The Entity Void: The Knowledge Graph Barely Knows These Businesses Exist

Forty-eight local leaders, twelve familiar names, and one method

Over August 2026, Raveneye Global measured eight local service trades, chartered accountant, dentist, gym, interior designer, digital marketing agency, coaching institute, physiotherapist, and wedding photographer, across six ordinary, non-metro Indian cities: Jaipur, Indore, Lucknow, Surat, Kochi, and Nagpur. Data came from live APIs: Google SERP and AI Overview data, Google Places, Google Ads search volume, the Google Knowledge Graph Search API, Google PageSpeed Insights, and DNS-over-HTTPS. This piece is one cut of that dataset, built entirely on the Knowledge Graph Search API results.

On August 10, 2026, Raveneye Global ran a targeted probe against that one instrument. For each of the eight trades in each of the six cities, 48 city-trade pairs in total, the business identified as the review leader, using the same Google Places data pulled across this field study, was submitted to the Knowledge Graph Search API by name. A second, parallel probe submitted 12 business names a large share of Indian internet users would likely recognize on sight: IndiaMART, Apollo Hospitals, MakeMyTrip, Livspace, Sulekha, Urban Company, Justdial, Aakash Educational Services, Practo, Zomato, Cult.fit, and Clove Dental. A business counted as resolved only when the API returned an entity whose name matched the query; an empty response, or a response naming something else entirely, counted as no match.

The two probes split sharply. Of the 48 local review leaders, one resolved to a matching entity, about 2%. Of the 12 familiar national names, roughly eight resolved, close to two-thirds. Both figures carry the same qualification, stated plainly here because it governs everything that follows: this is a single-day, name-match probe of one public interface, not a repeated measurement and not a claim about everything Google's systems privately know. What it measures, precisely and only, is whether one specific, documented, developer-facing entry point into the Knowledge Graph can name a given business back when asked.

The Knowledge Graph Search API was chosen for this cut of the study because it sits closest to the newest layer of discovery, the one where a generative system decides not which links to show, but which business to name outright. Ranking in a list of results and holding a well-reviewed listing are older, better understood surfaces, and this field study measures those elsewhere. Being a resolvable entity in the graph a machine consults before it commits to naming something in an answer is the newer, less understood layer, and the one this piece measures directly.

What a knowledge graph is, and why an entity is now the unit that matters

A knowledge graph, in Google's own framing, is not a list of web pages. It is, in the phrase the company used when it introduced the idea on the Official Google Blog on May 16, 2012, a shift from strings to things: from matching the text of a query against the text of a document, to recognizing that a query names a specific, real entity, a person, a place, a business, and answering from what the system knows about that entity directly. Google described the Knowledge Graph at launch as holding roughly 500 million entities and more than 3.5 billion facts and relationships connecting them. An entity, in this sense, is not a name. It is a node the system is confident exists, with attributes attached to it and corroborating sources behind it.

That distinction has only grown more consequential since. Google's Knowledge Panel documentation describes the Knowledge Graph as the company's database of billions of facts about people, places, and things, compiled from a mix of open-web sources, licensed data, and information supplied by an entity's own representatives, and states that a panel is created automatically, algorithmically, when there is enough information available on the open web. The same entity layer that has generated knowledge panels since 2012 is the layer generative systems increasingly draw on to decide what to name in an answer rather than merely what to list. A business that has never accumulated enough corroborating information to become an entity in this sense is not simply missing a knowledge panel. It is missing the more basic thing a knowledge panel depends on: a system confident the business is a distinct, nameable, real-world thing at all.

One match in forty-eight, and what the other forty-seven looked like

Forty-seven of the 48 local review leaders in this sample returned no matching entity. That is the finding this piece takes its name from: an entity void, not a gap. The businesses involved were not marginal. Each was the review leader, by the same measure used across this field study, in its own trade in its own city, an actual market leader by the metric that matters most to a paying customer. None of that standing translated into recognition by the API standing in for the Knowledge Graph.

The non-matches took two distinct forms, and the difference is worth stating precisely, because it changes what a miss actually means. Most returned nothing: an empty response, the plain outcome when a system has no candidate entity to offer. A smaller number returned something, just not the business asked about. A chartered accountancy firm named T.Nagar and Co. returned the entity 'Crime and Punishment,' the Dostoevsky novel, the query apparently sharing no meaningful ground with the firm beyond enough surface text for a low-confidence system to surface some answer rather than none. A dental clinic named Dental Seva returned 'Dental implant,' a generic medical concept rather than the clinic itself, a closer but still wrong answer, one step removed from useful and still not the thing that was asked about.

Both examples reveal something about the mechanism itself. The Knowledge Graph was built, in Google's own 2012 framing, to move past exactly this kind of behavior, an engine matching on shared text rather than resolving to the thing actually meant. A dental clinic returning a generic medical term, or an accountancy firm returning a nineteenth-century novel, is a graph falling back, at the edge of its coverage, to something close to the string-matching logic it was designed to replace. That is not a system reporting that it does not know a business. It is a system reaching for the nearest thing it does know and offering that instead, a distinction that matters, because a business owner reading a single headline number might assume a miss simply means silence. In the cases sampled here, it sometimes meant the graph guessing.

Composition matters too. The 48 queries spanned all eight trades measured across this field study, in each of the six cities, and no single trade or city concentrated the misses; the void was even. A gap this consistent across eight unrelated trades and six separate cities is more likely to describe something structural about how these businesses accumulate an online footprint than something incidental about any one market or profession.

The one match was never a local business to begin with

The single resolution in this sample of 48 belonged to Livspace, the national interior design and renovation marketplace that operates across Indian cities. In the interior-designer vertical, in one of the six cities measured, Livspace's own local presence was the review leader by the same measure applied everywhere else in this study, and when its name was submitted to the Knowledge Graph Search API, it returned a matching entity. Nothing about that result is surprising once it is named plainly. Livspace is not a small, single-location, independently owned business. It is a large, heavily documented national operator, and it was submitted a second time, independently, as one of the 12 familiar national names tested in the parallel probe, where it also resolved. The same entity answered to both queries, which is exactly what a well-formed Knowledge Graph entry is supposed to do.

That double appearance is why this piece treats the raw 2% figure as generous rather than as the real finding. A rate built on a single match, and that match a national brand that happened to out-review the independents in its category in one city, is not evidence that roughly one in fifty local businesses in India has an entity of its own. It is evidence that a national brand, wherever it turns up, brings its national-scale entity with it, and that not one of the remaining 47, and by extension not one of the genuinely independent, single-location, locally owned businesses this sample actually set out to measure, resolved to an entity at all. Read that way, the real number for local, independent businesses in this sample is not 2%. It is functionally zero.

The gap is not only local: four familiar brands also returned nothing

The comparison set produced misses too, worth reporting with the same directness as the misses on the local sample. Eight of the 12 national brands and aggregators tested resolved: IndiaMART, Apollo Hospitals, MakeMyTrip, Livspace, Sulekha, Urban Company, Justdial, and Aakash Educational Services all returned a matching entity. Four did not. Practo, Zomato, Cult.fit, and Clove Dental, four names with a substantial national footprint and extensive independent press coverage in their own right, returned no matching entity in the same probe, run the same day, with the same method.

That result complicates any simple story. It would be convenient, but not defensible, to conclude that Google's Knowledge Graph simply does not know brands of that scale. The more defensible reading, given what Google's own documentation says about the instrument, is that the public Knowledge Graph Search API is a narrower, less complete window into Google's entity knowledge than the graph believed to sit behind knowledge panels and AI Overviews internally. A miss on this specific API is not proof an entity does not exist anywhere in Google's systems. It is proof that this particular, publicly documented probe could not confirm one, for a brand as recognizable as Zomato exactly as it could not for a single-location dental clinic. The uncertainty runs in one direction, and it favors caution: a resolved entity here is strong, positive evidence; a non-match is weaker evidence of absence than it would be from a more complete instrument, which is precisely why every measured figure in this piece carries the emerging tier rather than the established one.

The full split, stated plainly:

• Resolved: IndiaMART, Apollo Hospitals, MakeMyTrip, Livspace, Sulekha, Urban Company, Justdial, Aakash Educational Services.
• Did not resolve: Practo, Zomato, Cult.fit, Clove Dental.

Why an engine cannot cite what it cannot name

The practical stakes of this gap sit downstream of the mechanism, in the moment a generative system actually assembles an answer. A 2024 peer-reviewed study that coined the term Generative Engine Optimization tested which properties of a source changed whether it got cited inside a generated answer, and found that adding cited statistics, quotations, and authoritative sourcing measurably raised a source's likelihood of appearing in the answer. The thread connecting that finding to this one is corroboration. A system deciding what to name, whether a search engine building a knowledge panel or a generative model composing an answer, favors what it can verify from more than one place over what it can only read on a single page. An entity is, at bottom, a bet a system is willing to make that a real, distinct thing exists behind a name. A business the Knowledge Graph has never resolved has not yet earned that bet, regardless of how it performs on surfaces that do not require one, a full calendar of reviews, a strong position in a list of links.

This is also why a strong position in a list of links does not substitute for entity status. A ranking is a judgment about relevance to a query. An entity is a claim about existence and identity, verified enough for a system to stake its own credibility on naming the thing outright rather than merely pointing to a page about it. Generative answers increasingly make exactly that kind of claim, naming a business inside a sentence rather than listing a link a person can click through and judge for themselves. The businesses in this sample that have not yet cleared that bar are not being ranked poorly. They are, from the graph's perspective, not yet a thing it is willing to name.

What accumulates into an entity, and what remains unproven

Google's own documentation ties automatic entity and knowledge panel creation to there being enough information available on the open web, language that points toward corroboration rather than any single action a business can take. The mechanisms the industry has built around that idea are documented and public: schema.org's Organization and LocalBusiness structured-data types, which Google Search Central's own guidance recommends implementing with the most specific subtype available, and the sameAs property, which schema.org defines as a way to point a page or a listing at the URL of the item's Wikipedia page, Wikidata entry, or official website, a machine-readable thread tying a business's separate profiles back to one identity.

None of that is a guarantee, and this study did not test whether applying it changes a specific business's outcome in a future Knowledge Graph Search API probe. Doing so would require a second, longitudinal measurement this piece does not have. What the evidence here supports is narrower and more precise: the businesses that did resolve, the national brands with a large public footprint, press coverage, and listings across multiple platforms that agree with each other, are also the businesses with the most corroborating information for a system to find. That correlation is consistent with Google's own stated criterion. It is not, on its own, proof that any single fix moves a specific business from one side of this line to the other.

It is also fair to note what a repeat of this probe might show. Google updates the Knowledge Graph continuously, and the public API layered on top of it is itself openly described by Google as being migrated toward a newer enterprise product. A business absent from this snapshot in August 2026 is not thereby permanently absent, and a second probe run months later, on the same businesses, would be the correct way to test whether any specific intervention moved the needle. This piece does not have that second measurement, and says so rather than implying one.

What this leaves is a measurement, not a verdict. On August 10, 2026, the businesses already leading eight local trades across six Indian cities were, with one exception that was not really an exception, absent as entities from the public interface Google itself documents as a window into its Knowledge Graph. That is a snapshot of a system that updates continuously, not a permanent state, and the next step is not a guarantee that any specific action changes it, but a measured read of where a given business stands today, across this surface and the others this field study has examined.