Why AI-generated company answers need an evidence lifecycle
AI has made it cheap to generate a plausible company answer. It has not made the answer accountable.
The unit of trust is the claim
An answer might say that a company is active, incorporated on a particular date, owned by a named parent and generating a stated level of revenue. Those are not one fact. They are separate claims, supported by different sources and updated on different schedules.
Verifying the entire paragraph with one confidence score hides that structure. An enterprise system needs a verdict for each material claim, the official value that was observed and the source that supported the decision.
Correct today can become stale tomorrow
Most verification products end when the API response is returned. That is sufficient for a transient lookup. It is not sufficient when the answer is published in a company profile, retained in an underwriting decision or used by a long-running agent.
The registry may later record a status change, a new director, a different ownership path or a fresh financial statement. The original answer was not necessarily wrong. It may have been correct when published and become stale later.
Evidence needs continuity
A defensible system must preserve the published value, the observed value, the source and the time of verification. It must then keep that record connected to subsequent changes and notify the application that relied on it.
Verification answers: “Was this supported?” The evidence lifecycle also answers: “Is it still supported?”
The operational consequence
Once evidence is structured at claim level, software can act. A verified claim can pass. A contradicted claim can be corrected. An unsupported claim can be removed or escalated. A stale claim can reopen a decision or update a published profile.
That is the purpose of CompanyProof: not to evaluate whether language sounds plausible, but to connect material company claims to official evidence for as long as the claim remains in use.