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

Answer Capsule

Search authority and machine authority are related but increasingly distinct forms of digital influence. Search authority describes a page or domain’s ability to earn visibility in traditional search results. Machine authority describes the likelihood that AI systems retrieve, trust, cite or rely on a source when constructing an answer. A website can therefore have strong traditional search visibility but limited influence over AI-generated recommendations—or meaningful AI influence without dominating the conventional search results.

For more than twenty years, digital marketers have had a reasonably clear mental model of authority.

A website accumulates links.

It develops topical depth.

It earns mentions.

People search for the brand.

Its pages acquire history.

Search engines develop confidence in the domain.

The result is what the SEO industry generally describes as authority.

The exact terminology varies.

PageRank.

Domain Rating.

Domain Authority.

Trust.

Link equity.

Topical authority.

Brand authority.

None of the commercial SEO metrics are themselves Google ranking factors, but they are all attempts to model roughly the same phenomenon:

How difficult is this website to displace in traditional search?

Generative AI introduces a second question.

How difficult is this source to displace from what machines believe?

Those are not necessarily the same thing.

I believe digital authority is beginning to split into two overlapping systems:

Search Authority

and

Machine Authority.

Understanding the difference may become one of the most important problems in LLM optimization.

What Is Search Authority?

For the purposes of this framework, Search Authority describes a website's ability to achieve persistent visibility within conventional search-engine results.

Links remain part of that history.

Google's foundational PageRank research treated hyperlinks as a form of citation and used the structure of the web graph to help estimate importance.[1]

Google's ranking systems have obviously evolved enormously since then.

Today, Google describes Search as relying on multiple ranking systems rather than one single authority score.[2]

Those systems consider many different kinds of information.

But the traditional outcome remains familiar:

A user enters a query.

Google ranks a set of documents.

Some websites repeatedly appear near the top.

SEO developed around understanding and influencing that ranking environment.

That gave us an industry obsessed with questions such as:

• How many websites link to this domain?
• How strong are those domains?
• Which keywords does the site rank for?
• How much organic traffic does it receive?
• How comprehensive is its topical coverage?
• How established is the brand?
• How frequently do users search for it?

Those remain useful questions.

But they don't completely answer the question AI search creates.

What Is Machine Authority?

I use Machine Authority to describe:

The degree to which a source influences the information an AI system retrieves, trusts, cites or uses when forming a conclusion.

That definition is intentionally different from:

“Does this website rank #1?”

A machine can encounter a document without displaying it as the first blue link.

A retrieval system can select passages rather than entire pages.

An AI system can compare information across many sources and produce a synthesized answer.

A source might influence that synthesis without becoming the final visible citation.

Another source might be cited frequently because its information is particularly easy to extract and verify.

Machine Authority is therefore not just about visibility.

It is about influence over synthesis.

And that creates an entirely different analytical problem.

The Difference Between Ranking and Retrieval

Traditional search gives us a ranked list.

Generative systems increasingly give us a synthesized conclusion.

That architectural difference matters.

Consider the query:

“What is the best medical alert system for an active senior who lives alone?”

Traditional search might return:

1. Forbes
2. U.S. News
3. National Council on Aging
4. Consumer Reports
5. The Senior List

The user then decides which result to open.

A generative system may instead produce:

“Medical Guardian is generally the strongest choice for an active senior living alone because of its mobile GPS options, fall detection and monitoring capabilities.”

The machine has already performed part of the comparison.

The user's question therefore becomes:

What information caused the system to reach that conclusion?

That evidence may have come from:

• one highly ranked review,
• several lower-ranked pages,
• manufacturer specifications,
• structured product data,
• independent testing,
• community discussion,
• institutional sources,
• or a mixture of them.

The ranking environment and the synthesis environment overlap.

But they are not identical.

Retrieval-Augmented Generation Changed the Information Flow

One of the important technical developments behind this change is Retrieval-Augmented Generation, commonly abbreviated RAG.

The foundational RAG work published by Patrick Lewis and colleagues demonstrated a system in which a language model retrieves external documents and uses them to help generate answers to knowledge-intensive questions.[3]

The broad idea is simple:

Retrieve information first. Generate from that information second.

Modern AI-search implementations vary considerably, and no public paper should be treated as a complete description of how a commercial system like Google AI Mode, ChatGPT Search or Perplexity works internally.

But the retrieval-generation architecture illustrates something strategically important.

The machine is no longer simply deciding:

“Which webpage should I rank first?”

It may be deciding:

“Which pieces of information should I use to construct the answer?”

Those are different competitions.

And they create different forms of authority.

A Page Can Be Useful to a Machine Without Being the Biggest Website

Imagine two publications.

Publication A

• Twenty years old
• Millions of backlinks
• Enormous organic traffic
• Covers thousands of categories
• Publishes a 5,000-word product review

Publication B

• Three years old
• Much smaller link profile
• Focuses exclusively on one vertical
• Publishes a regularly updated dataset
• Clearly identifies methodology
• Separates first-party claims from independent evidence
• Includes a structured comparison table
• Identifies contradictory evidence
• Provides direct citations beside individual claims

Traditional search may have strong reasons to trust Publication A.

But for a particular generative question, Publication B may offer something unusually useful:

a better evidence object.

It may be easier to retrieve.

Easier to interpret.

Easier to quote.

Easier to attribute.

More specific to the question.

More current.

More transparent about provenance.

That does not mean Publication B automatically outranks Publication A.

It means the two publications may possess different kinds of authority.

Publication A could have greater Search Authority.

Publication B could have greater Machine Authority for a particular question or category.

Machine Authority Is Probably More Granular

One reason I am hesitant to talk about “AI Domain Authority” is that the phrase may encourage us to import the wrong mental model from SEO.

Machine Authority may be far more query-specific, passage-specific and category-specific than conventional domain-level metrics suggest.

A website could be extraordinarily influential for:

“best medical alert systems for seniors living alone”

while having almost no influence over:

“best mobility scooters”

even though both topics fall under senior care.

Likewise, one page on a domain could become disproportionately influential while the rest of the site contributes little.

That suggests the unit we should study may not always be:

domain → authority

It may increasingly be:

source → claim → query → machine conclusion

That is a much more detailed graph.

The Evidence Graph

Traditional SEO has spent decades analyzing the link graph.

Who links to whom?

How many links?

How important is each linking node?

Which pages accumulate authority?

How does authority flow?

For AI search, I believe we increasingly need to map an additional structure:

the evidence graph.

The evidence graph asks:

• Which source supports which claim?
• Which claims appear across multiple sources?
• Which AI systems retrieve those sources?
• Which brands appear inside them?
• Which sources influence the final recommendation?
• Which sources are independently corroborated?
• Which claims originate with manufacturers?
• Which claims are repeated through secondary publishers?
• Which evidence remains stable over time?

The link graph measures relationships among webpages.

The evidence graph attempts to measure relationships among:

sources, claims, entities and machine conclusions.

They overlap.

But again, they are not the same thing.

Google Itself Shows Why the Distinction Matters

Google's own documentation for AI-powered search features is revealing.

Google says publishers generally do not need special AI markup or an entirely separate optimization strategy to become eligible for AI features.[4]

Pages still need to satisfy ordinary Search requirements.

They need to be crawlable.

They need to be indexed.

And standard SEO fundamentals still matter.

That tells us something important:

Machine visibility is not detached from the traditional web.

But it does not follow that the ordering of AI citations must reproduce the ordering of traditional blue links.

Eligibility is one problem.

Selection is another.

Citation is another.

Influence is another.

That is precisely why I think separating Search Authority from Machine Authority is useful.

GEO Research Suggests Generative Visibility Has Its Own Dynamics

The academic paper GEO: Generative Engine Optimization attempted to measure source visibility within generative-engine answers and test ways publishers could improve that visibility.[5]

The researchers experimented with content modifications including:

• adding citations,
• incorporating statistics,
• including quotations,
• improving fluency,
• and other presentation changes.

Their results suggested that content characteristics can affect visibility inside generative responses.

The paper should not be interpreted as a universal formula for modern commercial AI-search systems.

The field is moving too quickly for that.

But its importance is conceptual.

It demonstrates that we can treat generative visibility as its own measurable outcome.

That immediately raises the possibility that the variables associated with conventional ranking may not perfectly predict generative-source visibility.

And that is exactly the distinction Machine Authority attempts to describe.

Authority Is Becoming Two-Dimensional

A useful way to think about the emerging environment is with a simple matrix.

High Search Authority + High Machine Authority

These are the dominant incumbents.

They rank strongly in traditional search and are repeatedly surfaced, cited or relied upon by AI systems.

Think:

established authority reinforced across both systems.

High Search Authority + Low Machine Authority

These are what I would call legacy visibility assets.

They may rank extremely well but contribute surprisingly little to machine recommendations for a particular category.

Possible reasons could include:

• overly broad content,
• weak evidence structure,
• outdated information,
• low extractability,
• insufficient category specificity,
• limited third-party corroboration,
• or simply stronger competing evidence elsewhere.

The important point is not why any specific site falls into this quadrant.

The important point is that we should measure whether the quadrant exists

Low Search Authority + High Machine Authority

This may be the most interesting category in LLM optimization.

These are machine-influential sources that conventional SEO analysis could underestimate.

They may be:

• niche publications,
• specialist reviewers,
• forums,
• technical documentation,
• structured databases,
• specific research papers,
• industry associations,
• individual high-value pages,
• or emerging indexes.

A traditional link-building campaign might overlook them.

An LLMO campaign should not.

Low Search Authority + Low Machine Authority

These sources have limited influence in either system.

They may still be valuable for other reasons, but they probably do not currently sit near the center of the information ecosystem for the category being analyzed.

This Is Why Domain Authority Alone Is Incomplete for LLMO

I want to be precise here.

I am not arguing that Domain Authority is dead.

Nor am I arguing that links don't matter.

And I am certainly not claiming that AI systems somehow ignore the reputation of established publishers.

The argument is much narrower:

Traditional domain-level authority metrics do not directly measure whether a source influences machine answers.

Moz's Domain Authority does not tell us how frequently ChatGPT retrieves a page.

Ahrefs' Domain Rating does not tell us whether Gemini relies upon a publication when comparing products.

A site's backlink count does not tell us whether Google AI Mode repeatedly uses it as evidence for one commercial claim.

Those metrics were designed to model a different problem.

That doesn't make them bad metrics.

It makes them incomplete metrics for LLM optimization.

The New Question Isn't “How Strong Is This Domain?”

Suppose I am advising a company in a competitive market.

Traditional SEO analysis might produce the following report:

Competitor A has 14,000 referring domains.

Competitor B has 8,000.

Your brand has 4,000.

Useful.

Now suppose the company's real problem is:

“Why does AI keep recommending Competitor B?”

The referring-domain numbers don't answer the question.

We need another investigation.

Perhaps we discover that, across 500 recommendation queries:

• eight highly recurring sources consistently mention Competitor B,
• five describe it as the category leader,
• three independent tests validate the same benefit,
• two large community discussions reinforce the claim,
• and the company's own product information is consistently supported by third parties.

Meanwhile, our client has thousands of backlinks but weak representation inside those particular sources.

Now the problem becomes visible.

The client doesn't necessarily have an authority problem.

It may have a machine-evidence positioning problem.

That's different.

And that distinction changes the strategy.

From Authority to Citation Rating

Once Search Authority and Machine Authority are separated conceptually, another question follows naturally.

If not all sources influence machine conclusions equally, then:

Which sources matter most?

That is the problem I call Citation Rating.

Citation Rating applies network-centrality thinking to AI citation provenance.

The goal is not simply to count how often a website appears.

It is to identify which sources occupy disproportionately influential positions inside the evidence network.

A source appearing 100 times in one AI system may be less strategically important than a source appearing 40 times across five independent systems, particularly if those appearances occur close to final recommendations.

Read: Citation Rating — Why the Websites That Shape AI Opinion Matter More Than the Websites That Rank

This is where Machine Authority becomes measurable.

Brand Rating Adds the Company Layer

Once we understand source centrality, we can measure brands against that source network.

This leads to Brand Centrality.

Brand Centrality asks:

How strongly is a company represented within the information sources most influential to machine conclusions?

Consider:

Brand A

1,500 total web mentions.

But relatively few mentions among highly central sources.

Brand B

600 total web mentions.

But strong positive representation across seven of the ten most influential sources in the category.

A conventional visibility report may favor Brand A.

A machine-influence analysis may favor Brand B.

And if AI systems consistently recommend Brand B, that distribution could help explain why.

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

Consensus Gives Us Another Dimension

Our Consensus Index adds yet another layer.

Instead of measuring one machine answer, we compare multiple AI systems.

That lets us ask:

• Does the recommendation persist across models?
• Which sources recur?
• Which claims recur?
• Where do the models disagree?
• How concentrated is the supporting evidence?
• Which brands dominate the consensus?
• Which sources dominate the provenance?

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

Search Authority tells us how visible a source is within search.

Consensus tells us what multiple machines conclude.

Citation Centrality helps identify what sources influence those conclusions.

Brand Centrality identifies which companies are best positioned within those sources.

Together, these begin to form an analytical model for Machine Authority.

The Algorithmic Reciprocity Loop Connects the Two Systems Again

If authority is splitting, that does not mean the two systems remain permanently separate.

In fact, I suspect they will repeatedly feed one another.

In the first article in this series, I introduced the Algorithmic Reciprocity Loop.

The hypothesis is straightforward:

Machine recognition can lead to human discovery. Human discovery can produce conventional citations and links. Those web signals can increase future search and machine discoverability.

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

This creates an interesting possibility.

A publication could initially possess:

low Search Authority + high Machine Authority.

If the machines repeatedly expose that publication to journalists, companies, researchers and marketers, humans may begin citing it.

Those citations create conventional authority.

Over time, the publication could migrate toward:

high Search Authority + high Machine Authority.

That would be authority reciprocity in action.

It is a testable hypothesis.

And I believe it is worth measuring.

Why This Matters for PR

The split between Search Authority and Machine Authority has major implications outside SEO.

Traditional PR often asks:

“Which publications have the biggest audiences?”

Digital PR added:

“Which publications have the strongest domains and can provide authoritative links?”

Machine Relations adds:

“Which publications actually influence what machines conclude about our category?”

The answer may overlap with the biggest publications.

Sometimes it probably will.

But if the overlap is incomplete, PR strategy changes dramatically.

The objective becomes not just securing coverage.

It becomes understanding the information architecture of the market.

Which sources do machines trust?

Which sources introduce important claims?

Which sources are repeated?

Which sources challenge conventional narratives?

Which publishers dominate specific recommendation categories?

Where does misinformation originate?

Where is our competitor strongly represented?

Where are we absent?

That is what I mean by Machine Relations.

Read on CiteWorks Studios: Machine Relations — Why PR Is Becoming Citation Engineering

Machine Authority Should Not Become Another Fake Score

There is a danger here.

SEO has already created an industry full of convenient single-number metrics.

It would be easy to produce:

Machine Authority Score: 87/100

and pretend the problem is solved.

I don't think that's sufficient.

Machine Authority is probably multidimensional.

At minimum, useful measurement may need to include:

• query coverage,
• recommendation frequency,
• average recommendation position,
• cross-model agreement,
• citation frequency,
• Citation Centrality,
• source diversity,
• source concentration,
• brand representation,
• evidence quality,
• citation persistence,
• contradiction rate,
• and change over time.

Different companies will also care about different query classes.

Being authoritative for:

“What does Company X do?”

is not the same as being authoritative for:

“What is the best provider for my situation?”

And neither is identical to:

“What are the weaknesses of Company X?”

A useful Machine Authority model therefore needs to preserve intent.

Authority Is Also Becoming Temporal

Traditional SEO metrics are often treated as relatively slow-moving.

Machine authority may move faster.

A new product launches.

A large publication reviews it.

Customers begin discussing it.

A recall occurs.

A lawsuit appears.

An independent test contradicts a marketing claim.

An industry report publishes new data.

AI retrieval systems may encounter that information at different rates.

The result could be rapid changes in recommendation patterns.

This is why I believe longitudinal measurement will become essential.

A company's machine authority today is interesting.

Its machine-authority trajectory may be more valuable.

Did the brand rise from 12% recommendation share to 31%?

Did its Citation Centrality improve?

Did independent-source support increase?

Did its consensus ranking fall after a specific publication appeared?

Did one high-centrality source change its description?

Those are the kinds of questions future reputation teams may ask routinely.

Search Rank Tells You Where You Appear. Machine Authority Tells You Where You Matter.

That is probably the cleanest way I can describe the distinction.

Traditional ranking analysis asks:

Where does our website appear?

Machine-authority analysis asks:

Where does our information influence conclusions?

Those questions will frequently overlap.

But treating them as synonymous risks missing some of the most important nodes in an AI-mediated information environment.

This is why I believe LLM optimization will increasingly become less about “ranking inside ChatGPT” and more about mapping influence.

The objective is not simply to make a company's own website appear more often.

It is to understand the evidence ecosystem in which the company exists.

What We Are Trying to Measure at LLM Authority Index

The larger objective of LLM Authority Index is to make this environment observable.

We are developing a framework around several interconnected measurements.

Consensus Index

What do multiple AI systems conclude?

Citation Rating

Which sources disproportionately influence those conclusions?

Brand Rating

Which companies occupy the strongest positions within those influential sources?

Search Authority

How visible and established are the underlying sources within conventional search?

Machine Authority

How strongly do those sources and entities influence machine-generated conclusions?

Algorithmic Reciprocity

How does machine recognition generate human recognition, and how does that new human activity feed back into search and machine discoverability?

These are not six disconnected buzzwords.

They are six views of the same changing information system.

A Working Model, Not a Declaration of Algorithmic Fact

As with the other frameworks in this series, I want to separate observation from hypothesis.

There is no public Google metric called “Machine Authority.”

There is no evidence that ChatGPT calculates a score under that name.

There is no universal formula showing that a page with a particular Machine Authority value will receive an AI citation.

Machine Authority is an analytical framework I am proposing for understanding a new class of influence.

Several underlying observations are well established:

• traditional search engines use multiple ranking systems,
• links historically formed an important authority mechanism,
• modern language systems can retrieve external information,
• generative systems can expose source citations,
• and research suggests generative-source visibility can be measured.

The hypothesis is that these changes justify treating machine influence as a distinct authority dimension and measuring it separately from conventional search visibility.

That hypothesis should now be tested.

The Next Measurement Problem

Once we accept that Search Authority and Machine Authority may differ, the next question becomes unavoidable:

How do we identify the sources with the greatest machine influence?

Counting citations will not be enough.

Counting domains will not be enough.

Traditional Domain Authority will not be enough.

We need to understand the structure of the citation network itself.

That leads to the next article in this series:

Citation Centrality: Why the Websites That Shape AI Opinion Matter More Than the Websites That Rank

Because the future of LLM optimization may not be won by getting mentioned everywhere.

It may be won by understanding where a mention actually changes the information environment.

Sources and Research

1. 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 provides historical context for understanding web authority as a network problem in which hyperlinks function as citation-like signals.

2. Google Search Central — “A Guide to Google Search Ranking Systems.”
https://developers.google.com/search/docs/appearance/ranking-systems-guide

Google explains that Search relies on multiple automated ranking systems rather than one simple authority metric. This is useful context when distinguishing Google's actual ranking architecture from third-party metrics such as Domain Authority or Domain Rating.

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

Foundational research demonstrating how language generation can be combined with external document retrieval. Commercial generative-search systems differ in implementation, but RAG provides an important conceptual basis for understanding why AI answer generation creates a source-selection problem distinct from conventional ranked search.

4. Google Search Central — “AI Features and Your Website.”
https://developers.google.com/search/docs/appearance/ai-features

Google's official documentation explains that established Search fundamentals continue to apply to AI features and that publishers do not need special AI-specific files or markup simply to become eligible. This supports the view that traditional search infrastructure and generative features remain connected even though their final presentation and source-selection behavior can differ.

5. 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 visibility inside generative-engine answers as a measurable outcome. The work supports the broader premise that generative visibility constitutes an optimization environment distinct enough to study independently.

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

Research examining citation-supported generation by language models. This work contributes to the broader technical literature treating source attribution and citation as distinct components of generated answers.

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

This research examines whether claims produced by generative search systems are adequately supported by their accompanying citations. It reinforces the importance of separating the existence of a citation from the quality and completeness of the evidence behind it.

Methodology and Disclosure

Search Authority and Machine Authority are working analytical terms used by Mark Huntley and LLM Authority Index to distinguish two forms of digital influence.

Search Authority refers to persistent visibility and credibility within conventional search systems.

Machine Authority refers to the degree to which a source influences AI retrieval, citations and generated conclusions for a defined query set or category.

Machine Authority is not a publicly documented ranking factor used by Google, OpenAI, Anthropic, Perplexity or another AI provider.

The framework is intended to create testable questions, including:

• Do high-ranking websites consistently possess high machine influence?
• Which sources exhibit high machine influence despite modest traditional search visibility?
• How stable is machine influence over time?
• Which variables best predict repeated AI citation?
• How strongly does Citation Centrality relate to AI recommendation outcomes?
• Does machine visibility sometimes precede conventional backlink growth?

LLM Authority Index intends to evaluate these questions through repeated cross-platform measurement rather than assuming that conventional SEO metrics automatically explain generative-search behavior.