Tutorial: Build a Knowledge Bot for Confluence

In this tutorial, you’ll build an internal knowledge bot that your team can use to find information in your Confluence wiki instantly. Instead of searching through pages and spaces, your team just asks a question.

Scenario: Your company uses Confluence for internal documentation: engineering runbooks, product specs, HR policies, onboarding guides. Finding the right page takes time. You want a chatbot that knows everything in the wiki.

Time needed: About 20 minutes of setup. The crawl itself then runs in the background and can take a few hours on a large wiki.

What you’ll learn:

How to connect Confluence, including wikis behind SSO
How to crawl it as a website instead, when that fits better
How to supplement Confluence with PDF manuals
How to set up a private bot for internal use
How to connect via MCP so AI coding tools can search your wiki

Before you start: pick how to connect

There are two ways to get Confluence content into ChatbotIQ, and one is better in almost every case.

Use the Confluence connector. It reads your wiki through the Atlassian API using an API token. SSO is not a problem, you pick exactly which spaces are indexed, you get page hierarchy in search, and repeat syncs only re-read pages that actually changed. See How to connect Confluence for the setup, then skip to Step 2 of this tutorial.

Crawl it as a website. Only worth considering if your wiki is already readable without logging in. Open one of your pages in a private browser window: if you land on a login or SSO screen, crawling will collect nothing but login pages. Even when it does work, the crawler cannot see page hierarchy and has to re-read every page on each refresh.

The rest of Step 1 covers the crawl route. If you used the connector, go to Step 2.

Step 1: Add Confluence as a source

Confluence is a single-page application (SPA), it loads content via JavaScript after the page shell renders. This means basic crawlers get empty pages. ChatbotIQ handles this, but you need to use the right settings.

Go to Knowledge Base → ”+ Add Source” → Website.
Enter your Confluence URL (e.g., https://yourcompany.atlassian.net/wiki/spaces/DOCS).
For Discovery Mode, choose Link Crawling. Confluence sitemaps are often incomplete, so following links is more reliable.
For Configuration mode, choose Manual.
Open Advanced Options and under Site Preset, choose Confluence / Wiki.

What the preset configures for you

Setting	Value	Why
Reader Mode	On	Strips Confluence’s heavy UI chrome (sidebar, breadcrumbs, comments).
Concurrency	2	Confluence aggressively rate-limits. Higher concurrency triggers 429 errors.
Wait Time for Page Content	3 seconds	Confluence loads content via async API calls. Needs time to render.
Max Depth	10	Deep enough for most spaces.

What to adjust after applying it

Setting	Suggested value	Why
Content Area Selector	`#main-content`	Targets the article content area specifically. The preset leaves this empty.
Max Depth	15-20	Raise it if your wiki is deeply nested.
Max Pages	Start with 200	Test on a subset first, then increase.

Click Start Discovery, review the pages, then click Start Crawling.

Expect slower crawling. At concurrency 2, Confluence crawls at roughly 2-5 pages per minute. A 500-page wiki takes about 2-3 hours. This is normal, the low concurrency prevents Confluence from blocking you. The crawl runs in the background, so you can close the page.

Step 2: Add PDF supplements (optional)

If your team has documents outside Confluence (runbooks, policy docs), or if parts of your wiki sit behind a login and can’t be crawled, add them as PDF sources. See Add PDFs and Q&A Sources for the walkthrough.

Step 3: Create the internal bot

Go to Bots → “Create Bot”.
Name: “Wiki Assistant” or “Team Knowledge Bot”.
Sources: Select your Confluence source and any PDF sources.
Model: GPT-5 Mini works well for internal Q&A. If your documentation is technical and nuanced, consider GPT-5.2 or Sonnet 4.6 for better reasoning.
Personality: Here’s a template for internal bots:

You are a knowledgeable assistant for the [Company] team. You help team members find information from our internal wiki and documentation.

Guidelines:
- Be concise and direct - team members want quick answers
- Always cite the source page so people can read the full context
- If the information might be outdated, mention when the source was last updated
- For sensitive topics (HR, legal, finance), remind the user to verify with the relevant team

Click Create.

Step 4: Configure for internal use

Open bot settings:

Privacy: Keep it private

Set Privacy to Private. This means only workspace members can use the bot. No embed code needed, your team accesses it through the ChatbotIQ dashboard.

Alternatively, set it to Authenticated if you want any logged-in ChatbotIQ user to access it, or Public with Allowed Domains set to your intranet domain.

Source citations: Essential for trust

Keep Include source citations on. Team members need to verify information and read the full Confluence page for context.

General knowledge: Consider enabling

Toggle Allow general knowledge on if you want the bot to supplement wiki answers with general explanations. For example, if someone asks “What is Kubernetes?” and your wiki assumes prior knowledge, general knowledge mode can provide background context.

Step 5: Connect via MCP (for developers)

If your team uses AI coding assistants (Claude Desktop, Cursor, Windsurf), you can connect them to your knowledge bot via MCP.

MCP requires a paid plan. On the free plan, the MCP tab shows an upgrade prompt instead of the settings.

Open the bot and go to Playground, then the MCP tab.
Click With API Key. There is also a Public (no auth) option, but it lets anyone with the URL query your knowledge base, so avoid it for internal documentation.
Copy the API key straight away, it is shown only once. You can replace it later with Rotate Key.
Copy the MCP Endpoint URL, then pick your client under Client Configuration.
In Claude Desktop: Settings → Connectors → click ”+” → enter the name and URL, then expand Advanced settings to add the Authorization header with your API key.

Now your development team can ask their coding assistant questions about your internal docs without leaving their IDE.

Step 6: Test with real questions

Test with questions your team actually asks:

“How do I set up the development environment?”
“What’s the process for requesting time off?”
“Where is the architecture diagram for the billing service?”
“What are the runbook steps for a database failover?”

Check the RAG Debug panel for each answer:

Are the right Confluence pages being retrieved?
Are the relevance scores high?
Is the content clean (no navigation or sidebar text)?

If content looks noisy: Adjust the Content Area Selector. Try article or .wiki-content instead of #main-content if your Confluence theme uses different markup.

Step 7: Roll out to your team

Invite team members - go to Team and invite your colleagues. Give them Member or Viewer roles.
Share the link - send your team the ChatbotIQ dashboard URL.
Set expectations - let your team know what the bot can and can’t do. It’s great for finding information, but it’s only as good as the content in your wiki.

Keep it current

Confluence content changes frequently. Refresh your source weekly, and after major wiki updates. See Keep Content Up to Date for refresh strategies.

Configure Crawling for Your Site - all crawling recipes
Connect via MCP - detailed MCP setup guide
Keep Content Up to Date - refresh strategies