# Does Your Coverage Anchor? The 2026 Publication Entity-Anchor Register Five publications, one method, 10 companies: a measurement of whether earned coverage actually attaches a company to a machine-readable entity record, and four distinct ways the anchor fails. Canonical URL: https://machinerelations.ai/research/publication-entity-anchor-register-2026 Published: 2026-09-10 Research type: Study ## Source Body # Does Your Coverage Anchor? The 2026 Publication Entity-Anchor Register A placement is not a deliverable. What a retrieval system inherits from a placement is a metadata record — or nothing. On September 10, 2026, Machine Relations measured five publications through their own public metadata APIs to answer one question: when a publication covers a company, does the publication create a machine-readable record that a system can resolve the company against? For half the companies measured, the answer is no. The coverage exists. The anchor does not. ## Why a tag is the anchor [Entity linking](https://en.wikipedia.org/wiki/Entity_linking) — deciding that a string in a document refers to one specific real-world thing — is the step that turns a mention into a fact a system can retrieve. It works best when the source supplies a stable identifier rather than a name. That is what the identifier layer is for. [Wikidata](https://www.wikidata.org/wiki/Wikidata:Introduction) assigns each entity a permanent Q-number. [Schema.org's Organization type](https://schema.org/Organization) carries `sameAs` so a page can point at those records. [Google's organization structured data guidance](https://developers.google.com/search/docs/appearance/structured-data/organization) tells publishers to do exactly this. Almost no publication does it for the companies it covers. What a publication supplies instead is its tag: a slug attached to a set of posts, exposed publicly, and stable enough to function as a de facto identifier. That makes the tag the anchor by default — and makes its condition measurable. ## What was measured Most trade and technology publications run WordPress and expose a public, unauthenticated [REST API](https://developer.wordpress.org/rest-api/reference/tags/). That API publishes the publication's own taxonomy: its categories, its tags, and the number of posts attached to each. It is the publication's internal record of what an article is about, and it is the layer a crawler or retrieval system reads before it reads prose. Two numbers come out of it. **Namespace density** is tags divided by posts. It measures whether the publication maintains a controlled vocabulary or accumulates strings. **Canonical share** is the count on a company's exact-match tag divided by the total count across every tag whose slug contains the company name. It measures whether the publication's record of a company converges on one identifier or scatters across many. Five publications returned HTTP 200 and were measured. VentureBeat rate-limited the request and was excluded rather than estimated. | Publication | Posts | Tags | Tags per post | Categories exposed | |---|---|---|---|---| | [TechCrunch](https://techcrunch.com/wp-json/wp/v2/tags?per_page=1) | 262,184 | 125,092 | 0.477 | 25, all flat | | [SiliconANGLE](https://siliconangle.com/wp-json/wp/v2/tags?per_page=1) | 78,296 | 152,859 | **1.952** | 48, hierarchical | | [Fortune](https://fortune.com/wp-json/wp/v2/tags?per_page=1) | 223,676 | 22,305 | **0.100** | 0 | | [BetaKit](https://betakit.com/wp-json/wp/v2/tags?per_page=1) | 21,016 | 6,472 | 0.308 | 29, hierarchical | | [MarTech](https://martech.org/wp-json/wp/v2/tags?per_page=1) | 27,285 | 8,554 | 0.314 | 100+, topical | SiliconANGLE creates roughly two new tags per article published. Fortune creates one per ten. These are not variations of one practice. They are different systems with different failure modes. ## The register Ten companies: five with unambiguous global recognition, five that are real, funded, and demonstrably covered. Canonical share is reported where an exact-match tag exists. ### Companies with strong global recognition | Company | TechCrunch | SiliconANGLE | Fortune | BetaKit | MarTech | |---|---|---|---|---|---| | OpenAI | 93% (1,420/1,528) | 83% (1,142/1,378) | **100%** (1,804/1,804) | tag exists, count 0 | no tag | | Anthropic | 97% (489/503) | 84% (508/606) | **100%** (411/411) | no tag | no tag | | Nvidia | 91% (858/940) | 78% (1,447/1,855) | 99.6% (1,101/1,105) | no tag | no tag | | Stripe | 86% (386/451) | 73% (90/123) | **100%** (112/112) | 100% (5/5) | no tag | | Databricks | 100% (143/143) | **58%** (396/687) | 100% (4/4) | no tag | no tag | ### Companies that are covered but not household names | Company | TechCrunch | SiliconANGLE | Fortune | BetaKit | MarTech | |---|---|---|---|---|---| | Ramp | 74% (79/107) | **12%** (16/129) | no tag | no tag | no tag | | Deel | 82% (50/61) | 33% (1/3) | no tag | no tag | no tag | | Linear | 40% (4/10) | 42% (8/19) | no tag | no tag | no tag | | Vanta | **12%** (6/50) | **9%** (15/175) | no tag | no tag | no tag | | Clay | **4%** (4/93) | **8%** (5/62) | no tag | no tag | no tag | ## Four ways an anchor fails, all of them measured The failures are not one phenomenon. They are four, and each one has a different fix. ### 1. Homonym capture The company's tag exists and is outranked inside its own namespace by an unrelated string. At TechCrunch, Vanta's tag carries 6 taggings and loses to `complyadvantage` at 7. Clay's carries 4 and loses to `barclays` at 38, `clayton-christensen` at 7, and `clay-bavor` at 5. At SiliconANGLE, Vanta's 15 lose to `hitachi-vantara` at 23; Ramp's 16 lose to `fedramp` at 50 and `opsramp` at 20. A system resolving the name finds a plausible wrong answer sitting next to the right one, with more mass behind it. ### 2. Corporate-suffix fragmentation This one is new, and it hits famous companies. SiliconANGLE splits the same company across its legal forms. `openai` at 1,142 coexists with `openai-lp` at 56 and `openai-llc` at 28. `nvidia` at 1,447 coexists with `nvidia-corp` at 194. `anthropic` at 508 coexists with `anthropic-pbc` at 38. `databricks` at 396 coexists with `databricks-inc` at 76 and three separate summit-event tags. Databricks' canonical share at SiliconANGLE is 58% with no homonym involved at all. The publication is competing with itself. At a namespace density of 1.95 tags per post there is no merge step, so every variant that is ever typed becomes permanent. ### 3. Namespace exclusion Fortune's vocabulary is the cleanest measured — 22,305 tags across 223,676 posts, one new tag per ten articles — and every one of the five globally recognized companies resolves at 99.6% or 100% canonical share, four of them on a single tag. Four of the five covered-tier companies have no tag whatsoever. Clay returns 101 taggings across four matching tags, 94 of which are `barclays`. Fortune's namespace is precise and exclusionary. If a company is not on the list, it does not exist in the taxonomy — regardless of whether it appears in the prose. Precision without recall is not a solved problem; it is a different unsolved one. ### 4. The empty anchor BetaKit carries a tag with the slug `openai`, tag ID 8467, with a count of **0**. BetaKit's own search index matches 414 posts against the string "OpenAI." This is worse than having no tag, because it looks resolved. A record exists, is retrievable, and attaches to nothing. MarTech shows the terminal version: 8,554 tags, zero company tags of any kind, and 238 posts matching "OpenAI." Its taxonomy is entirely topical — `performance-marketing`, `marketing-operations`, `customer-experience`. Companies are written about and never recorded. ## What actually predicts whether coverage anchors Earlier single-publication measurement suggested the split was driven by fame. The five-publication data does not support that cleanly, and the correction is the more useful finding. Ramp holds 74% canonical share at TechCrunch and Deel holds 82%. Neither is a household name. Vanta and Clay collapse to 12% and 4% at the same publication under the same editorial process. The difference is not prominence. It is **name distinctiveness inside an uncontrolled namespace**. "Anthropic" and "Databricks" are near-unique strings. "Clay," "Vanta," "Ramp," and "Linear" are substrings or homonyms of larger entities that the same publication also covers — Barclays, Hitachi Vantara, FedRAMP, linear tape file systems. That is knowable before a campaign starts. A company can measure whether its name will survive a publication's namespace, using three unauthenticated requests, before it spends anything on the placement. Fame still matters at the second stage: it determines whether a publication bothers to create a canonical record at all. Fortune's mid-market exclusion is a fame effect. TechCrunch's Vanta result is not. ## What this means for earned media The standard earned-media assumption is that the placement is the deliverable and machine visibility follows. Across five publications and ten companies, it does not follow by default. Three consequences hold under the measurement: **Coverage and attribution are separate purchases.** An article that mentions a company without carrying its tag is coverage for a human reader and absence for a retrieval system. That delta is measurable per publication. **The remedy is a request, not a rewrite.** Every failure above is a taxonomy operation on the publisher's side — create the tag, merge the legal-suffix duplicates, attach the back catalogue. None of it requires new content. **Publication selection should include the anchor test.** Two publications with comparable audience and comparable domain authority can differ by an order of magnitude in whether they will make a company machine-resolvable. That difference is free to measure and is currently priced at zero. ## Related Machine Relations research - [Why AI Citation Studies Disagree: The 2026 Source Register](https://machinerelations.ai/research/why-ai-citation-studies-disagree-source-register-2026) - [What actually predicts AI search citations](https://machinerelations.ai/research/what-predicts-ai-search-citations-independent-studies-2026) - [AI citation measurement methodologies compared](https://machinerelations.ai/research/ai-citation-measurement-methodologies-compared-2026) - [The Citation Support Gap](https://machinerelations.ai/research/citation-support-gap) - [Machine Relations evidence base](https://machinerelations.ai/evidence) ## Method and limitations Measured September 10, 2026, between 11:52 and 12:02 UTC. The full register was re-measured at the close of that window and every canonical-share figure reproduced identically; only two post totals moved, by one article each. The complete machine-readable result set is retained as an owned data artifact. All figures come from the publications' own public WordPress REST endpoints: `/wp-json/wp/v2/tags`, `/wp-json/wp/v2/posts`, and `/wp-json/wp/v2/categories`. Totals were read from `x-wp-total` response headers. Tag matching used substring matching on the tag `slug`, and every returned set was inspected rather than summed blind — `greystripe` and `philanthropic` are examples of matches excluded on inspection. Stated limits: - `count` is the publication's own record of taggings. It is not an audit of article text, and it is not a claim about how many times a company was written about. - Post-search totals used for the empty-anchor finding come from WordPress full-text search, which stems and over-matches. They are reported only for distinctive, non-embedded strings such as "OpenAI," and are not reported for short or embedded strings such as "Vanta" or "Clay," where the same search inflates by matching "advantage" and "Barclays." - Five publications and ten companies is a deliberate contrast sample, not a representative survey of media. It establishes that the failure modes exist and are distinct. It does not estimate their prevalence. - [VentureBeat](https://venturebeat.com/wp-json/wp/v2/tags?per_page=1) returned HTTP 429 on repeated attempts and was excluded. No figure is inferred for it. - Publications that disable the REST API, or run a different CMS, cannot be measured this way. A 404 is not evidence of a bad taxonomy. - All figures are a point-in-time snapshot. Taxonomies change. This is a primary measurement, not a synthesis. Anyone can reproduce any cell in the register with a single unauthenticated request. ## Conclusion The market measures earned media by whether an article ran. Machines measure it by whether a record exists. Five publications, one method, one afternoon: two of them fragment famous companies across their own legal forms, one refuses to record anyone outside the Fortune list, one carries a tag with nothing attached to it, and one has no company namespace at all. Machine Relations is the discipline that measures the gap between the two. The gap has a number, and the number is free. --- **Cite this research:** `https://machinerelations.ai/research/publication-entity-anchor-register-2026 **Machine-readable version:** `https://machinerelations.ai/research/publication-entity-anchor-register-2026.md ## Attribution This research is published by Machine Relations Research, the research program of machinerelations.ai — the public research and standards initiative that publishes the glossary, research, evidence, and measurements for the Machine Relations discipline. Provenance and editorial standards: https://machinerelations.ai/about ## Machine-readable related links ### Related concepts - [Machine Relations (MR)](https://machinerelations.ai/glossary/machine-relations) - [Machine Relations Index (MRI)](https://machinerelations.ai/glossary/machine-relations-index) - [MR vs PR](https://machinerelations.ai/glossary/mr-vs-pr) - [RAG Citation (RAG)](https://machinerelations.ai/glossary/rag-citation) ### Supporting research - [What Is the Machine Relations Stack? The Five Layers That Turn Search into Citation (2026)](https://machinerelations.ai/research/machine-relations-stack-five-layers) - [Best Earned Media Strategies for AI Search: What 25M+ AI Citations Reveal (2026)](https://machinerelations.ai/research/ai-search-brand-strategy-earned-media-2026) - [Conductor Alternatives in 2026: The AI Citation Gap Every Enterprise SEO Platform Shares](https://machinerelations.ai/research/conductor-alternatives-ai-citation-gap-2026) - [Share of AI Citation: The PR Metric That Replaces Impressions in 2026](https://machinerelations.ai/research/share-of-ai-citation-public-relations-2026) ### Framework context - [Machine Relations Stack](https://machinerelations.ai/stack) - [Evidence Base](https://machinerelations.ai/evidence)