that needs its own copy of everything, the agent can work with the same customer-facing information the business already maintains on the site. Cisco also recommends keeping this focused: start with a specific section of the website, not the homepage, and expand only after reviewing the extracted content.

Where this becomes useful fast

Retail and e-commerce
A store can point the agent to pages about shipping, returns, warranty, and payment rules. That gives the AI agent enough information to answer the questions customers ask every day without sending them through five menu clicks. 

Healthcare and clinics
A clinic can use website pages that explain services, preparation steps, required documents, and opening hours. The AI agent handles routine questions first, then moves into tasks. 

Package tracking and service updates
This is another strong fit. The website can explain delivery rules, failed-delivery policies, or support procedures, while the agent can also call a live action to retrieve the current parcel status. 

What the setup actually looks like

The configuration is really straightforward. First, in AI Agent Studio, you go to Knowledge and add a new knowledge base. You need to provide its name then click Create

On the next screen click Extract Websites (or as alternative you can add source of type Websites).  

From there, Cisco documents several key settings: a Starting URL, optional URL patterns, optional subdomains, a depth limit, and a page limit.

The Starting URL is the page where the system begins reading content. Cisco recommends using a focused entry page for one topic area rather than a general homepage with mixed content.

The URL patterns setting lets you keep the source clean. In practice, that means you can tell the system to use only pages whose addresses match sections such as /help/, /faq/, or /returns/. Cisco allows up to 10 such patterns for a website source.

If the relevant content sits on more than one subdomain, Cisco also allows up to 10 subdomains to be included in the same source.

The depth limit controls how far the system follows links away from the starting page. Cisco documents three levels:
- 0 means only the starting page,
- 1 means the starting page plus the pages linked directly from it,
- 2 means one level deeper beyond that.

The page limit controls how many pages are included. Cisco documents a maximum of 250 pages for a website source.

In our demo, we connected the AI Agent to a live web page with information about our company, Knowledge Club. The bot can then use that content to answer customer questions such as “What is Knowledge Club?”, “What courses do they offer?”, and “Where are their offices located?” This is a practical example of how an AI Agent can use information from an existing, fully functional website instead of relying on content created separately for the bot.

We didn't want to include the information from the Navigation menu, Footer and Forms to this knowledge base, so we set appropriate checkboxes to exclude it. 


After the source is created,  a sync immediately starts to pull in the latest version of the site content. When the sync is completed the status of your knowledge base should be Processed and you can see the content of your page. 


Add your new knowledge base to your Autonomous AI Agent. Don't forget to save your changes and publish your bot:


Now let’s test it. Open the bot preview in chat mode and ask a few questions. You should see that the AI Agent is now using the information from your web page in its answers. Below are a few examples from our demo:



The real best practice

The best approach is not to connect the whole public website and hope the bot sorts it out.

The better approach is much simpler: choose the pages that already contain clear customer answers, keep the first scope small, review what was extracted, and update the source when the site changes. Cisco’s own guidance points in exactly that direction.

That is why this feature is useful. It does not ask teams to invent a new knowledge library from scratch. It lets them turn existing website content into practical self-service - and then combine that knowledge with live actions when the conversation needs to move from answers to outcomes. 

No comments:

Post a Comment