Maintaining accuracy in content is a fundamental requirement, especially for published material. For anyone producing or curating information, ensuring every claim is substantiated can be a time-intensive process. We are introducing a new AI agent skill designed to address this directly: fact-check. This skill is built for systematic verification, ensuring the integrity of your content before it goes live. It helps anyone who needs to confirm the veracity of information within a document, from researchers to content creators, by providing a robust system for validation.
How the fact-check Skill Operates
The fact-check skill functions through a methodical verification pipeline. First, it systematically analyzes an entire document to extract and categorize every verifiable claim. During this initial scan, it also tags any potential red flags that suggest immediate scrutiny. Once claims are identified, the skill proceeds to verify each one. This verification process is critical: it queries LIVE web sources for validation, ensuring that information is current and not reliant on potentially outdated training data. For facts derived from an internal data pipeline, it employs independent re-derivation paths to confirm accuracy, never simply re-running the original query that produced the claim. This independent verification path is a core principle. Beyond individual claims, it also checks for internal consistency across the entire document, identifying contradictions or discrepancies. After verification, the skill applies corrections directly to the text where necessary. Finally, it generates a comprehensive pass/fail report, which includes a confidence score for each claim on a 6-level scale, ranging from 'verified' to 'false'.
Verification
Rules and Scope
This skill operates under strict rules, especially concerning data-derived claims. For instance, when validating data, it ensures that an independent path is used for verification; it will never re-run the same query that initially produced a claim. This rule prevents self-validation bias. Additionally, it distinguishes clearly between authorship and mere affiliation when evaluating sources, ensuring that a source's direct expertise is properly weighed. Importantly, the skill hard-blocks any data-derived claims that cannot be adequately supported through its verification process, preventing unconfirmed information from being published. The fact-check skill can be triggered manually within a document using phrases like 'fact check this', 'verify the facts', or 'is this accurate'. It also runs automatically as a critical step before any document is published, acting as a final safeguard. It is important to note what this skill does not do: it is not intended for checking opinions, predictions, metaphors, or simply formatting citations. Its scope is strictly factual verification. A key aspect of its operation is that it never treats a fetched page or source as an unquestionable authority; all information is subjected to scrutiny.
Practical
Application and Tools Utilized
Consider a scenario where a document states: 'The recent economic growth in Region Y has been primarily driven by a 15% increase in agricultural exports over the last fiscal year.' The fact-check skill would identify this as a verifiable claim. It would then use its tools to query live government economic reports, agricultural export data from relevant agencies, and potentially news archives concerning Region Y's economy. If, for example, the skill finds that agricultural exports increased by only 8% and that manufacturing contributed more significantly to growth, it would flag the claim as incorrect, provide the correct data, and offer a suggested correction to the document. If it cannot find sufficient, independently verifiable data to support the 15% claim, it would hard-block it. This process ensures that the published content reflects accurate, externally verified information. To accomplish its tasks, the fact-check skill uses a suite of tools, including robust web search capabilities, advanced page fetching to retrieve content, direct queries to knowledge graphs ('brain queries'), and graph traversal to explore relationships between data points.
Frequently Asked Questions
Q: What types of claims can the fact-check skill verify? A: It verifies specific, verifiable claims against live web sources or through independent re-derivation for pipeline-generated facts. It does not verify opinions, predictions, or metaphors.
Q: How does it handle conflicting information or unreliable sources? A: The skill never treats a fetched page as unquestionable authority. It uses a 6-level confidence scale in its reports to reflect the certainty of its findings, and it will hard-block unsupported claims.
Q: When does the fact-check skill activate? A: It can be triggered by explicit commands like 'fact check this' or 'verify the facts' within a document. It also automatically runs before any document publication as a standard verification step.
This skill offers a systematic approach to ensuring content accuracy. It provides a reliable method for validating information, which supports the creation of trustworthy documents.
