The Knowledge Base is the source of truth of your brand in Genezio. Genezio compares what answer engines say about your brand with the Knowledge Base, and it shows which claims are correct. The Knowledge Base is a set of documents, pages, and notes that you keep current. These are the items that you show when a person asks “what is really true about your brand?”

​
Why a Knowledge Base exists

Answer engines make claims about brands all the time. Some claims are correct. Some claims are old. Some claims are completely wrong. Before, a person had to read each perception and compare it with an internal source to find the difference. The Knowledge Base does this comparison automatically. You give the truth one time. Then, the platform uses it as the reference each time it extracts a perception about your brand from a conversation with an answer engine. This is the base for an accuracy audit loop:
  1. The platform records what answer engines say about you.
  2. The Knowledge Base gives what is really true.
  3. The platform compares the two, and you see where they agree and where they do not agree.

​
Where to configure the Knowledge Base

The Knowledge Base is in Brand Settings → Knowledge Base. You configure it one time for each brand. After that, it operates in the background. The platform automatically compares each new perception from conversations with the Knowledge Base.

​
You start with one item

Your Knowledge Base is not empty at the start. When you create a brand, Genezio automatically crawls the website of the brand and adds it as the first item of the Knowledge Base. Thus, the grounded-perception checks operate from day one. They use your public site as the initial source of truth, also when you did no other configuration. After that, add to the Knowledge Base the documents, URLs, and free-text snippets that your website does not cover.

​
What types of input you can add

The Knowledge Base accepts three types of input. Thus, you can use the format that your truth has today:
  • Documents: Upload datasheets, FAQs, positioning documents, internal one-pagers, or other files that have the truth about the brand.
  • URLs: Add pages that already show your official messages (your product pages, your about page, your pricing page, partner pages).
  • Free-text snippets: Paste facts, definitions, or correction notes that are not yet in a document (“our pricing starts at $X”, “we do not offer a free tier”, “our HQ is in Berlin, not San Francisco”).
You can use all three types together. A usual Knowledge Base for a mature brand has these items:
  • Some authoritative documents.
  • Links to some canonical pages.
  • A list of free-text corrections that increases over time.

​
What to put in the Knowledge Base

Think about the claims that an answer engine must make correctly when it describes your brand:
  • Products and capabilities: What you sell and what each product really does.
  • Prices and plans: The current tiers, what they include, and what they do not include.
  • Target audience and use cases: The customers that you are for and the customers that you are not for.
  • Competitive differentiators: The messages that make you different.
  • Facts that answer engines get wrong: The things that answer engines already said incorrectly about you. This is a very good signal. Put here each item that you corrected in a sales call.
  • Official statements and brand guidelines: Positioning documents, key messages, taglines, and do-not-say lists.
Quality is more important than quantity. A small, correct, and current Knowledge Base is more useful than a large and old one.

​
How the platform uses the Knowledge Base

When you have a Knowledge Base, it supplies two functions in the platform today.

​
Grounded perceptions

The platform compares each perception from a conversation with an answer engine with your Knowledge Base. The perceptions that the platform can verify get a grounded badge with one of three states:
  • Grounded-correct: The perception agrees with your Knowledge Base.
  • Grounded-incorrect: The perception does not agree with your Knowledge Base.
  • Not grounded: The platform could not verify the perception with your Knowledge Base.
This badge shows the accuracy audit. To see how grounded perceptions appear in the platform, refer to Perceptions.

​
Reports and insights

The reports of the platform use the grounded perceptions. You can see the percentage of perceptions about your brand that are correct, incorrect, or that the platform cannot verify. You can also examine the specific perceptions that are wrong. Then, you can correct the source: your website, your documents, or the Knowledge Base. The Knowledge Base is also a source of claims for the Fact Checker agent. The Fact Checker tests if answer engines confirm these claims when a user asks about them directly.

​
Strength of the Knowledge Base

Genezio crawls the website of each brand at the start. Thus, the grounded layer is active from day one. But the strength of the signal depends on how much truth your Knowledge Base has. If your website is detailed and current, the initial crawl gives much value. Your website can have little content or mostly marketing content. It can also have no prices, specifications, or positioning details. In these cases, the platform can verify fewer perceptions, and you see more “not grounded” badges. Then, you must extend the Knowledge Base. Add the documents, URLs, and snippets that fill the gaps of your website.

​
Maintain the Knowledge Base

The Knowledge Base is a living asset. Use it as a brand book that you edit:
  • Update it when your brand changes: For example, a new product, new prices, or a new positioning.
  • Add to it when you find an incorrect perception: Answer engines can say an untrue thing again and again. Write the correction in the Knowledge Base. Then, the platform can flag the next occurrences.
  • Remove old documents: Old truth is worse than no truth.
When the Knowledge Base is more current and correct, you can trust the grounded signal more.

​
The Knowledge Base in the public API

The public API reads and writes the knowledge bases of a brand. Refer to List the knowledge bases.