Your locations drift apart in the schema, not in the copy.

Google's revised AI guidance names structured data as something a person must review. For a business with multiple locations, that is the layer where branches quietly stop describing the same company.

· 9 minute read

On October 1, 2026, Google revised its guidance on generative AI content, and most readers took one instruction from it: check AI writing before you publish. The more consequential edit is a list. Google says the review covers metadata too, and it names title elements, meta descriptions, structured data and alternate text for images.

Google's October revision puts structured data under human review

Google's generative AI content guidance, updated October 1, 2026, tells site owners to manually fact-check and review all AI-generated content for accuracy and trustworthiness before publishing. The same passage extends that review to metadata and lists the items by name: title elements, meta description elements, structured data, and alternate texts for images. It adds that structured data should follow Google's general guidelines and the policies for each search feature, and that the markup should be validated.

The common reading of this update is about prose: a human editor should stand between a model and the publish button. For a business with many branches, that reading misses the part that matters. Visible copy on a location page already gets read, by the regional manager, by the franchisee, by customers who call when the hours are wrong. The four items on Google's metadata list are different. They sit in the page source, they are produced in bulk, and on most multi-site businesses nobody is assigned to read them. Google's guidance now names that unread layer as something a person is responsible for.

Schema drifts because nobody reads it

Schema drifts between location pages because visible copy has an audience and structured data does not. A wrong address in a paragraph gets noticed within days. A wrong business type in a block of JSON can sit for years, because the page looks identical with or without it.

Multi-location sites also accumulate their markup in layers. The first fifty location pages come from one template. The next thirty come from a redesign, a different plugin, or a franchisee who runs a separate site. A store locator generates a third set. Then a generative tool is asked to fill in descriptions, titles and markup for new branches, and it produces output that is plausible for each page and consistent with none of the others. Each layer was reasonable on the day it shipped. Nobody compared them to each other, because there was no reader for whom the comparison was visible.

Google's own Local Business documentation sets the expectation that this layer is per branch: it says to define each local business location as a LocalBusiness type, and to use the most specific sub-type possible. Every branch carries its own machine-readable statement of what it is. A hundred locations is a hundred statements, and the brand has usually reviewed none of them side by side.

Two valid blocks, two different businesses

Two location pages from the same dental group can look identical to a customer and describe two different businesses to a crawler. The blocks below are an illustration, trimmed to the six identity fields that define a branch: @type, name, parentOrganization, sameAs, telephone and branchCode.

Branch A: {"@type": "Dentist", "name": "Brightwater Dental", "parentOrganization": {"@type": "Organization", "name": "Brightwater Dental Group"}, "sameAs": ["group profile link", "group profile link"], "telephone": "Branch A front desk", "branchCode": "NORTH"}

Branch B: {"@type": "LocalBusiness", "name": "Brightwater Dental Care of Eastside LLC", "telephone": "regional call center"}

A customer opening both pages sees the same header, the same brand, the same services and a different address, which is exactly what should differ. In the source, Branch A declares a Dentist that belongs to the group, points to the group's official profiles and carries its own branch code. Branch B declares a generic LocalBusiness under a legal name, with no parent, no profile links, no branch code and a call center for a phone line. Every field Branch B has is filled in and syntactically valid, and a validator would pass it.

Schema.org defines LocalBusiness as a particular physical business or branch of an organization, and gives it parentOrganization and branchCode for exactly this job. Branch A uses that vocabulary to say it belongs to something. Branch B, in the only layer a machine can parse without guessing, says it stands alone.

Check two of your own location pages in two minutes

The drift check on a multi-location site needs two location pages and a browser. Open each page, view the source, and search for application/ld+json. Copy the block whose @type is LocalBusiness or one of its subtypes from each branch, and set the six identity fields side by side: @type, name, parentOrganization, sameAs, telephone, branchCode.

Then read the pair against two references. Google's Local Business documentation says to use the most specific subtype possible, so two dental branches should both declare Dentist. Schema.org treats a LocalBusiness as a branch of an organization, so every branch should name the same parent and point to the same official profiles. Brand name, parentOrganization and sameAs should match across branches; telephone and branchCode should differ, and each should belong to that branch. If view source shows no block at all because a store locator writes it in the browser, record that as well. A location page a crawler cannot read is a location that crawler cannot place.

Drift changes the entity, not a value

A side-by-side comparison of location pages lines up the same fields across every branch and reports where they disagree. Read that way, schema drift is a different business, not a typo. A typo is a difference inside one value: a transposed digit, a misspelled street. Drift is a difference in which entity the page claims to be. When two location pages disagree on business type, name, parent organization and outbound identity references, there is no single field to correct. The pages are making two different claims.

Those claims carry weight. Google describes structured data as something it uses both to understand a page and to gather information about the wider world, including the companies the markup names. Structured data is read as a statement about an entity, not as decoration on a page.

A validator checks one page, and Branch B passes it. Only a side-by-side read of the identity fields, branch by branch, catches two valid pages that disagree.

Other signals do not excuse markup that misstates a branch

The strongest case against treating schema drift between locations as serious is that search and AI systems do not depend on markup alone. Google has other evidence about a branch: its business profile, the visible text, links, and years of accumulated data. Google's own structured data policy says a manual action for a structured data issue removes eligibility for rich results and does not affect how the page ranks in web search. Language models read the visible prose as well as the source. A mismatched business type on one location page may be reconciled silently.

The answer starts with Google's own standard, not with penalties. The same policy requires structured data to be a true representation of the page content. Branch B's page shows a branch of Brightwater Dental; its markup names a standalone LLC with no parent. That markup fails the standard whether or not anything is penalized, and other signals smoothing it over do not make it accurate. When the on-page signals agree, a machine has nothing to resolve. When they disagree, the business has handed the decision about what its branch is to a system it cannot inspect. The October 2026 revision removes the remaining excuse: Google now says in plain terms that generated structured data is something a person reviews. Drift is the part of this problem a business fully controls, and it lives on its own pages.

A drift report states what it read

A drift report across a brand's location pages states its coverage as N of M sampled, and lists every location page it could not read as a finding of its own. It reports what each page publicly declares, not what any AI model answered about a branch, and that is enough to show whether the branches still describe one business.

A typo is a difference inside one value; drift is a difference in which business the page claims to be.

Google's October 2026 revision will mostly be remembered as a note about editing AI prose. For a business with multiple locations, the practical instruction is narrower and more useful: the title, the description, the alternate text and above all the structured data on every location page are published statements about who that branch is, and somebody now has to read them. Reading them one page at a time will find the broken ones. Reading them against each other is the only way to find the valid ones that no longer agree.

Further reading

  1. Google Search Central, general structured data guidelines, the requirement that markup truly represent the page and the scope of a structured data manual action developers.google.com/search/docs/appearance/structured-data/sd-policies
  2. Schema.org, the LocalBusiness type definition, which covers a branch of an organization and carries the parentOrganization and branch code properties schema.org/LocalBusiness