Findings · 2026-09-16 · By VSNARY | Emmanuel Orta · 0 views

The parent declared the chain. None of the locations declared it back.

One site we operate points at twelve businesses under one brand. Six of those businesses, read on their own, named no parent at all. A page-subject check now reads the relation from both sides, in the vocabulary the recovered Maps schema uses.

We operate a personal site that declares a brand and twelve businesses under it: a tree service, four fencing companies, a concrete contractor, an excavation firm, a landscaper, a reptile shop, two software products and an old experiment. The declaration is precise. The site's structured data carries a Brand node with a stable identifier and points at each business by its own identifier. A reader that parses the graph learns that every one of those twelve belongs to the same brand.

Then it reads any of the twelve. Not one of them said so. The tree service declared a membership of an arborists' association and nothing about a parent. The fencing companies declared themselves and nothing else. The relation existed from the parent side only, which is the same as a family tree in which the parent lists the children and no child has heard of the parent.

Why one-way is not enough #

The recovered Google Maps schema published in August 2026 names the relation explicitly: RELATION_MEMBER_OF_CHAIN, alongside SUBSIDIARY_OF, DEPARTMENT_OF, INDEPENDENT_ESTABLISHMENT_IN, BUSINESS_MOVED and BUSINESS_REBRANDED. A chain is an object in that schema, with its own record of what its members' categories are supposed to be. The same study measured 21 stores of one retail brand in one city and found 11 carrying one primary category and 10 carrying another. The chain declared an ideal state; its members disagreed with it in the wild.

A one-way declaration leaves the reader to resolve the other direction. It has to decide whether six similarly named home-service sites in one metro area, all pointing at one phone operation, are branches of a thing, or six things, or one thing appearing six times. Entity SEO is the work of not leaving that decision to the reader. The site is a named data source in that same schema, so what it declares about itself is not decoration; it is an observation with provenance, and an observation that is missing cannot win a conflation.

What the scanner now reads #

Since 16 September every scan carries a page-subject section. It finds the business entity the page is about, by mainEntity where the page declares one, otherwise by the entity whose URL is on the scanned host, otherwise the only business entity present. Every other business entity on the page is then classified by its declared relation to that subject, in the vocabulary above, or reported as having none. The verdict is one of four words: single, related, diluted, ambiguous. It is reported, not scored, until it has been read across the corpus.

Two readings from the first hour taught it something. The tree service's page declares fourteen Service nodes, and the first version counted each as a competing business and called the page diluted. A service is an offering of a business, never a rival to it, and the type was removed from the test. The same page declares memberOf an arborists' association, and the first version mapped that to chain membership. A trade body is not a chain; the relation is now reported as a declared membership and nothing more. Both corrections came from reading real output before shipping, which is the only place either would have shown up.

How it was fixed, and where it was not #

The fix is the parent's own Brand node, byte for byte, written as brand on each location's business entity: same identifier, same name, same URL. Nothing new is claimed; the location says what the parent already said, from its own side. It is written only when the node carries no brand or parent of its own, and only for hosts the parent page names. It was written at the edge, in the two Workers that already complete these sites' structured data, rather than in six content-management sessions.

The first deploy branded four of five. The fifth site types its subject as a plain Organization, which the Worker's business test did not match, and the verification that fetched every page on a forced cache miss found the gap before anything was reported. The test now accepts a business node or the Organization whose URL is the host, and the association node on the tree site, whose URL is elsewhere, stays untouched. The sixth site has no edge layer in front of it and remains the one location that does not declare its parent; that is stated in the table rather than left off it.

The inconsistency the reciprocity check exposed #

Once the six locations and the parent agree on the chain, the next question is whether they agree on anything else. They do not. The Denver tree service declares 37 service areas in its structured data; the Arvada tree service, the same operation one town over, declares one. A chain whose members describe their territory that differently is one the reader has to reconcile, and it will reconcile it with whatever value has priority. The service-area check reads the other half of that: whether the areas a site declares have pages behind them. Two of the four fencing sites had pages for one.

What this post does not claim #

It does not claim a chain relation improves a ranking; the recovered schema names chain membership as a signal and the researchers were explicit that no weight was recovered with it. It does not claim these six sites are now correctly modelled, only that they now say the same thing from both sides. And it does not claim the fix generalises: a business whose parent is a franchisor, a holding company or a department of something else needs a different relation from the same vocabulary, and declaring the wrong one is worse than declaring none. What it claims is that the direction matters, that we had it one way for a year without noticing, and that the check which noticed is now part of every report.

Questions this post answers

What does the page-subject section report?

Which business entity the page is about, chosen by mainEntity, then by the entity whose URL is on the scanned host, then by the only business entity present; and how every other business entity on the page relates to it, in the recovered schema's vocabulary or as undeclared. The verdict is single, related, diluted or ambiguous, and it is reported rather than scored.

Is a trade-association membership a chain?

No. memberOf an arborists' association is a declared membership and is reported as one. Chain membership is read from brand, parentOrganization, subOrganization or branchOf pointing between the two entities.

Why was the fix made at the edge?

Two Workers already complete these sites' structured data from values the pages declare. Writing the parent's own Brand node there, verbatim, reached five sites in one deploy and claims nothing new. The sixth site has no edge layer and stays open.

Does declaring a chain improve rankings?

The recovered schema names chain membership as a signal, and the researchers recovered names, not weights. This post claims only that the relation should be stated from both sides, and that six sites did not.

Related findings

Comments

Comments are read before they appear. Nothing is published automatically, and no account is needed.

Writing about this? Facts, live figures and marks — every number on that page is dated and traceable to a scan.

All findings · The dataset · How the dataset works