Fact Checker agent

The Fact Checker agent tests if answer engines confirm specific claims about your brand, such as prices, features, or certifications. It is the third main axis of measurement in Genezio, together with Visibility and Recommendation. Visibility asks how frequently you occur. Recommendation asks how frequently the AI selects you. The Fact Checker asks a different question:

When answer engines talk about your brand, are their facts correct?

You give specific claims about your brand that a person can verify. Examples are pricing, features, founding year, customer counts, certifications, and all other items that can be true or false. The platform tests the claims on each supported answer engine.

​

Why AI fact checking is necessary

Answer engines hallucinate. They mix old information with current information. They confuse a brand with similar brands. For most brands, this is a small problem. For some brands, it is a real risk. Examples are security companies, financial services, regulated industries, healthcare, and all companies whose product details must be accurate. Before the Fact Checker, you did not have a method to monitor in a systematic way what answer engines say about specific facts. You learned about a hallucinated claim only when a customer mentioned it, or when a person asked the correct question by chance. The Fact Checker changes this into a measurement. You make a list of the claims that are important to you. The platform gives a result for each claim on each answer engine. The answer engine confirms the claim, contradicts it, or does not give an answer about it.

​

How the Fact Checker agent works

A Fact Checker topic has a set of claims to verify. These are claims that you say are true about your brand. For each claim, the platform runs conversations with each supported answer engine. It puts the answer into one of three results:

• True: the answer engine confirms the claim.
• False: the answer engine contradicts the claim.
• Indecisive: the answer engine does not take a clear position.

The result applies to one claim and one answer engine. Thus, ChatGPT can confirm a claim, Perplexity can contradict it, and Claude can ignore it. The platform shows all three results.

​

Where the claims come from

You can add claims to a Fact Checker topic in two ways.

​

Claims from the Knowledge Base

Get claims from the Knowledge Base of your brand. The Knowledge Base has the documents, URLs, and snippets that are already your source of truth. This is the usual method after you add content to your Knowledge Base. The truth is already in the documents, and the Fact Checker only measures if answer engines show it.

​

Custom claims

Type claims directly on the Fact Checker topic. Custom claims are useful in these cases:

• One-time measurements, for example: “does ChatGPT know we acquired Company X last quarter?”
• Tests of claims that are not in the Knowledge Base yet
• The monitoring of specific high-risk facts, for example a certification, a regulatory status, or a pricing detail

You can use the two methods on one topic. Some claims can come from the Knowledge Base, and you can type other claims directly.

​

Where you see the Fact Checker results

​

On the conversation

Each conversation has a Fact Checker tab. The tab shows which claims the answer engine stated and which claims it did not state. Each card has a clear CLAIMED or NOT CLAIMED label. Thus, you can quickly examine one conversation.

​

On the topic and scenario drawers

Open the View More drawer on a topic or on a scenario. The drawer shows the Fact Checker charts. One chart is “Claimed metric by answer engine”. It shows which answer engines support your claims and which answer engines do not. Use this view for reports to management. It gives a snapshot of which facts the answer engines confirm.

​

Effect of the Fact Checker on the KPIs

Fact Checker conversations do not count for AI Visibility or AI Recommendations. The prompt names the brand and the claims explicitly. Thus, the conversations do not measure organic discovery. They measure accuracy. In the KPIs, these conversations make the metrics incorrect. The Fact Checker is a separate measurement tool, as are the Introspector and the Comparer. It is separate from visibility and recommendation, but it is equally useful for the correct brand.

​

The accuracy stack: Knowledge Base, perceptions, and Fact Checker

The Fact Checker is the last part of the accuracy stack of Genezio:

1. Knowledge Base: defines what is true about your brand.
2. Grounded Perceptions: show if the claims from organic conversations match your Knowledge Base.
3. Fact Checker: tests if answer engines confirm specific claims when a user asks about them.

Together, these give you a complete view of accuracy:

• The Knowledge Base is the source of truth.
• Grounded Perceptions measure what answer engines say about you without a question about the claim.
• The Fact Checker measures what answer engines say when a user asks directly.

Grounded Perceptions observe. The Fact Checker probes. When you read them together, you know the narrative that answer engines give without a request. You also know the answers that they give when a user asks directly.

​

When to use the Fact Checker

The Fact Checker is most valuable for brands where factual accuracy has a direct effect on buyer trust or regulatory standing:

• Security and financial services: incorrect statements about certifications, compliance, or features can decrease trust.
• Healthcare and regulated industries: claims about products have legal weight.
• Brands that answer engines frequently confuse with competitors: answer engines mix your facts with the facts of another brand.
• Brands after a major change, for example new pricing, an acquisition, or a rebrand: old information stays in answer engines for months.
• All brands that are concerned about hallucinated misinformation: the Fact Checker changes that concern into a measured signal.

​

Typical Fact Checker workflow

​

Related pages

• Knowledge Base
• Perceptions
• Prompter agent
• Recommender agent
• Introspector agent
• Comparer agent

​

Fact check data in the public API

The public API reads the fact check results over time, by topic, and by answer engine: Fact check over time, Fact check by topic, and Fact check by engine.