5LLMs monitored
6perception metrics
30+Academy lessons
EUAI Act ready
Back to Guides
IntermediateHallucination Management

Building a Fact-Checking Knowledge Base

How to create the ground truth repository that powers accurate AI responses about your brand.

10 min readFebruary 25, 2024

What is a knowledge base?

In VectorGap, your knowledge base is the authoritative source of truth about your brand.

When AI makes a claim about your company, we compare it against your knowledge base. Matches are accurate. Mismatches are potential hallucinations.

The more comprehensive your knowledge base, the better we can assess AI accuracy and identify what needs correction.

Essential information to include

Start with facts AI commonly gets wrong:

Company basics - Founded date, headquarters, employee count, funding

Product information - Features, pricing, integrations, limitations

Customer facts - Number of customers, notable logos (with permission), industries served

Competitive positioning - What you compete with, key differentiators

Team - Founders, executives, key hires (public information only)

Timeline - Major milestones, product launches, pivots

Document structure best practices

Format documents for easy extraction:

Use clear headings - "Pricing Plans" not "Our Value Proposition"

State facts directly - "Starter plan: €49/month" not "affordable plans available"

One topic per document - Easier to update and process

Include dates - "As of January 2024, we serve 10,000 customers"

Avoid marketing language - Facts only, no superlatives

Sourcing accurate information

Pull from authoritative internal sources:

Product documentation (current version)

Pricing pages (verify with finance)

Press releases (factual content only)

Annual reports or investor materials

Legal filings (if public company)

Don't upload: Marketing collateral with soft claims, outdated documents, speculative content.

Handling sensitive information

Your knowledge base is used for comparison, not shared publicly.

Safe to include: Public pricing, announced features, published customer counts

Be careful with: Unreleased products, internal metrics, customer names without permission

Never include: Customer data, credentials, trade secrets

When in doubt, include only information you'd put in a press release.

Maintenance schedule

Outdated knowledge bases create false accuracy signals. Update:

Immediately - Pricing changes, new products, acquisitions

Monthly - Customer counts, employee numbers, feature additions

Quarterly - Full review of all documents

Annually - Major refresh with all current information

Set calendar reminders. An outdated knowledge base is worse than no knowledge base.

Ready to apply what you learned?

Run a free perception audit and see how these concepts apply to your brand.

Start Free Audit