Methodology · Network detection
How we detect hidden dental ownership networks
The commercial question: can your team trust — and act on — ownership links that no directory shows? Many networks keep a separate brand, website and legal entity per location, invisible to every list built from self-identification.
Short answer
DentalCensus detects ownership networks bottom-up from evidence: it compares three classes of signals between companies — shared contact infrastructure, shared web and media assets, and shared practitioners — then passes every candidate cluster through independent AI verification. Only confirmed relationships enter the dataset; 12.7% of candidate clusters were rejected at verification.
Source: DentalCensus network graph · Snapshot: August 2026 · Denominator: 125,564 analyzed US dental-industry company records · Refresh: monthly
From 125,564 records to a verified graph
Detection is a narrowing funnel: each stage discards more than it keeps, and nothing is published from an intermediate stage. All counts are from the August 2026 snapshot.
125,564 company records
US dental-industry entities analyzed: clinics, labs, brand sites, education and related companies. August 2026 corpus snapshot.
≈323,000 distinct signals
Contact, web, media and practitioner signals extracted from public evidence and normalized per company.
61,547 shared-signal pairs
Raw cases where the same signal appears on two otherwise unrelated companies — the input, not the answer.
2,160 candidate clusters
Connected groups of shared-signal pairs, each treated as a hypothesis to be tested — never published as-is.
Verified network graph
≈4,500 confirmed cross-signal links; the full relationship graph holds 19,418 records across all detection methods. August 2026.
Three classes of evidence, compared between companies
Directories look at what a company says about itself. Network detection looks at what companies unintentionally share with each other. A single shared signal is never sufficient — clusters form where independent classes of evidence reinforce one another.
What a directory sees
What detection sees
Shared contact infrastructure
The same contact channels appearing across separately branded clinics — the strongest and most common class of network evidence.
Shared web & media assets
Distinct clinic brands operating on common web properties, media channels or published materials.
Shared practitioners
The same doctors surfacing across supposedly independent practices, cross-checked against the other classes.
Every cluster is verified. Rejections are published.
A shared signal can have innocent explanations: a common marketing vendor, a regional provider, a coincidence. That is why no candidate cluster reaches the dataset directly — each one passes an independent AI verification pass that weighs alternative explanations before any relationship is recorded. Clusters that fail are rejected, not counted.
We publish the rejection rate because a detection method that never says no is not a method. Filtered-out explanations include shared marketing vendors, regional service providers and coincidental overlaps.
12.7%
of candidate clusters rejected at verification
275 of 2,160 clusters · August 2026 snapshot
Three detection methods, kept distinct
The relationship graph records where each link came from, and the classes are never blended into one score. A cross-signal inference, a fact extracted from public sources and a human-confirmed relationship are three different answers — and stay visible as such.
Cross-signal detection
Inferred — a verified conclusion from signals compared between companies. Carries cluster-level confidence.
Deep research
Extracted — a relationship stated in public sources and captured during company-level research.
Manual investigation
Observed — confirmed by a human researcher reconstructing a specific group from primary evidence.
What absence means — two different answers
For company records covered by the latest verification snapshot, an empty network view means the analysis ran and found no links. For companies or signals newer than that snapshot, it means not analyzed yet. DentalCensus keeps these states distinct on every surface — unknown is never converted into a negative finding.
Covered by snapshot
"No links found" — a real result of a completed analysis, dated August 2026.
Newer than snapshot
"Not analyzed yet" — the monthly refresh closes the gap on the next cycle.
Limits, and what we deliberately do not publish
A confirmed link narrows research; it does not replace it. Coverage depends on public evidence, confidence varies by detection method, and a derived grouping is not a legal ownership registry. The following never reaches a public surface:
- Raw evidence records: the underlying contact channels and documents stay out of every public surface.
- Clinic-level rows, addresses or contact data — public pages carry aggregates only.
- The scoring thresholds, exclusion lists and clustering mechanics that tune the pipeline — the proprietary layer that keeps precision high.
Use the network layer
See market structure the directories can't.
Keep exploring
Frequently asked questions
How does DentalCensus detect hidden dental ownership networks?
By comparing evidence between companies, not inside one profile: shared contact infrastructure, shared web and media assets and shared practitioners across 125,564 analyzed company records. Candidate clusters produced by this cross-signal comparison then pass independent AI verification, and only confirmed relationships enter the dataset. As of the August 2026 snapshot, 12.7% of candidate clusters were rejected at verification.
Why do some clinics show no network links?
Two different states are kept distinct. For records covered by the latest verification snapshot, no links means the analysis ran and found none. For companies or signals newer than that snapshot, absence means not analyzed yet — the graph refreshes monthly, so the answer arrives with the next cycle. Unknown is never presented as a negative finding.
Is a detected network link proof of legal ownership?
No. A confirmed link is an evidence-based inference from public signals, labeled with its detection method and confidence. It is market intelligence for research and qualification — not a legal ownership registry, and not a substitute for diligence.
How often is the network graph updated?
Monthly. Each detection run closes after the corresponding collection cycle, so the graph reflects signals gathered through the prior month. Every published number carries its snapshot date.
How to cite
Cite as "DentalCensus — Network Detection Methodology", with a link to this page. Counts on this page are from the August 2026 snapshot of a rolling, incremental collection; each figure states its denominator. Detection provenance (inferred · extracted · observed) is preserved in the dataset and should be preserved in citation.