Entity Resolution Rate: Why AI Recommends Some Brands and Ignores Others
Entity resolution rate measures how confidently AI systems can identify, verify, and cite a brand across sources. Brands that resolve cleanly get recommended. Those that don't get ignored or misrepresented.
Entity resolution rate measures how confidently an AI system can identify a brand as a single, coherent entity across multiple sources. When someone asks ChatGPT which payments platform to trust for enterprise clients, the AI doesn't just find text that mentions brands. It runs a resolution process: cross-referencing names, descriptions, founding dates, product categories, and publication mentions to determine whether the different signals it finds all refer to the same entity. Brands that resolve cleanly get cited. Brands that produce conflicting, thin, or ambiguous signals get ignored or misidentified.
This is not a technical edge case. Harvard Business Review's March 2026 analysis of brand readiness for agentic AI documented how LLM data on brands was "often incomplete or incorrect" in ways companies only discovered after AI systems had already started influencing buyer decisions. One major spirits brand found that a popular AI model had miscategorized an affordable product as a prestige offering. The brand had not failed at marketing. It had failed at machine legibility.
Entity resolution rate, introduced by Jaxon Parrott as part of the Machine Relations measurement framework, captures this directly. It measures the percentage of AI queries in which a brand is correctly identified, attributed, and represented as intended. It replaces share of voice as the primary measurement for the AI search era, because volume of mentions means nothing if the AI cannot confidently resolve which entity is being referenced.
Key takeaways
- Entity resolution rate measures how often AI systems correctly identify and represent a brand when the brand is relevant to a query
- Resolution confidence requires corroboration: multiple independent, trusted sources all describing the brand consistently
- Earned media from Tier 1 publications is the primary signal that moves entity resolution rate because AI engines weight third-party editorial sources over brand-owned content
- A brand can have extensive online presence and still resolve poorly if that presence is concentrated in owned channels rather than independent editorial sources
- Entity resolution rate is part of the Machine Relations measurement layer; measuring it is how brands move from subjective "AI visibility" to a trackable, improvable metric
- Brands with high resolution rates get recommended; those below the confidence threshold get passed over regardless of what they've invested in owned content
What entity resolution actually means for brands
AI language models build their understanding of the world from training data and real-time retrieval. When they encounter a brand name in a query, they search across everything they know about that name: Wikipedia entries, Wikidata records, news coverage, Crunchbase profiles, LinkedIn descriptions, press releases, industry reports, analyst mentions, and editorial coverage in publications they've indexed and trust.
If all of those sources agree on what the brand does, who founded it, what category it belongs to, and what it's known for, resolution confidence is high. The model can cite the brand accurately and specifically. If those sources conflict, are sparse, or describe the brand in ways that don't connect into a coherent picture, resolution confidence drops. Below a confidence threshold, the model either omits the brand or provides a hedged, generic mention that doesn't help the brand or the person asking.
The academic framing of this problem is well documented. Research from Dong Liu and Sreyashi Nag at arXiv (February 2025) on query brand entity linking in e-commerce search documented this process directly: entity linking requires both detecting a brand mention and disambiguating which entity is meant when multiple matches exist. The resolution fails most often when the entity's signals are inconsistent or when the gap between a brand's real identity and its documented identity is wide.
For practical brand strategy, this translates to one concrete question: does your documented identity online match your actual identity, and is that identity documented in sources AI systems treat as authoritative?
The resolution confidence threshold
AI engines don't cite brands they're uncertain about. The design is intentional: hallucinating a brand recommendation is worse than omitting one. AI systems operate with resolution confidence thresholds below which they simply won't surface the brand by name in a response, even when the brand is genuinely relevant to the query.
As documented in the AuthorityTech entity resolution glossary entry, this confidence floor sits at roughly 60%: brands that fail to meet the threshold get omitted or appear as vague, unspecific mentions. The threshold exists because AI engines are built to surface specific, confident answers. A brand that cannot be clearly resolved produces a worse user experience than no mention at all.
Research on how AI systems make citation decisions shows this selectivity clearly. The GEO-16 framework analysis by Kumar et al. (September 2025) studied 1,702 citations across Brave, Google AIO, and Perplexity using 70 industry-targeted prompts. Pages that scored above a quality threshold of G greater than or equal to 0.70 and hit at least 12 structural quality pillars achieved a 78% cross-engine citation rate. Pages below the threshold were omitted even when their content was directly relevant to the query. The odds ratio for quality as a predictor of citation was 4.2 across the full dataset.
For brand entities, the equivalent logic applies. Strong, consistent signals in trusted sources produce confident resolution. Weak, scattered, or contradictory signals produce omission.
| Resolution confidence level | What triggers it | AI system behavior |
|---|---|---|
| High (80%+) | Multiple high-DA sources with consistent, specific entity signals; Wikidata entry; editorial coverage in Tier 1 publications | Brand cited by name with accurate specifics; high probability of appearing in relevant recommendations |
| Medium (60-80%) | Some editorial coverage; basic structured data; brand-owned descriptions that partially match external sources | Brand cited in some contexts; may be misattributed in edge cases; inconsistent performance across engines |
| Low (<60%) | Sparse third-party coverage; conflicting descriptions; primarily brand-owned signals; no Wikidata or knowledge graph anchor | Brand omitted from AI responses even in relevant queries; may appear as a generic, hedged mention without specifics |
Why earned media determines resolution rate
Brands that improve entity resolution rate almost always do it through the same mechanism: earning placements in publications that AI systems already treat as authoritative sources.
This is not a formatting or technical optimization problem. AuthorityTech's research on earned vs. owned AI citation rates found that earned media produces 325% more AI citations than owned content distribution. The reason is that AI systems are built to be skeptical of brand-owned content. A company's own website, press releases, and social media are signals, but they carry lower weight precisely because they're unverified by any independent editorial judgment.
Earned placements in publications like Forbes, TechCrunch, or the Wall Street Journal function differently in an AI engine's signal hierarchy. These are sources the AI has been trained to trust as editorially independent. When they describe a brand's category, founding, and capabilities, those descriptions carry enough authority to move resolution confidence in a meaningful way.
The Muck Rack analysis of over one million AI prompts (July 2025) found that 85.5% of AI citations come from earned media sources and over 95% come from non-paid sources. This is not incidental. It reflects the underlying architecture of how AI systems decide what to cite: they lean on the same sources that human researchers and journalists have trusted for decades.
Ahrefs' analysis of ChatGPT citation behavior across 75,000 brands confirmed this pattern quantitatively: brand web mentions correlated with AI Overview visibility at 0.664, compared to 0.218 for backlinks. Brand mentions in trusted third-party editorial sources, not technical SEO signals, determine how AI systems resolve and recommend brands.
Moz's 2026 analysis of 40,000 AI Mode queries found that 88% of AI Mode citations were not in the organic top 10 search results. The population of sources AI systems trust for citation is substantially different from the population of pages that rank well in traditional search. This means that optimizing for search ranking and optimizing for entity resolution are distinct activities requiring distinct strategies.
The foundational GEO research from Princeton and Georgia Tech (Aggarwal et al., SIGKDD 2024) established that adding statistics to content improves AI citation rates by 30-40%, and that citing credible sources increases the probability of the page itself being cited. The mechanism works in both directions: trusted sources citing a brand raise its entity resolution confidence; the brand citing trusted sources improves the extractability of its own content.
An independent study from Fullintel and the University of Connecticut (IPRRC, February 2026) analyzed AI citation patterns across major AI engines and found that 47% of all AI citations in responses came from journalistic sources, 89% of links cited were earned media, and 95% were unpaid. These numbers are consistent across multiple independent research streams and point to a single conclusion: AI systems have a structural preference for editorial, third-party sources that cannot be overridden through owned content investment alone.
The entity graph AI systems build
Understanding what AI systems actually use to resolve brand entities clarifies what's worth building. The signals that matter most, in rough order of weight:
- Wikidata and knowledge graph entries. These are structured, machine-readable facts about entities: founding date, founder name, product category, headquarters, key relationships. When a Wikidata record exists and is accurate, it gives AI systems an anchor point that makes all other signals more interpretable. Without it, AI systems have to reconcile competing signals without a reference point.
- Editorial coverage in high-DA publications. Forbes, TechCrunch, Reuters, Bloomberg, the Financial Times. When these sources consistently describe a brand in the same terms, resolution confidence rises. The more independently corroborating sources use the same entity signals, the stronger the anchor.
- Schema markup on owned properties. Organization schema, Person schema, and FAQPage schema give AI crawlers structured data they can use directly. This doesn't substitute for earned media, but it makes the brand's own signals machine-readable in a way that raw prose doesn't.
- Consistency across brand-owned properties. When a company's website, LinkedIn profile, Crunchbase entry, and press materials all describe the brand in the same terms, it reduces the risk of ambiguity during resolution. Inconsistency across owned channels is one of the most common sources of poor resolution rate in companies that otherwise have strong reputations.
- Cross-domain corroboration. When multiple independent domains, not just multiple articles on the same platform, all cite the same entity facts, resolution confidence gets its biggest single boost. AI systems look for corroboration across different domains rather than accumulating evidence from one source.
This architecture is what the HBR analysis pointed to when it described brands finding that AI representations were "often incomplete or incorrect." The problem wasn't that the brands were obscure. It was that the entity graph AI systems had built for those brands was thin, contradictory, or misaligned with how the brands actually wanted to be described.
What low entity resolution rate costs
The cost of low resolution rate has become direct and measurable. B2B buyers now conduct most of their vendor research independently before contacting a sales team, and an increasing share of that research runs through AI answer engines rather than traditional search. When buyers run those AI queries, brands with low entity resolution rate are simply absent from consideration.
This is different from the traditional problem of low search ranking. A brand that ranks position 8 for a keyword still appears in search results. A brand that falls below the AI resolution threshold doesn't appear at all. The buyer never sees it as a candidate.
The brands that were building entity signals through earned media during this period are the ones that arrive at that transition with high resolution rates already in place.
Oxford and Stupid Human's ChoiceEval research (2026) on auditing brand preferences in LLMs found consistent geographic and recency biases in which brands AI systems recommend. Applied to Gemini, GPT-5, and DeepSeek across 10 topics and more than 2,000 questions, the study found that U.S.-developed models showed marked favoritism toward American entities, while brand preferences persisted consistently across different user personas. Brand recommendation in AI responses is not random: it follows the entity signal graph.
How to measure entity resolution rate
The measurement methodology follows directly from the definition. To measure resolution rate for a brand:
- Define the query set. Identify 20-30 queries that a prospect or buyer would ask when researching the brand's category. These should include the brand's primary value proposition, use cases, and comparison queries ("best [category] for [use case]," "[brand] vs [competitor]," "who leads [category]").
- Run the queries across AI engines. ChatGPT, Perplexity, Gemini, Claude, and Google AI Mode will each produce different responses to the same query. Run each query across at least three engines.
- Score each response on three dimensions:
- Presence: was the brand mentioned at all?
- Accuracy: was the entity described correctly (category, founding, capabilities)?
- Sentiment accuracy: did the description match how the brand wants to be positioned?
- Calculate rate and identify patterns. Resolution rate is the percentage of relevant queries where the brand was present and accurately described. Patterns in where accuracy fails point to specific entity graph gaps.
The AuthorityTech visibility audit runs this process automatically, identifying current resolution rate, the specific entity signal gaps driving it, and the earned media opportunities most likely to move the metric.
Improving entity resolution rate
Resolution rate improves when the entity graph becomes denser, more consistent, and more independently corroborated. The sequence matters:
Start with entity clarity on owned properties. Before building external signals, make sure the brand's own web presence, schema markup, and cross-platform descriptions are consistent and accurate. The Wikidata entry, if it exists, should match the website. The LinkedIn description should use the same category terms as the Crunchbase profile. Inconsistency in owned signals undermines the value of external corroboration because AI systems can't resolve the contradiction.
Earn placements in publications AI engines trust. This is the primary lever for resolution rate improvement. A placement in a publication with DA 70+ that consistently describes the brand's category, founding, and value proposition is worth more to entity resolution than dozens of lower-quality mentions. The publication doesn't just create a citation. It creates a trusted, independently verified data point that AI systems can use to anchor their resolution of the brand entity.
This is the mechanism at the foundation of Machine Relations: earned media from trusted publications is the primary infrastructure through which AI-mediated brand discovery is built. The publications that shaped human brand perception for decades are the same publications AI systems treat as authoritative. What changed is the reader.
Build corroboration across domains, not just publications. Multiple publications on the same platform corroborate each other, but cross-domain corroboration produces stronger resolution signals. An entity described consistently in Forbes, a university case study, a Wikidata entry, and a regulatory filing has higher resolution confidence than the same entity described in twenty articles on the same news platform.
Anchor the entity graph to knowledge base records. Wikidata entries, Wikipedia citations when available, and schema markup on owned properties give AI systems structured records to anchor all other signals to. These are the hardest entity signals to fake and therefore the ones AI systems weight most heavily for disambiguation.
Entity resolution rate vs. share of voice
| Metric | What it measures | Era it belongs to | What drives it |
|---|---|---|---|
| Share of voice | How often a brand is mentioned relative to competitors in a defined media set | Traditional PR and human-read media | Volume of coverage, journalist relationships, PR spend |
| Entity resolution rate | How often AI systems correctly identify and accurately represent a brand when relevant | AI search and answer engine era | Consistency and corroboration of entity signals in sources AI systems trust |
Share of voice is not obsolete. It continues to matter for human readers and human brand perception. But it measures a different thing than resolution rate, and for brands trying to understand their AI search visibility, it's an incomplete proxy. A brand can have high share of voice and low entity resolution rate if its coverage is concentrated in lower-DA sources, is internally inconsistent, or describes the brand in terms that don't align with how AI systems have categorized the relevant market.
The brands with the highest AI citation rates are often not the ones with the highest traditional media coverage. They're the ones with the most consistent, independently corroborated entity signals across the sources AI systems weight most heavily. Coverage volume and citation accuracy are related but not the same thing.
Frequently asked questions about entity resolution rate
What is entity resolution rate?
Entity resolution rate is the percentage of AI search queries, relevant to a brand, in which the AI system correctly identifies and accurately represents that brand. It measures machine legibility: not just whether a brand is mentioned, but whether the AI can confidently resolve the brand as a specific entity with accurate attributes. A high entity resolution rate means the brand gets cited correctly in relevant AI-generated answers. A low resolution rate means the brand gets omitted or misrepresented even when it's directly relevant to the query.
Who introduced entity resolution rate as a brand metric?
Jaxon Parrott, founder of AuthorityTech, defined entity resolution rate as part of the Machine Relations measurement framework. The term adapts entity resolution from its database management origins to the specific challenge of brand legibility in AI search. In database management, entity resolution is the problem of determining whether two records refer to the same real-world entity. In Machine Relations, it's the problem of whether AI systems can determine with confidence which brand they're being asked about and whether they can describe it accurately.
How does entity resolution rate relate to share of citation?
Share of citation measures how often a brand is cited across all AI-generated answers in a given category. Entity resolution rate measures the quality of those citations, specifically whether the citations are accurate and correctly attributed. Both are part of the Machine Relations measurement layer. Share of citation tells you volume; entity resolution rate tells you whether the volume is actually working for the brand or creating noise.
How do AI systems decide whether to cite a brand?
AI systems use a resolution process that cross-references multiple signals: knowledge graph records, editorial coverage, structured data, and cross-domain corroboration. The GEO-16 framework research (Kumar et al., 2025) found that pages hitting a quality score of G greater than or equal to 0.70 and at least 12 structural quality pillars achieved a 78% cross-engine citation rate. Below the threshold, pages were omitted from citations regardless of their topical relevance. For brand entities specifically, the threshold is primarily determined by the consistency and authority of the external signals that describe the brand.
Can a brand have high Google rankings and low entity resolution rate?
Yes, and this is one of the most common gaps for brands entering the AI search era. Google's PageRank algorithm and AI entity resolution weigh different signals. PageRank rewards backlinks, technical on-page optimization, and click-through behavior. AI entity resolution rewards consistent, independently corroborated entity signals in editorially trusted sources. Machine resolution and search ranking are related but distinct processes. A brand can rank well organically and still fail the entity resolution confidence threshold in AI systems, particularly if its SEO strategy focused on technical optimization rather than building entity signals through earned editorial coverage.
What's the fastest way to improve entity resolution rate?
Earn placements in publications AI systems index and trust. Specifically: publications with DA 70+ that editorially describe the brand's category, capabilities, and positioning. The Muck Rack analysis of one million AI prompts found that 85.5% of all AI-cited sources were earned media and over 95% were non-paid. That distribution reflects how AI systems are built: they weight third-party editorial judgment over brand-owned signals because editorial judgment is independently verified. A single placement in a trusted Tier 1 publication that accurately describes a brand's category and founding can move resolution confidence more than months of owned content production.
The measurement shift that's already underway
The brands that will define their categories in AI-mediated discovery are building entity resolution rate now, while most of their competitors are still optimizing for traditional search metrics.
Bain's 2025 consumer research found that about 80% of search users rely on AI summaries at least 40% of the time on traditional search engines, and about 60% of searches now end without the user progressing to a website. What appears in AI summaries doesn't come from whoever spent the most on keywords. It comes from whoever built the strongest, most consistent, most corroborated entity signal graph in the sources AI systems trust.
The convergence between PR and AI-mediated discovery has been documented by independent observers outside the AT ecosystem. Stacker's February 2026 analysis of the AI discovery shift quoted Gab Ferree, founder of Off the Record, directly: "Media relations are becoming machine relations. It's on the comms professionals to learn the patterns of AI and then take action on them." The article, syndicated to 200+ publications, concluded that the same editorial relationships that drive traditional media coverage are now the infrastructure for AI citation.
The mechanism has not changed from how PR built brand credibility in the era of human readers. A placement in a respected publication, earned through a genuine editorial relationship, is the most powerful trust signal available. It was true when buyers were human. It remains true now that AI systems are doing the first layer of research on buyers' behalf. What changed is the reader.
This is what Machine Relations names as a discipline: the work of making a brand legible, retrievable, and citable to machine readers. Entity resolution rate is how you measure whether that work is landing. Brands with high resolution rates get recommended. Brands that fall below the confidence threshold get passed over, regardless of how much they've invested in channels that optimize for a different era's signals.
Start your visibility audit to see your brand's current entity resolution rate and the specific gaps driving it. →