Search Engine Land counts 72 Geostore ranking signals in a recovered Google binary, 25 deprecated

A count of signals is a count of names in a schema. It says what the system can refer to, and nothing about what any one of them does to an ordering. The examination puts the count at 72 Geostore ranking signals, of which 25 are marked deprecated ^[3][5]. That leaves 47 live names and about 35% of the vocabulary already retired ^[15][16]. Twenty-five dead entries in a long-lived codebase is unremarkable.

Provenance matters more than the totals here. Researchers obtained a binary exposing a non-public Geostore scope, then cross-referenced it against Maps protocols, network traffic, the web index, mobile services, style tables, on-device components and material from Google's 2024 leak ^[2]. Each identification is an inference from structures in shipped code, corroborated across those other surfaces. Search Engine Land says the work is not an official Google disclosure and does not reveal the weights that determine Maps rankings ^[6].

The data model changes the work. The analysis describes a local listing as a geographic Feature that can carry identity, geometry, source information, websites, brand relationships, Knowledge Graph references, concepts and ranking information ^[9]. External references reach the entity through the Knowledge Graph and a webref layer that links documents to entities ^[10]. If a document can be attached to a Feature, a page stating what a business does and where it operates is an input to the same record as the field a manager edits in the profile.

For the 72 to become a task list, three things would have to be published: which of the 47 live signals read inputs a business can change, the weights, and where in the pipeline each one is applied. The analysis places the signals in Oyster Rank, described as an internal ranking vocabulary for Geostore. That vocabulary sits inside a pipeline that also covers query understanding, semantic matching, candidate generation, geography and quality assessment, and reranking ^[8]. A signal applied at reranking does nothing for a place that never entered the candidate set. Search Engine Land says the numbers should not be read as a list of 72 tactics or a confirmed Maps ranking formula ^[7].

What the analysis actually recommends is cheap: consistent name, address, phone, website and attributes; website text explaining what the business is and which services are associated with that location; accurate brand associations and external entity references; and important details stated clearly in pages Google can connect to the entity instead of implied through marketing language ^[12]. It also holds that accurate core Business Profile information remains essential ^[11]. The recovered data does not provide coefficients, weights or a direct causal ranking model, and the analysis says it is not proof that any single change will improve a Maps position ^[13].

The claim that a polished profile with thin supporting information may be harder for an entity system to interpret than a consistently documented one is untested in this material ^[14]. I would still take that bet for a small local site, because agreement between the site, its citations and its brand references survives whatever the weights turn out to be. The recovered scope also names 793 data source providers ^[3], and Search Engine Land's framing is that the profile matters while being only one visible input into a much broader representation of the place ^[17].