Why Machine-Readable Business Truth Needs an Owner
Machine-readable business truth is not whatever a crawler extracts or a model happens to summarize. It is the small set of facts a business is prepared to stand behind, published in forms that humans and machines can retrieve without guessing.
That truth needs an owner. Automation can collect candidate facts, compare pages, and flag contradictions. It cannot decide that an inferred service, old phone number, or vague credential should become the official public record. Someone responsible for the business must approve what is current, what is supported, and what should remain unknown.
This governance step is what turns structured content into dependable infrastructure.
Extraction is useful, but it is not approval
A website already contains many possible facts: names, locations, products, policies, hours, staff biographies, claims, and contact details. A scanner can reduce setup work by finding them. A model can help normalize them. Neither process establishes authority to publish.
Pages become stale. Marketing copy compresses important qualifications. A service-area business may show a mailing address that is not a storefront. A professional profile may mention a credential that has expired or applies only to one person. Even accurate extraction can produce the wrong operational answer when context is missing.
The safe workflow is proposal, review, correction, approval, and publication. Scanner and model output remain drafts until the owner accepts them. The distinction is practical, not bureaucratic: it prevents a machine-facing system from turning an observation into a promise.
For the core facts worth governing, see The Business Facts AI Must Agree On.
A truth profile should be compact
The goal is not to copy an entire website into a database. A useful business truth profile focuses on facts that answer engines and customers repeatedly need:
- Canonical business name and domain.
- Primary category and real offerings.
- Address, service area, or online operating model.
- Contact methods and applicable hours.
- Supported specialties, credentials, and differentiators.
- Boundaries around price, availability, eligibility, and professional judgment.
Narrative knowledge still matters. Frequently asked questions, service explanations, policies, and background can make answers more useful. That material belongs in an approved business corpus with source and version context, not in an ungoverned pile of generated copy.
Boundaries are part of the truth
Many business errors are not incorrect facts. They are confident answers given beyond the available facts.
A contractor may not quote a project without an inspection. A clinic may not offer medical advice through a public answer layer. A restaurant may have changing availability. A regional provider may serve one county but not the next. If those limits are absent, a fluent system may fill the gap with a plausible answer.
An owner-approved truth layer should state what can be answered, what requires confirmation, and what must be refused. Unknown, unavailable, and needs-human-review are healthy operating states. They protect the customer and the business from false precision.
This is why a Site Agent should not be described as a general website chatbot. A useful Site Agent is a controlled, machine-readable truth and answer layer built from approved facts and explicit boundaries. Building a Safe AI Agent for a Local Business explains that narrower job.
Publication should create an immutable release
Drafts change while people work. Public truth should not change accidentally with every edit.
A strong publication model freezes the approved facts, boundaries, instructions, and selected corpus version into one release. External machine interfaces answer from that release until the owner approves and publishes another. A draft correction can be previewed without silently changing what is public.
Immutability also improves diagnosis. When an answer was wrong on a particular date, the business can identify which release was live and which evidence supported it. Without a version boundary, teams end up comparing a current draft with a historical observation and cannot explain the difference.
Owned truth is the beginning, not independent proof
An owner-approved profile answers, “What does this business say is true?” It does not answer, “What does the independent public web corroborate?”
That second question belongs to authority measurement. Profiles, reviews, credentials, publishers, and citations must be captured and qualified independently before they can support a trust conclusion. An owned truth layer gives those checks a stable entity to match, but it cannot award itself outside credibility.
The next article in this series, Independent Corroboration Is Not an Owned Website Claim, separates those responsibilities.
Treat business truth as an operating asset
Machine-readable truth is not a file that gets published once and forgotten. Hours change. Service areas expand. Policies mature. Credentials renew. New questions reveal missing knowledge.
The owner needs a repeatable process: review proposals, resolve conflicts, approve material changes, publish a new release, and verify that machines can retrieve it. That process makes the business easier to understand while preserving accountability.
The technology matters, but the ownership model matters more. A business should always be able to answer three questions: what is public, who approved it, and which version produced the answer.