Defined term

Entity Resolution Rate

Entity Resolution Rate is the share of observed answer-engine responses that mention a target entity and match a declared set of identity attributes, under a disclosed query set, engine panel, collection window, and matching rule.

Defined at the home of Machine Relations: Entity Resolution Rate in the Machine Relations glossary

Entity resolution rate is resolved in-scope brand mentions divided by all in-scope brand mentions. A mention is resolved when it matches the identity attributes and matching rule declared before scoring. The rate describes a disclosed answer sample; it is not a provider confidence score, name-recall rate or citation rate.

The Machine Relations definition and formula are the formula. Machine Relations Research and AuthorityTech are both Bramwell companies. This entry explains the term. The measurement guide supplies a scoring worksheet; the entity-recognition solutions guide maps observed failures to possible fixes.

What belongs in the denominator

Declare the target brand, accepted aliases, identity facts, query panel, engine or product, collection dates and inclusion rule before scoring. An in-scope mention remains in the denominator when it names the wrong company, gives the wrong founder or category, or leaves a required attribute uncertain. It does not enter the numerator unless it meets the matching rule.

A successfully collected answer that does not name the target brand is a non-mention, not an unresolved mention. Keep it in the separate mention-rate calculation. A failed request, empty result or unavailable engine is missing collection evidence, not a non-mention and not a failed identity match. Report those missing cells separately.

Do not calculate entity resolution rate when there are no in-scope mentions. The result is unmeasured, not a zero score. A brand can be accurately described whenever it appears and still be absent from most answers. Reporting entity resolution rate beside mention rate prevents that conditional accuracy from being mistaken for broad visibility.

What counts as a match

Write a dated reference record with sources for the facts the assessment checks: for example the canonical name and domain, company category, founder, product or ownership relationship. Choose only the fields relevant to the question, and apply the same rule to every eligible answer. A common-word name or rebrand needs an explicit alias and ambiguity rule.

Record each required field as match, mismatch, absent or uncertain. A required founder fact that is missing is not a match. An accurate identity accompanied by unfavorable opinion is not automatically an identity error: sentiment, recommendation and factual matching need separate labels. If an answer gives contradictory identity facts, preserve the contradiction instead of selecting only the favorable sentence.

The measurement guide uses a disclosed response-level convention: at most one target-brand assessment per answer, so repeated names do not multiply the sample. An occurrence-counted mention protocol is a different unit and must not be silently pooled with it.

Entity resolution rate is not mention or citation rate

ObservationIts denominatorWhat it tells you
Collection coverageAttempted collection cellsWhether a usable answer was obtained
Mention rateSuccessfully collected, in-scope answersWhether the target brand was named
Entity resolution rateIn-scope target-brand assessmentsWhether named instances met the declared identity rule
Citation or recommendationThe separately declared answer populationWhether a source was cited or the brand was offered as an option

The same answer can name a brand without citing it, cite it with an incorrect category, or describe it accurately without recommending it. Neither share of citation nor share of voice is replaced by an accuracy calculation; they describe different observations.

What a low rate does—and does not—establish

A lower rate establishes that fewer eligible mentions matched the declared identity rule in that sample. Preserve the actual mismatched facts and inspect the pages the answer cites. Stale first-party records, conflicting third-party descriptions, ambiguous names and collection changes are hypotheses to investigate, not causes proved by the rate.

For example, Liu and Nag’s query-brand linking research addresses brand disambiguation in e-commerce search. It does not measure brand representation in consumer answer engines. The GEO-16 study concerns observed page quality and citation, not a brand-level resolution cutoff. Neither establishes that buying coverage, adding schema or reaching a fixed score makes an engine cite a brand.

There is no universal engine confidence threshold to infer from this output metric. Test source corrections against the same panel, product, matching rule and collection conditions; report any change as an observation rather than attributing it to one intervention without separate evidence.

Frequently asked questions

How is entity resolution rate calculated? Divide resolved in-scope brand mentions by all in-scope brand mentions under a declared matching rule. Disclose the counting unit, required identity fields, query panel, engine, dates and exclusions. Non-mentions and failed collection are reported separately, not added to this denominator.

What if the brand is never named? Entity resolution rate is unmeasured because there are no in-scope mentions to score. Report the non-mention result over successfully collected answers and disclose any missing collection cells. Do not label absence as an identity mistake or invent a zero accuracy rate.

Can a brand have a high rate but low AI visibility? Yes. The rate can be high among the few answers that name the brand while its mention rate is low. Citation and recommendation outcomes can also differ. Report each observation with its own denominator.

What should a team fix first? Start from the preserved failure: incorrect identity, stale attributes, absence, missing citations or no recommendation. Those are different problems. The entity-recognition solutions guide explains the checks and candidate actions; none is a guaranteed citation remedy.

See how your brand performs in AI search

Free AI visibility audit: your AI visibility score, the publications already covering you, the gaps, and the story angles to close them.

Run Free Audit