Glossary / Entity confidence

Entity confidence

Also: Entity resolution readiness, Entity confidence score

Entity confidence is a free diagnostic on every graded Lantad report that scores, out of 100, how completely the JSON-LD in your crawler-visible HTML declares one identity and how far that identity is anchored by identifiers which resolve outside the page.

  • Identity declared 35 pts the gate: type, name, absolute url
  • External identifiers (sameAs) 25 pts 16 of 24 sampled pages
  • Stable identifier (@id) 15 pts 11 of 24 sampled pages
  • Logo declared 15 pts 19 of 24 sampled pages
  • Category declared 10 pts 9 of 24 sampled pages
The five signals and their weights, read from core/src/entity.ts. They sum to 100.

Entity confidence answers a narrower question than it sounds like: not whether an AI system understands your brand, which nobody outside that system can observe, but whether the structured data a no-JavaScript crawler receives declares one identity clearly enough for anything to be said about it at all. It is read from the same raw-HTML fetch as Prose Parity and Schema, so it costs nothing extra to run and it never scores markup a crawler would not receive.

It is diagnostic. It appears on every graded report, free, and it never feeds the composite AI Visibility Score, exactly like Retrievability. It carries no letter grade, because grade bands calibrated for a continuous composite manufacture precision when they are applied to a five-item checklist. It carries a tier instead, and every tier is printed with the sentence that defines it.

What it actually reads

Lantad parses every application/ld+json block in the HTML the crawler fetch returned, descends nested property values as well as @graph arrays, and looks for an identity node: an Organization-family type, including the LocalBusiness and Store subtrees, that carries both a name and an absolute url. All five signals are then evaluated on that node and only that node.

That scoping is deliberate and it costs a little credit on purpose. A sameAs sitting on a Product, or on an Article about someone else, is not a statement about you, so it earns nothing. The alternative was worse: an earlier version collected sameAs from any node in the document, and on a Wikipedia article it printed a sentence about "your identity" connecting to a profile on a page where it had also just reported that no identity existed. Under-crediting is acceptable. Two sentences that contradict each other on one screen are not.

Sample Illustrative, not a measurement of any real site.

application/ld+json, crawler view

  • "@type": "Organization" identity
  • "name": "Acme Analytics", "url": "https://acme.com/" identity
  • "@id": "https://acme.com/#organization" stable id
  • "logo": "https://acme.com/logo.png" logo
  • "sameAs": ["https://www.wikidata.org/..."] identifiers
  • "description": "Analytics software" not a category
Sample identity node. Four of the five signals are read from this node; category also accepts a Product, Service or SoftwareApplication node declared elsewhere in the same markup.

Why these five, and why these weights

The weights are set by dependency first and measured frequency second. Identity is the gate at 35 points, because every other signal is a statement about an identity node and none of them can be true without one, and it stays under half so that the mere existence of a node is never most of the score. External identifiers take the largest non-gate weight at 25, because sameAs is the only property in the set whose own schema.org definition is an identity statement: a reference page that, in schema.org's words, "unambiguously indicates the item's identity".

The frequencies come from a hand-picked sample of 30 pages fetched on 26 July 2026 as raw HTML with no JavaScript executed: 24 returned JSON-LD and are the denominator, 6 returned none at all, and 9 more refused the fetch and were excluded. A stable @id appears on 11 of those 24 identity nodes, the closest to half of anything measured and so the best single discriminator in the sample. A logo appears on 19 of 24 and is the only property in the whole set for which any platform owner documents a consumer: Google's Organization documentation, updated 15 April 2026, lists it among 23 recommended properties and says the markup can influence which logo is shown in Search results and the knowledge panel. Category is the rarest signal kept, at 9 of 24 across every machine-readable form, so it takes the smallest weight. Thirty pages chosen by hand is enough to rank one property against another and nowhere near enough to support a sentence beginning "most sites".

  • Identity declared 35 pts An Organization-family type with a name and an absolute url, found by descending nested nodes as well as @graph.
  • External identifiers (sameAs) 25 pts One or more absolute sameAs URLs on that node, counted once each.
  • Stable identifier (@id) 15 pts An @id that is an absolute URI. A document-local fragment is reported, not credited.
  • Logo declared 15 pts An absolute http or https logo URL. A relative path is reported, not credited.
  • Category declared 10 pts A specific subtype, an additionalType URL, naics or isicV4, knowsAbout, or a SoftwareApplication, Product or Service node.
What each signal requires. A description satisfies none of them.

Three tiers instead of a letter grade

The tier is computed from the signals, never from the score. Not declared means there is no identity node in the HTML a crawler receives. Declared means one is present and complete, and nothing outside the page points at it. Anchored means it is present and carries at least one identifier that resolves outside the page.

It has to work that way, because 50 points is reachable in both the declared and the anchored configuration, so the number alone cannot name the state. The letter grade that used to sit here was worse than imprecise: it put two grades on different scales on one screen. Lantad's own stored capture of astro.build renders 88.8 out of 100, grade B, for the composite next to 0 out of 100 for entity confidence, because that page ships no JSON-LD whatsoever. Both numbers are correct and the pair reads as the report contradicting itself.

  • Not declared no identity node Nothing in the crawler-view JSON-LD declares who publishes the page.
  • Declared identity, unanchored An identity node is present and complete. Nothing points at it from outside this page.
  • Anchored identity plus an identifier Present, and carrying at least one identifier that resolves outside this page.
The three tiers. Each one ships with the sentence that defines it.

What it does not claim

Entity confidence measures markup presence. It does not measure, model, or predict what any AI system does with that markup, and no sentence on the report says a signal earns a mention, a citation or a place in an answer. That limit is not modesty, it is the state of the evidence: OpenAI, Anthropic and Perplexity publish nothing about structured data in their crawler documentation, and Google's own AI-features documentation, updated 10 December 2025, states that no special schema.org structured data is needed to appear in those features. The one thing Google does keep recommending is that structured data match the visible text on the page.

So the honest case for this diagnostic is the one the score actually supports: labelled facts are extractable and unambiguous, an identity with an absolute identifier can be pointed at, and a page that declares no identity at all leaves a machine nothing to attach anything to. Two named absences were removed from the score for the same reason they are named here. A founder was worth 20 points until we measured that 5 of 24 sampled pages declare one, that no platform owner documents consuming it, and that our own check counted an article author as a founder. A name was worth 10 points until we measured that across all 12 stored captures it never once appeared on a page where an identity node did not; it was folded into the gate.

Common questions

Does entity confidence affect my AI Visibility Score?
No. It is diagnostic and never feeds the composite, exactly like Retrievability. It is on every graded report, free, and it is scored from the same crawler-view fetch the composite already used, so it costs nothing extra to produce.
Why is there no letter grade any more?
Because the grade bands were calibrated for a continuous composite and were being applied to a five-item weighted checklist, which manufactures precision and puts two grades on different scales on one screen. Lantad's stored capture of astro.build shows the problem: composite 88.8 grade B beside entity confidence 0. A tier computed from the signals, printed with its definition, cannot be mistaken for the composite grade.
Why does a description no longer count as a category?
A description is prose, and prose is the thing structured data exists to replace. Counting any non-empty description made a property that 9 of 24 sampled pages declare in a machine-readable form look common, and it printed "your markup states what you are" over a marketing sentence. A subtype, an additionalType URL, an industry code, knowsAbout, or a Product-style node all still count.
My page has markup but scores zero. Why?
The most common reason is that the JSON-LD is injected by JavaScript, so it exists in the rendered DOM and not in the HTML a crawler receives, which is the only source this check reads. The report says so explicitly in that case. The other reasons are a node with a type but no name or no absolute url, or JSON-LD that declares only generic types such as WebPage and WebSite with no identity node anywhere in it.

See if AI can read your site

Paste a URL and see exactly what AI crawlers can and cannot read, with a graded report and ranked fixes. Free, no signup.

Scan your site free How the score works