Brand Knowledge Base — Ground Your AI-Generated Content in Real Facts

What Is the Brand Knowledge Base?

The Brand Knowledge Base is Citadex's way of teaching the AI content generator what's really true about your brand — your actual product features, your real pricing, the case studies that actually happened, the positioning you've decided on. Without it, the AI writes from generic public knowledge and tends to invent details. With it, every article, summary, and outreach piece is grounded in YOUR facts.

Internally this is called a RAG (Retrieval-Augmented Generation) layer. Every time you generate content via Citadex's multi-Agent pipeline, the system retrieves the most relevant chunks from your Knowledge Base and feeds them to the Writer model as authoritative context. The result: generated content that sounds like your team wrote it, with the correct numbers, the right product names, and zero invented features.

When You Should Use It

The Knowledge Base matters most when:

• You're about to generate content at scale — RAG is the difference between "AI-generated content that's mostly accurate" and "AI-generated content that sounds like your brand."
• Your product has unique terminology — feature names, internal metrics, certifications. The AI doesn't know these without the KB.
• You have specific numbers — pricing tiers, statistics from your own research, customer counts. Without the KB, AI will guess or invent.
• You want to track how competitors position themselves — the competitor namespace lets you index competitor pages without polluting your brand facts.

If you only ever generate one or two pieces of content total, the KB might be overkill. For ongoing content programs, it's transformative.

How to Add Sources — Three Methods

Open Citadex → left sidebar → Knowledge Base (under More Tools).

You'll see three input cards at the top. Use whichever fits your source.

Method 1: Paste a Fact or Note (Fastest)

Best for: short authoritative statements, FAQ answers, key positioning copy, internal notes.

1. Title (optional) — give the snippet a short name so you can find it later
2. Content box — paste your text (minimum ~50 characters)
3. "This is a competitor's content" — leave unchecked unless this is from a competitor (more on this below)
4. Click Add to Knowledge Base

The system indexes it in seconds. You'll see "Indexed — added N chunks" confirmation.

Method 2: Crawl a Page from Your Site

Best for: your About page, product pages, pricing page, docs, FAQ pages — anywhere the canonical truth lives on your site.

1. Paste a full URL (e.g., https://yourbrand.com/pricing)
2. Click Add
3. Citadex fetches the page, strips navigation and footers, and indexes the main content

A site URL also automatically harvests usable images from that page — those images become available in Citadex's hero image picker the next time you generate content.

Method 3: Upload a File

Best for: PDFs of product one-pagers, .docx brand guidelines, .md technical specs, .txt research notes, .html exports.

• Supported file types: .pdf, .docx, .md, .markdown, .txt, .text, .html, .htm, .csv
• Maximum file size: 8 MB

1. Click Choose file
2. Pick your file
3. Indexing takes a few seconds for short files, up to a minute for large PDFs

For scanned PDFs (images, no text layer), the upload will fail with a clear message. Run OCR locally first, then re-upload as text.

Competitor Sources — Separate Namespace

When you check "This is a competitor's content" before adding a source, that source is tagged competitor-only. The difference:

• Brand sources (the default) feed the Writer and Researcher agents — they treat this content as authoritative facts about YOU.
• Competitor sources feed only the Strategist agent — used as positioning reference (so the article knows what competitors claim) but NEVER treated as your brand facts.

This prevents the disaster where a competitor's marketing copy accidentally ends up in your article describing YOUR product.

Examples of what to mark as competitor:

• Competitor's homepage
• Competitor's pricing page
• Industry reports that compare you against others

Test Retrieval — Preview What the Generator Will See

Before generating real content, you can ask the same question your article will ask the KB:

1. Scroll to the Test retrieval card
2. Type a topic (e.g., "how does our pricing work")
3. Click Test

You'll see the exact chunks the retrieval pipeline would surface for that query. This helps you spot:

• Coverage gaps — if no relevant chunks appear for a topic you'll write about, add more sources first
• Stale information — old chunks ranking higher than new ones; delete and re-upload the canonical version
• Wrong source winning — sometimes a competitor's page is more topically relevant than your own; consider strengthening your own page

How the KB Affects Generated Articles

When you run Strategy Article generation (Prompt Tracking → Generate Strategy Article → Multi-Agent Pipeline):

1. Researcher agent retrieves up to 6 relevant KB chunks for the article's topic
2. Strategist agent retrieves an additional 4 competitor chunks (if any exist) for positioning awareness
3. Writer agent writes with both sets of chunks as context, citing your brand facts where appropriate

After the article generates, you can see exactly which KB sources grounded it: open the draft in Content Drafts → expand the article details → "Sources used from your knowledge base."

Best Practices

1. Start with your top 5–10 canonical pages. Pricing, About, top 3 product pages, top 3 FAQ entries. This covers 80% of typical content generation needs.

1. Add primary research with attribution. If you have proprietary data ("we tracked 50 brands across 8 engines for 30 days"), paste it as a snippet with the methodology so the AI can cite it accurately.

1. Refresh periodically. When your pricing or feature set changes, delete old chunks and re-upload — the cache key is content-based, so old versions don't auto-update.

1. Don't dump your entire website. Quality beats quantity. 20 well-curated sources beats 200 thin ones — the retrieval pipeline picks top-K, so noise hurts you.

1. Use Test retrieval before generating critical content. If retrieval returns nothing relevant for the topic you're about to write about, that article will read generic.

Frequently Asked Questions

Q: Is the Knowledge Base private to my project?

Yes. Every source is scoped to the specific project that added it. Other workspace members on the same project see the same KB; other projects see nothing.

Q: Will adding a competitor's URL violate their terms of service?

You're crawling a single public page (the same as any browser would). For most public marketing pages this is acceptable. For sites with explicit anti-scraping ToS or paywalled content, don't add them.

Q: How long does indexing take?

Pasted text: 2–5 seconds. URL crawl: 5–15 seconds. PDF upload: 10–60 seconds depending on size.

Q: Can I update a source without deleting it?

Site URLs have a Re-index button — useful when the source page changes. Pasted text and uploaded files don't update in place; delete and re-add.

Q: How big can my Knowledge Base get?

There's no hard cap, but practical limits apply. The retrieval pipeline picks the top 6 chunks per query, so adding more sources past ~100 doesn't significantly improve generation quality — it just adds noise.

Q: Does the KB affect my prompt tracking results?

No. The KB only affects content generation. Prompt tracking (the Rankings page) measures your actual brand visibility in AI engines based on what those engines have learned from the public web — your private KB doesn't influence what ChatGPT or Claude say about you.

Q: What if Knowledge Base isn't enabled in my workspace?

The Knowledge Base is in beta. If you see a notice saying it's not enabled for your workspace yet, contact support — we're rolling it out progressively.