Citation Rating: Why the Sources That Shape AI Opinion Matter More Than the Sources That Simply Rank

Answer Capsule

Citation Rating is a proposed framework for measuring how influential a source is within the evidence networks that shape AI-generated answers. Unlike raw citation counts, Citation Rating considers factors such as cross-model appearance, source independence, recommendation proximity, persistence over time and the importance of the sources connected to it. The goal is to identify which pages and publications disproportionately influence machine conclusions within a specific commercial category.

For years, digital marketers have asked:

Which websites have authority?

In AI search, I think the more useful question is becoming:

Which sources actually shape machine conclusions?

Those are not the same thing.

A website can have enormous traditional authority and still play a relatively small role in how AI systems answer a particular commercial question.

Another source can have modest conventional visibility yet appear again and again behind recommendations generated by ChatGPT, Gemini, Perplexity, Google AI Mode or other retrieval-driven systems.

That second source may be more strategically important than it first appears.

This is the problem I call Citation Rating.

Citation Rating is not simply a count of how often a source appears.

It is an attempt to measure where that source sits inside the information network that machines repeatedly use to form conclusions.

And if LLM optimization is going to mature beyond basic citation tracking, I believe this is one of the measurements we will need.

What Is Citation Rating?

Citation Rating measures the relative influence of a source within a defined AI citation and evidence network.

In practical terms, it attempts to answer:

Which sources appear to matter disproportionately when AI systems decide what to recommend, compare or believe?

A source with high Citation Rating may:

• appear across multiple AI platforms,
• support high-intent recommendation queries,
• recur over time,
• sit close to final recommendations,
• provide evidence later repeated by other sources,
• connect to other highly influential sources,
• and contribute information that is independently corroborated rather than merely duplicated.

That is different from asking:

“Which website got cited the most?”

Citation frequency is one signal.

Centrality is about structural importance.

Centrality Is Not a New Idea

The term centrality already has a long and rigorous history in network science.

I am not claiming to have invented centrality.

Researchers have spent decades developing ways to measure which nodes matter most inside networks.

Linton Freeman's classic 1978 paper formalized several important concepts, including degree, closeness and betweenness centrality.[1]

Phillip Bonacich developed measures in which the importance of a node depends partly on the importance of the nodes connected to it.[2]

Google's original PageRank system took a similar recursive idea into web search:

a link from an important page can be more valuable than a link from an unimportant page.[3]

Jon Kleinberg's HITS algorithm similarly distinguished between authorities and hubs, showing that the structure of relationships among pages could reveal different kinds of importance.[4]

The concept behind Citation Rating is therefore not:

“Let's invent a fancy new number.”

It is:

“Let's apply established network-centrality thinking to the provenance networks emerging around AI-generated answers.”

The new part is the object we are trying to measure.

Not merely links.

Not merely pages.

But relationships among:

queries → AI systems → sources → claims → brands → recommendations.

Why Raw Citation Counts Are Not Enough

Imagine we monitor 1,000 AI-generated answers in the home-security market.

Two sources emerge.

Source A

Cited 180 times.

All 180 citations came from one AI platform.

Most relate to general informational queries.

The source rarely appears when users ask which company they should choose.

Source B

Cited 76 times.

It appears across five AI platforms.

Its citations cluster heavily around comparison and recommendation queries.

Several of its claims appear directly before specific brands are recommended.

Its information is independently supported by government data and technical testing.

It continues appearing month after month.

Which source has more influence?

A simple citation count says:

Source A.

Citation Rating may say:

Source B.

That difference matters.

Because marketers do not actually want “citations.”

They want to understand which information environments affect decisions.

Frequency Is Visibility. Centrality Is Influence.

This distinction is critical.

A website might appear everywhere because it publishes thousands of pages.

That creates citation volume.

But another publication might appear only occasionally and consistently affect the final decision.

That creates citation influence.

In network terms, not all nodes perform the same role.

Some are common.

Some are bridges.

Some are authoritative reference points.

Some connect otherwise separate information communities.

Some introduce claims that propagate outward.

Some merely repeat what another source originally established.

If we count all citations equally, those differences disappear.

Citation Rating tries to preserve them.

Citation Rating Should Be Query-Specific

I do not believe a useful Citation Rating score should simply say:

“Forbes has a Citation Rating score of 92.”

That is probably too broad to be strategically meaningful.

Citation Rating should be measured within a defined information market.

For example:

Citation Rating for medical alert system recommendations.

Or:

Citation Rating for enterprise payroll software comparisons.

Or:

Citation Rating for “best credit card for international travel” queries.

A publication may be highly central in one evidence network and nearly irrelevant in another.

This is similar to how expertise itself works.

A cardiologist may be highly authoritative regarding heart disease.

That does not automatically make the same person authoritative regarding semiconductor manufacturing.

Machine influence should be treated with similar specificity.

The Unit of Analysis May Be the URL, Not the Domain

Traditional SEO often aggregates authority at the domain level.

That is useful.

But AI retrieval may force us to become much more granular.

A single URL can become disproportionately important.

Consider a highly cited research report hosted on a university website.

The university domain contains millions of pages.

Yet perhaps one particular PDF or research page repeatedly appears when machines answer a narrow question.

The domain itself is not the interesting object.

The specific source document is.

This suggests that Citation Rating may need to operate at multiple levels:

URL-level centrality

Which exact documents influence machine answers?

Domain-level centrality

Which publishers repeatedly provide influential documents?

Entity-level centrality

Which organizations repeatedly generate evidence that propagates across the network?

Claim-level centrality

Which specific statements become foundational to repeated machine conclusions?

That final level may prove especially interesting.

Claims Can Have Centrality Too

Imagine a company claims:

“Our response center answers calls in eight seconds on average.”

The manufacturer publishes the number.

Three review sites repeat it.

A comparison article repeats it.

Several AI systems later cite those sites while recommending the company.

It now appears that five independent sources support the claim.

But provenance analysis may show:

Manufacturer → Review Site A → Review Site B → Comparison Site → AI systems

The claim itself has become central.

But its evidence independence is low.

Now imagine another claim:

“The device achieved 99% location accuracy.”

That result comes from:

• an independent testing organization,
• a university study,
• a consumer publication,
• and a regulatory filing.

The second claim may have fewer total mentions but much stronger independent support.

This is why Citation Rating should eventually incorporate claim provenance, not simply publication frequency.

Source Independence Is One of the Hardest Problems

This is where AI citation analysis becomes significantly more complicated than counting footnotes.

Suppose five publications say the same thing.

Are there five sources?

Maybe.

Or maybe one press release has propagated through five websites.

The network might look like:

Company press release

↓
Publication A

↓
Publication B

↓
Aggregator C

↓
Review Site D

↓
AI answer

A naive system counts:

4 independent third-party sources.

A provenance-aware system may conclude:

1 primary claim source + 3 propagation nodes.

That distinction is enormously important.

Especially in commercial categories where marketing claims are frequently repeated across the web.

A strong Citation Rating framework should therefore distinguish:

citation diversity

from:

evidence diversity.

They are not interchangeable.

Cross-Model Breadth Matters

Another important dimension is cross-model appearance.

Suppose a page is cited by one AI system 300 times.

Another page is cited only 90 times but appears consistently across:

• ChatGPT,
• Gemini,
• Claude,
• Perplexity,
• Copilot,
• and Google AI Mode.

The second source may have stronger cross-model centrality.

Why?

Because repeated retrieval by substantially different systems suggests the source possesses characteristics useful across multiple retrieval environments.

Those characteristics might include:

• specificity,
• clarity,
• factual density,
• strong provenance,
• topical relevance,
• freshness,
• extractable structure,
• or established web authority.

We should not assume why the systems selected it.

We can measure the pattern first.

Then investigate.

This is an important principle across LLM Authority Index:

Measurement before explanation.

Recommendation Proximity Matters

Not every citation inside an AI answer performs the same role.

Consider two citations.

Citation A

Supports:

“The medical alert industry has expanded over the past decade.”

Citation B

Supports:

“For seniors living alone, Company X is the strongest overall choice.”

From a commercial standpoint, Citation B is much more important.

Both citations count as one citation.

But they do not have equal recommendation proximity.

A useful Citation Rating model should therefore consider where a source appears relative to:

• recommendations,
• rankings,
• buying decisions,
• comparative claims,
• negative claims,
• and category-defining statements.

A source cited behind a recommendation may exert more commercial influence than a source cited behind background context.

Citation Rating Is Different From Sentiment

This distinction also matters.

A publication can be highly central while being negative toward a company.

Centrality measures influence, not favorability.

Suppose Consumer Reports publishes a major investigation critical of a product.

AI systems repeatedly retrieve that investigation when users ask:

“Is Brand X reliable?”

The publication may possess extremely high Citation Rating.

The brand's sentiment within that source may be negative.

Those are separate measurements.

A complete system might therefore represent a source like this:

Citation Rating: High
Brand presence: High
Brand sentiment: Negative
Recommendation influence: High

That is far more useful than simply saying:

“You were mentioned.”

Persistence Over Time Matters

AI retrieval is not static.

Sources enter and leave the evidence environment.

Pages get updated.

Products change.

Models change.

Search indexes change.

News cycles change.

A publication that dominates this month may disappear six months later.

That creates another dimension:

Citation Persistence.

Suppose Source A appears heavily for two weeks after publishing a viral article and then disappears.

Source B appears consistently for eighteen months.

Those patterns should not receive identical treatment.

Persistent citation may indicate something closer to structural influence.

Temporary citation may reflect a news cycle.

Both matter.

But they matter differently.

Freshness Can Compete With Authority

This also raises an interesting tension.

Traditional authority often accumulates slowly.

But AI systems answering current product questions may need fresh information.

A five-year-old authoritative review can be less useful than a three-week-old specialist dataset if:

• the product changed,
• pricing changed,
• the company changed ownership,
• the service discontinued a feature,
• or new independent testing became available.

This suggests Citation Rating may include an inherently temporal element.

A source can gain or lose centrality as the information environment changes.

That is why repeated measurement is essential.

A Working Citation Rating Model

I do not believe the field is mature enough to declare one universal formula.

And I would be skeptical of anyone claiming there is already a perfect one-number measurement.

But conceptually, a Citation Rating framework could evaluate several dimensions.

1. Citation Frequency

How often does the source appear across the monitored query set?

2. Cross-Model Breadth

How many distinct AI systems retrieve or cite it?

3. Query Coverage

Across how many distinct query classes does the source appear?

4. Recommendation Proximity

How frequently is the source connected to buying recommendations, rankings or other consequential conclusions?

5. Source Independence

Does the source contribute original evidence, or is it primarily repeating another node?

6. Persistence

Does the source remain influential over repeated measurement periods?

7. Category Specificity

Is the source influential specifically within the commercial category being analyzed?

8. Network Position

Does the source connect to other influential sources, claims or entities within the citation graph?

9. Evidence Quality

Does the underlying information come from:

• independent testing,
• government data,
• original research,
• expert analysis,
• company claims,
• secondary summaries,
• or unattributed repetition?

10. Cross-Source Corroboration

Are important claims independently supported elsewhere?

None of these dimensions should automatically be treated as a known ranking factor inside any AI system.

They are measurement variables for our analysis of the external evidence network.

That distinction is important.

Citation Rating Is Not PageRank for ChatGPT

The analogy to PageRank is useful.

But it should not be taken literally.

I am not claiming:

“ChatGPT secretly calculates PageRank over citations.”

Nor am I claiming that one fixed citation graph exists across all AI systems.

Different platforms may use:

• different indexes,
• different retrievers,
• different ranking models,
• different freshness mechanisms,
• different search providers,
• different source filters,
• different generation systems.

Citation Rating is therefore an observer-side model.

We are building a map from repeated outputs.

We are not claiming access to the machine's private ranking function.

That makes the exercise similar to much of traditional search analysis.

SEO tools do not possess Google's complete ranking algorithm.

They measure outputs and infer patterns.

Citation intelligence will likely develop in the same way.

PageRank Shows Why Relationship Structure Matters

Google's original PageRank paper remains relevant because it introduced an important principle:

not every link should count equally.

A link from a page receiving many important links itself could carry more weight than a link from an isolated page.[3]

Citation networks deserve similar thinking.

Suppose:

Source A is cited by six AI systems.

Source B is cited by three AI systems.

But Source B is also:

• cited by Source A,
• relied upon by two industry reports,
• referenced by a government publication,
• and consistently appears as the original source behind an important market statistic.

Source B may occupy a more foundational position in the network.

Simple frequency would miss that.

Centrality analysis is designed to find it.

Hubs and Authorities May Reappear in AI Evidence Networks

Kleinberg's HITS model introduced another useful distinction.[4]

Some pages function as authorities.

Others function as hubs pointing toward useful authorities.

We may observe something similar in AI citation environments.

Evidence Authorities

Sources containing original data, testing or definitive documentation.

Evidence Hubs

Sources that synthesize and organize many important evidence authorities.

A Consensus Index, for example, may function primarily as an evidence hub.

It might not physically test every product.

Instead, it organizes:

• model outputs,
• independent testing,
• company claims,
• pricing,
• disagreement,
• and provenance.

That synthesis can itself become valuable.

This is precisely the model we are exploring with Aging in Place Index.

The publication does not need to pretend it generated every underlying fact.

Its value comes partly from making the relationship among those facts observable.

The Consensus Index Produces the Data Citation Rating Needs

In the previous article in this series, I introduced the Consensus Index.

Read: The Consensus Index — What Happens When We Stop Asking One Reviewer Who Is Best?

A Consensus Index repeatedly asks multiple AI systems the same structured questions.

That produces:

• recommendations,
• rankings,
• cited sources,
• disagreements,
• and longitudinal observations.

Once we accumulate enough of those citations, we can build a network.

That network is where Citation Rating becomes measurable.

In other words:

Consensus tells us what the machines conclude. Citation Rating helps us understand which sources appear to shape those conclusions.

The two measurements are complementary.

Search Authority and Citation Rating Are Different

In the previous article, I argued that digital authority is splitting into two related systems:

Search Authority

and:

Machine Authority.

Read: Authority Is Splitting in Two — Search Authority vs. Machine Authority

Citation Rating helps operationalize part of that distinction.

Traditional authority tools ask:

“How strong is this website?”

Citation Rating asks:

“How strategically positioned is this source inside a particular machine evidence network?”

Those answers may correlate.

But we should not assume they are identical.

The interesting opportunities for LLMO may exist precisely where they diverge.

The Citation Rating Gap

This suggests a potentially useful diagnostic concept:

the Citation Rating Gap.

Imagine a company performs a traditional authority audit.

Its marketing team has relationships with:

• large publishers,
• high-DR websites,
• major industry media,
• prominent blogs.

Everything appears healthy.

Then the team performs an AI evidence audit.

It discovers that most AI recommendations are being shaped by:

• three niche review sites,
• two Reddit communities,
• one technical publication,
• one industry association,
• and a comparison page nobody on the PR team had previously prioritized.

That difference is the Citation Rating Gap.

The company's communications strategy is focused on one information network.

The machines are relying heavily on another.

That is an actionable problem.

This Changes Digital PR

For twenty years, link acquisition encouraged marketers to ask:

“Can we get a link from this site?”

Citation Rating encourages a more strategic question:

“Does this source matter to the information ecosystem surrounding our buyer's decision?”

A website with extremely high traditional authority may be excellent.

But if it has almost no Citation Rating for the relevant category, it may not be the first target for an LLMO campaign.

Conversely, a niche source with modest traditional metrics may be critical if it repeatedly influences machine recommendations.

That means source targeting can become much more precise.

The objective moves from:

maximum placements

to:

maximum information influence.

This Does Not Mean Manipulating Central Sources

There is an obvious bad version of this idea.

A marketer identifies five highly central publications and attempts to buy, spam or manipulate mentions across all five.

That would recreate the worst parts of legacy link building.

Citation Rating should support the opposite behavior.

If a source is influential, the question should become:

“What evidence can we legitimately contribute that improves this source's understanding of the market?”

That could include:

• original research,
• technical documentation,
• corrected data,
• transparent pricing,
• product testing,
• expert interviews,
• verifiable customer evidence,
• case studies,
• regulatory information,
• or access to relevant company experts.

Centrality tells you where accuracy matters most.

It does not excuse manipulation.

Citation Rating Makes LLMO More Efficient

This has direct budget implications.

Suppose a company has $250,000 to invest in digital PR.

Without influence mapping, the campaign may pursue 300 publications.

But Citation Rating analysis identifies 18 sources that account for a disproportionate share of the evidence behind valuable AI recommendations.

That doesn't mean the other 282 sites are worthless.

It means the company now has priority intelligence.

It knows:

• which publications deserve immediate attention,
• where its information is missing,
• where competitor claims dominate,
• which inaccurate claims need correction,
• where independent evidence is weak,
• and where original research could have the greatest downstream value.

That is fundamentally different from buying another backlink report.

Citation Rating Can Also Find Risk

The same model can identify reputational vulnerabilities.

Suppose one highly central publication says:

“Brand X has unusually difficult cancellation terms.”

That claim is then repeated across:

• comparison websites,
• Reddit threads,
• AI answers,
• and buyer guides.

If the claim is true, the company has a product or policy problem.

If it is outdated, the company has an information-distribution problem.

If it is false, the company has a correction problem.

Without Citation Rating, the company may spend months responding to dozens of downstream mentions.

With provenance analysis, it may discover one highly central upstream source.

That is a much more efficient diagnosis.

The Goal Is to Find Upstream Influence

This may be the simplest way to explain Citation Rating.

Traditional monitoring often captures downstream mentions.

Citation Rating tries to locate upstream influence.

When 50 pages repeat the same claim, the valuable question is not:

“How do we get mentioned on all 50?”

It is:

“Where did the claim originate, which sources amplified it, and which nodes are machines repeatedly using?”

That is the influence graph.

And once the graph becomes visible, strategy changes.

From Citation Rating to Brand Rating

Citation Rating measures the source side of the network.

But brands occupy the same network.

That creates the next question:

Which brands are most strongly represented across the sources with the greatest Citation Rating?

That is what I call Brand Rating.

Imagine two companies.

Company A

Mentioned positively across 900 web pages.

Company B

Mentioned positively across only 350.

At first glance, Company A dominates.

But now weight those mentions by Citation Rating.

Suppose Company B appears positively in:

• nine of the ten most central sources,
• six independent testing reports,
• the dominant industry association,
• and multiple high-intent comparison sources.

Company A's 900 mentions are mostly scattered across low-influence sites.

Which brand occupies the stronger machine evidence position?

Possibly Company B.

Read next: Brand Rating — Measuring a Company's Position Inside the AI Evidence Graph

That is the brand-side implication of Citation Rating.

Citation Rating Also Connects to the Algorithmic Reciprocity Loop

The first article in this series introduced the Algorithmic Reciprocity Loop.

Read: The Algorithmic Reciprocity Loop — How AI Citations Can Become Real-World Authority

The core hypothesis is:

Machine recognition → human discovery → human citation → conventional web authority → increased machine discoverability.

Citation Rating may help identify which sources are most likely to initiate or accelerate that loop.

A highly central source that begins citing a new report may expose that report to:

• companies,
• journalists,
• researchers,
• agencies,
• creators,
• and other publishers.

Those humans create additional citations.

The influence spreads.

In network science terms, some nodes matter more because they are positioned to propagate information.

That may be equally true in AI-mediated information networks.

Measuring the Invisible Market

This is ultimately what interests me most.

Every commercial category contains an invisible information hierarchy.

Some sources create facts.

Some verify them.

Some interpret them.

Some amplify them.

Some repeat them.

Some organize them.

Some machines retrieve them.

Some companies dominate within them.

Some brands are nearly absent.

Traditional SEO has given us excellent tools for mapping links and rankings.

Social listening has given us tools for mapping conversations.

Citation Rating attempts to map something else:

the architecture of evidence behind machine-generated conclusions.

That is a new analytical layer.

Why Citation Rating Could Matter More Than Another Backlink

Suppose a marketing agency secures 100 new backlinks.

Traditional reporting looks excellent.

But none of those sites appears in the evidence ecosystem surrounding the client's high-value AI queries.

Another agency earns five placements.

All five occur within highly central sources regularly used across AI recommendation environments.

Which campaign generated more machine influence?

We cannot answer that with backlink counts alone.

That is exactly why Citation Rating matters.

The metric is not designed to declare backlinks irrelevant.

It is designed to measure something backlinks were never intended to measure:

proximity to machine belief formation.

The Future Question for CMOs

I believe one of the most important questions a CMO will eventually ask is:

“Which twenty sources most influence what AI systems believe about our category?”

Not:

“Which websites have the highest DR?”

Not:

“Where does our competitor have backlinks?”

Not even:

“Where are we mentioned most frequently?”

Those will remain useful inputs.

But the strategic question is about influence concentration.

Which sources matter disproportionately?

Which claims do they reinforce?

Which brands dominate them?

Where does our evidence originate?

How persistent are those relationships?

That is Citation Rating.

What We Are Building Toward

At LLM Authority Index, I see the analytical stack developing like this:

Consensus Index

Measures what multiple AI systems conclude.

Search Authority vs. Machine Authority

Separates conventional search prominence from machine influence.

Citation Rating

Identifies which sources disproportionately influence machine conclusions.

Brand Rating

Measures which companies occupy the strongest positions within those influential sources.

Algorithmic Reciprocity

Measures how machine recognition can produce human citation and new conventional authority.

Machine Relations

Turns those measurements into an operating strategy for communications, digital PR and LLM optimization.

The purpose is not to replace every existing marketing metric.

It is to answer a set of questions those metrics were never built to answer.

A Measurement Framework, Not a Secret AI Ranking Factor

I want to end with an important disclosure.

Citation Rating is a proposed analytical framework.

It is not:

• an OpenAI ranking factor,
• a Google ranking factor,
• a Perplexity metric,
• an Anthropic metric,
• or an established industry standard.

It is an attempt to apply ideas from network science and citation analysis to a new observable system: AI-generated answers with identifiable sources.

The hypothesis is that some sources occupy disproportionately influential positions inside those networks.

That hypothesis is testable.

We can collect:

• repeated AI outputs,
• citation data,
• source relationships,
• query classes,
• recommendation outcomes,
• and historical observations.

Then we can test whether sources with high measured centrality are disproportionately associated with persistent AI conclusions.

That is what research should do.

Measure first.

Publish the methodology.

Show the limitations.

Then see whether the model predicts anything useful.

The Next Question: Where Does the Brand Sit?

Once we know which sources are central, the next question becomes much more commercially important:

Which companies dominate those sources?

A brand can have enormous visibility and weak central positioning.

Another can have fewer total mentions but occupy exactly the sources machines appear to rely upon.

That difference is the subject of the next article:

Brand Rating: Measuring a Company's Position Inside the AI Evidence Graph

Because in AI search, visibility may not ultimately be about how often the internet mentions you.

It may be about where you sit inside the evidence machines use to make decisions.

Sources and Research

1. Freeman, Linton C. — “Centrality in Social Networks: Conceptual Clarification.” Social Networks, 1978/1979.
https://doi.org/10.1016/0378-8733(78)90021-7

A foundational work formalizing major concepts in network centrality, including degree, betweenness and closeness. Citation Rating borrows from this broader intellectual tradition but applies centrality thinking to AI citation and evidence networks.

2. Bonacich, Phillip — “Power and Centrality: A Family of Measures.” American Journal of Sociology, 1987.
https://doi.org/10.1086/228631

Bonacich's work develops recursive measures in which a node's importance depends partly on the importance and structure of the nodes around it. This principle is useful when considering why raw citation frequency alone may fail to capture source influence.

3. Page, Lawrence; Brin, Sergey; Motwani, Rajeev; Winograd, Terry — “The PageRank Citation Ranking: Bringing Order to the Web.” Stanford InfoLab, 1999.
https://ilpubs.stanford.edu:8090/422/

The foundational PageRank paper demonstrates one of the most commercially significant applications of recursive network importance: links should not automatically be weighted equally.

4. Kleinberg, Jon M. — “Authoritative Sources in a Hyperlinked Environment.” Journal of the ACM, 1999.
https://doi.org/10.1145/324133.324140

Kleinberg's HITS framework distinguishes between authoritative pages and hubs that point toward authorities. The distinction provides useful conceptual precedent for examining different roles within AI evidence networks.

5. Lewis, Patrick et al. — “Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks.” NeurIPS, 2020.
https://arxiv.org/abs/2005.11401

Foundational research on combining language generation with external information retrieval. It provides technical context for why source-selection and evidence networks matter in systems that combine retrieval with generation.

6. Gao, Tianyu et al. — “Enabling Large Language Models to Generate Text with Citations.” 2023.
https://arxiv.org/abs/2305.14627

Research examining how language models can generate answers supported by external citations. It contributes to the technical foundation for treating citation behavior as an observable component of generative systems.

7. Liu, Nelson F. et al. — “Evaluating Verifiability in Generative Search Engines.” 2023.
https://arxiv.org/abs/2304.09848

Research examining whether citations in generative search adequately support the claims attached to them. The study reinforces an important principle behind Citation Rating: citation presence and evidence quality are not the same thing.

8. Aggarwal, Pranjal et al. — “GEO: Generative Engine Optimization.” arXiv:2311.09735; presented at KDD 2024.
https://arxiv.org/abs/2311.09735

An early academic study treating source visibility within generative-engine responses as a measurable outcome. Its relevance here is not that it validates Citation Rating specifically, but that it supports studying generative-source visibility as an independent measurement problem.

Methodology and Disclosure

Citation Rating is a working framework proposed by Mark Huntley and LLM Authority Index for analyzing source influence within AI-generated citation and evidence networks.

The term draws explicitly from established network-science concepts and should not be interpreted as a claim that centrality itself is new.

Citation Rating is not a publicly documented metric used by Google, OpenAI, Anthropic, Perplexity, Microsoft or another AI provider.

The current research hypothesis is that source influence within AI recommendation environments can be better understood by combining multiple observable variables rather than relying on citation frequency alone.

Candidate variables include:

• citation frequency,
• cross-model breadth,
• query coverage,
• recommendation proximity,
• evidence independence,
• citation persistence,
• category specificity,
• network position,
• corroboration,
• and source quality.

The relative weighting of these variables remains an empirical question.

LLM Authority Index intends to evaluate Citation Rating longitudinally and compare the resulting measurements against actual AI recommendation behavior.