The direct answer
Fix one canonical record for each entity (the business, each product, each location): its name, stable identifier, core facts, and relationships. Then extract the structured data from every page that describes that entity and compare, field by field. Every mismatch is a finding. The audit's output is a disagreement list with a source for each item, not a score and not a syntax report.
Syntax validation still matters, and it is a different check. A validator confirms one page's markup is well-formed. The consistency audit confirms the well-formed pages agree with each other, which is the property machines rely on when they resolve your pages into one entity.
Run the audit in five passes
Inventory the entities
List what the site describes: the organization, people, products, services, locations, and the pages that describe each one.
Fix the canonical record
Decide the correct facts from authoritative sources (legal name, address of record, official product names) before comparing anything.
Compare page by page
Extract each page's structured data and diff it against the canonical record, including identifiers and sameAs targets, not just visible fields.
Classify each disagreement
Separate typos from template errors from genuine changes nobody propagated. The class decides the fix.
Correct at the source
Fix the model, template, or feed that produced the error, then re-extract and confirm the disagreement is gone everywhere.
The disagreements that matter most
- Business name variants across pages, including old names surviving a rebrand in some templates
- Different identifiers, or different sameAs targets, offered for the same entity
- Address or phone differences between contact pages, location pages, and footer schema
- Product names, prices, or availability in schema that differ from the visible page
- Ratings or review counts in markup that the visible page does not support
- The organization typed differently on different pages, or described with conflicting founding and leadership facts
Order findings by resolution risk, not by count. A wrong identifier does more damage than ten punctuation differences, because identifiers are what machines merge on.
The Entity & Knowledge Graph hubWhat to fix first, and how to keep it fixed
The audit produces a list. The list is not the work. The work is deciding the order, and the order is set by what a system does with the disagreement. Identity first: the business name, the address, and the phone number, because every other description hangs off those three. Then the products and services, because a wrong service description sends a customer to the wrong offer. Then the supporting pages, which matter only after the core descriptions agree.
Fix each disagreement at its source, not on the page where it was found. If the name is wrong on six pages, the fix is not six edits. The fix is one edit to the canonical record, and then the six pages are corrected from it. A canonical record is a single entry that states the business name, address, phone, hours, services, and the links that confirm them. Every page, listing, and profile is checked against it. Without a canonical record, the same disagreement returns the next time anyone edits a page.
The audit is a comparison, and a comparison is only useful if it is repeated. Run it again after any rebrand, any new service, any site change, and on a fixed cadence — quarterly at minimum for an active site. SchemaGraph examines the graph the pages describe, which is where cross-page disagreement becomes visible; a syntax validator only checks that the markup parses, which is a different question. The audit answers whether the pages agree with each other and with the canonical record.
Drift is the normal state of a site that changes. The goal is not to reach a day with zero disagreements and stop. The goal is a routine that finds them early, fixes them at the source, and keeps the canonical record as the place where the business is described once, correctly.
The canonical record is not a document that is written once and filed. It is a living entry, updated whenever the business changes — a new service, a new address, a new phone number — and the update is the moment the audit is re-run, not a separate task for later. The discipline is small: change the record first, then let the pages follow. A team that changes the pages first and the record later will always be auditing its own edits.
Common questions
How often should this audit run?
After any redesign, migration, or rebrand, and on a regular cadence for facts that change: offers, locations, people. Templates drift even when nobody intends them to.
Does inconsistent schema trigger a penalty?
The cost is quieter than a penalty. Machines merge conflicting descriptions cautiously, so the entity is understood more weakly and represented less reliably, including in AI-generated answers.
What tooling does the audit need?
Per-page validators check syntax. The comparison step is the audit itself: SchemaGraph examines the graph your pages describe, which is where cross-page disagreement becomes visible.