What Sources Does Claude Cite for High-Intent Buying Questions? A 12,595-Citation Analysis

LLM Authority Index analyzed 12,595 citation events produced by Anthropic Claude across 150 high-intent buying scenarios in 10 consumer categories. Review sources represented 47.9% of all observed citation events, while company sources represented 37.7%. In detailed company-fit evaluations, 57.7% of citations were classified as independent and 34.8% as company-owned.

But the aggregate numbers conceal a much more important finding.

Claude did not maintain the same evidence pattern across industries.

In aging, safety, mobility and home-related purchase scenarios, 64.6% of fit-stage citations were independent.

In consumer credit and financial-service scenarios, that pattern reversed: 74.7% of citations were company-owned.

The model stayed the same.

The commercial environment changed.

The observable citation mix changed with it.

For brands trying to understand AI search optimization, that may be more important than any universal claim about whether Claude "prefers" first-party or third-party sources.

Key Findings From 12,595 Claude Citation Events

Answer Capsule

Across 150 high-intent buying scenarios, Claude produced 12,595 observable citation events from 1,505 normalized domains. Review sources represented 47.9% of citations, more than any other source type. In company-fit evaluations, 57.7% of citations were independent, but source ownership changed dramatically by market.

Questions This Section Answers

• What types of sources does Claude cite for high-intent buying questions?
• Does Claude cite independent sources more often than company websites?
• How much does Claude's citation behavior change across industries?

The aggregate result suggests a citation environment in which independent and review-oriented sources played a substantial role.

But the category-level data shows why aggregate AI statistics can be misleading.

Claude's citation behavior was highly dependent on the commercial market being researched.

Further Reading:

• Compare Claude’s citation behavior with frontier AI model citation patterns for high-intent buying questions.
• See how OpenAI GPT citation sources for high-intent buying questions differ from Claude’s source preferences.
• Explore how Gemini citation sources for high-intent buying questions compare with Claude across commercial research queries.
• Review how Perplexity citation sources for high-intent buying questions differ in the sources used to support recommendations.
• Compare Claude with Grok citation sources for high-intent buying questions to see how their citation patterns vary.

What Did LLM Authority Index Test With Claude?

Answer Capsule

LLM Authority Index submitted 150 narrowly defined commercial buying scenarios to Anthropic Claude across 10 consumer categories. Claude independently ranked products or services for each buyer use case, after which recommended companies received deeper fit evaluations. The analysis includes 1,354 recommendations and 12,595 citation events.

Questions This Section Answers

• How was the Claude citation study conducted?
• How many high-intent prompts were tested?
• Which Claude models were included?
• What counts as a high-intent buying question in this research?

The study focused on buyer decisions rather than broad informational questions.

A typical ranking prompt followed a structure similar to:

Identify and rank the best [product or service] for the following narrowly defined buyer need.

The prompts supplied commercial context including:

• target buyer
• specific use case
• product or service requirements
• geography where relevant
• important buying criteria
• research year
• maximum number of recommendations

The dataset covered 10 categories.

Aging, Safety, Mobility and Home-Related Categories

• Medical alert systems
• Home safety
• Senior technology
• Stairlifts
• Walk-in tubs

Consumer Credit and Financial-Service Categories

• Credit repair
• Credit building and rebuilding
• Credit monitoring and scores
• Debt relief
• Personal and debt consolidation loans

The ranking layer contained 150 Claude responses.

The configured Claude model was Claude Haiku 4.5 for 148 of the 150 ranking scenarios. Two ranking scenarios used Claude Sonnet 5 configurations.

The deeper fit stage included 1,069 company evaluations, with the large majority also generated through Claude Haiku 4.5.

For that reason, this study describes the results at the Claude family level while disclosing the exact model composition in the methodology.

The research was collected between July 27 and September 9, 2026.

What Types of Sources Does Claude Cite?

Answer Capsule

Review sources were Claude's largest observed source category, representing 47.9% of all 12,595 citation events. Company sources represented 37.7%. Directories, government sources, other sources and journalism collectively accounted for the remaining 14.4%.

Questions This Section Answers

• Does Claude cite review websites or company websites more often?
• What percentage of Claude citations come from independent review sources?
• What source types appear most frequently in Claude buying recommendations?

Across both stages of the research, Claude's citation events were classified as follows:

Review sources were the largest category.

That is notably different from the OpenAI results in the same research dataset, where company sources represented the dominant source type.

This does not establish that Claude internally "trusts reviews more."

The research cannot observe Anthropic's proprietary retrieval weighting or source-selection logic.

What it establishes is narrower:

Review sources appeared more frequently than company sources in the Claude citation events observed across these high-intent buyer scenarios.

Does Claude Cite More Independent or Company-Owned Sources?

Answer Capsule

Independent evidence represented the majority of Claude's detailed company-fit citations. Of 10,326 fit-stage citation events, 57.7% were classified as independent, 34.8% as company-owned and 7.6% as unclear. The aggregate result, however, changed sharply when the data was separated by industry.

Questions This Section Answers

• What percentage of Claude citations are independent?
• How often does Claude cite company-owned websites?
• Does Claude rely primarily on third-party evidence when evaluating companies?

The company-fit stage contained explicit source-ownership classifications.

Across 10,326 citations:

At the aggregate level, independent evidence clearly exceeded company-owned evidence.

Approximately:

58 of every 100 Claude fit-stage citations were independent.

About:

35 of every 100 were company-owned.

But those percentages should not be converted into a universal AI optimization rule.

When the same data is separated by commercial category, Claude's evidence mix changes substantially.

Does Claude's First-Party vs. Independent Citation Mix Change by Industry?

Answer Capsule

Yes. Claude's source-ownership mix varied dramatically across the 10 commercial categories. Independent evidence represented 69.9% of medical-alert citations and 69.1% of senior-technology citations. In personal and debt consolidation loans, however, 84.9% of citations were company-owned.

Questions This Section Answers

• Does Claude use the same evidence mix in every industry?
• Which categories have the highest independent-source citation rates?
• Which categories have the highest company-owned citation rates?

The category-level results reveal two very different citation environments.

The difference is substantial.

Claude's company-owned citation rate was:

21.3% for medical alert systems

but:

84.9% for personal and debt consolidation loans.

That is a difference of more than 63 percentage points.

The model did not simply exhibit one consistent "third-party citation preference."

Its observable evidence environment changed materially with the commercial category.

Does Claude Use Different Sources for Consumer Products and Financial Services?

Answer Capsule

Claude showed almost opposite source-ownership patterns across the two research cohorts. In aging, safety, mobility and home-related categories, 64.6% of fit-stage citations were independent and 26.5% company-owned. In consumer credit and financial services, 74.7% were company-owned and only 24.0% independent.

Questions This Section Answers

• Does Claude citation behavior differ between consumer products and financial services?
• Does Claude rely more on first-party evidence in finance?
• How much can citation behavior change without changing the AI model?

Aggregating the 10 categories into two broader commercial groups makes the reversal especially clear.

This may be one of the most important findings in the Claude dataset.

In one cohort, independent sources outnumbered company-owned citations by more than two to one.

In the other cohort, company-owned citations outnumbered independent citations by more than three to one.

The model family remained Claude.

The prompt framework remained standardized.

The commercial domain changed.

And the observable evidence mix changed dramatically.

This is why model-level averages alone may be insufficient for AI search optimization.

Brands may need to understand citation behavior at the intersection of:

model + industry + buyer intent

rather than simply asking what "Claude tends to cite."

What Sources Does Claude Cite for Medical Alert Systems?

Answer Capsule

Medical alert systems produced one of Claude's most independent-source-heavy citation environments. Independent sources represented 69.9% of fit-stage citations, while company-owned sources represented 21.3%. Review-oriented publishers such as SeniorLiving.org, TheSeniorList and NCOA appeared repeatedly in Claude's broader aging-related research.

Questions This Section Answers

• Does Claude rely heavily on independent sources for medical alert recommendations?
• What types of sources appear around Claude's medical alert recommendations?
• Why might third-party visibility matter for medical alert brands?

Medical alert systems generated:

1,992 fit-stage citation events

Of those:

• 69.9% were independent
• 21.3% were company-owned
• 8.8% were unclear

Across the larger aging-related dataset, Claude repeatedly surfaced established review, senior-living and consumer-information publishers.

This creates a substantially different observable evidence environment from categories where first-party sites dominate.

For a medical alert company, optimizing only its own website would therefore ignore a large portion of the sources Claude surfaced in this dataset.

The appropriate question becomes:

Which independent domains repeatedly appear when Claude evaluates the high-intent buyer scenarios our company wants to win?

That is measurable at the prompt level.

What Sources Does Claude Cite for Credit and Financial-Service Questions?

Answer Capsule

Claude's financial-service citation pattern was far more first-party oriented. Company-owned sources represented 74.7% of fit-stage citations across the five credit and financial categories. The rate exceeded 82% for credit monitoring, credit building and rebuilding, and 84% for personal and debt consolidation loans.

Questions This Section Answers

• Does Claude cite financial companies' own websites?
• How first-party-heavy is Claude's evidence mix for credit products?
• Which financial categories had the highest company-owned citation rates?

Across the consumer credit and financial-service cohort:

1,766 fit-stage citations were recorded.

The ownership distribution was:

• 74.7% company-owned
• 24.0% independent
• 1.3% unclear

Several categories were even more concentrated.

Personal and Debt Consolidation Loans

84.9% company-owned

Credit Building and Rebuilding

82.5% company-owned

Credit Monitoring and Scores

82.3% company-owned

Credit Repair

76.0% company-owned

Debt relief was the major exception, with:

54.8% company-owned

and:

44.8% independent

This suggests that even within one broad vertical such as consumer finance, the evidence mix can change based on the specific purchase decision.

Which Domains Did Claude Cite Most Often?

Answer Capsule

Claude cited 1,505 normalized domains across the full dataset. The most frequent included SeniorLiving.org, RetirementLiving.com, TheSeniorList.com, NCOA.org, SafeHome.org, ConsumerAffairs.com, SafeWise.com, FTC.gov and USNews.com. Its 10 most frequently cited domains accounted for 24.8% of all citation events.

Questions This Section Answers

• Which websites did Claude cite most frequently?
• What are the most visible citation domains in Claude's high-intent responses?
• How concentrated is Claude's citation activity among leading domains?

The most frequently observed normalized domains across all Claude citation events included:

Claude's ten most frequently cited domains collectively represented:

24.8% of all Claude citation events

This means a relatively small group of publishers accumulated meaningful citation volume.

At the same time, the dataset contained:

1,505 normalized domains

So the overall source universe was still broad.

This distinction matters for AI optimization.

A brand may need to understand both:

High-Frequency Citation Sources

Domains that accumulate substantial citation activity across many responses.

Prompt-Specific Evidence Sources

Domains that become relevant only for a particular product, buyer need or commercial scenario.

The first can reveal broad citation authority.

The second can reveal the evidence environment around an actual purchase decision.

Which Domains Appeared Across the Most Claude Buying Scenarios?

Answer Capsule

SeniorLiving.org appeared in Claude ranking citations across 45 of the 150 high-intent buyer scenarios. RetirementLiving.com and TheSeniorList.com each appeared across 36, NCOA.org across 35 and LendingClub.com across 28. Prompt breadth provides a different measure from raw citation frequency.

Questions This Section Answers

• Which websites appear across the broadest range of Claude buying questions?
• What is the difference between citation frequency and prompt coverage?
• Which domains have broad Claude citation visibility?

At the initial ranking stage, several domains appeared across a large number of distinct buying scenarios.

This gives us a second way to think about citation authority.

A domain may receive many repeated citations within a smaller number of evaluations.

Another domain may appear less frequently per response but surface across a much broader range of buyer scenarios.

Those are different signals.

Citation Frequency

How often is the domain cited?

Prompt Breadth

Across how many distinct high-intent buyer scenarios does the domain appear?

Cross-Model Citation Authority

Across how many different AI systems does the domain surface?

A useful AI citation measurement system should distinguish among all three.

How Many Sources Does Claude Surface Per High-Intent Buying Question?

Answer Capsule

Claude's 150 initial ranking responses averaged 15.1 citation events and 10.2 unique normalized domains per high-intent buying scenario. The median response contained 16 citation events and 10 unique domains. Eleven responses contained no ranking-stage citations.

Questions This Section Answers

• How many citations does Claude provide in a typical commercial response?
• How many unique websites appear in a Claude buying answer?
• Does every Claude response contain citations?

Across the 150 standardized ranking responses, Claude produced:

2,269 ranking-stage citation events

That equals an average of:

15.1 citation events per response

The median was:

16 citation events

When citations were reduced to unique normalized domains, Claude averaged:

10.2 unique domains per ranking response

The median was:

10 unique domains

Eleven of the 150 responses contained no ranking-stage citation.

These raw counts are useful for understanding Claude's responses, but they should not be used by themselves to compare Claude with another model.

Different models can generate different response lengths and numbers of citations.

For cross-model comparisons, normalized measurements such as:

• prompt-level domain overlap
• source-type percentages
• source ownership percentages
• prompt coverage
• recommendation rate

are generally more informative than raw citation totals alone.

Is Claude's Citation Ecosystem More Diverse Than Other Frontier Models?

Answer Capsule

Claude surfaced 1,505 normalized domains, the largest raw domain count among the seven model families in this dataset. It also produced the highest raw citation-event total at 12,595. However, Claude generated more citations per response than several other models, so raw volume should not be interpreted as proof that Claude inherently uses a broader retrieval system.

Questions This Section Answers

• Did Claude cite more domains than other frontier models?
• Does Claude have a broader citation ecosystem?
• Why should raw citation counts be normalized before comparing AI platforms?

In the full seven-model dataset, Claude produced:

12,595 citation events

and:

1,505 normalized domains

Both were the largest raw totals among the seven model families studied.

But there is an important methodological limitation.

Claude also produced relatively citation-rich responses.

At the ranking stage, Claude averaged approximately:

15.1 citation events per response

A model that generates more citations has more opportunities to produce unique domains.

Therefore, it would be inappropriate to conclude from the raw totals alone that:

Claude has the broadest underlying retrieval system.

We cannot observe the underlying retrieval system.

What we can say is:

Claude surfaced the largest raw number of citation events and normalized domains in this dataset.

Comparative claims require normalization for response structure and citation volume.

Does Claude Cite the Same Sources as OpenAI, Gemini or Other Frontier Models?

Answer Capsule

Only partially. In the broader matched-prompt study, Claude's average domain overlap with OpenAI was 15.0% and its overlap with Gemini was 12.9%. Claude's average overlap with Kimi was only 7.3%. These results indicate substantial source divergence even when models answer the same commercial question.

Questions This Section Answers

• Does Claude cite the same sources as OpenAI?
• How much source overlap exists between Claude and Gemini?
• Can citation visibility in another AI platform predict Claude visibility?

Because all seven model families answered the same 150 high-intent ranking scenarios, their domain sets can be compared directly at the prompt level.

Selected Claude pairings included:

Even Claude's highest average overlap in this group was only 15%.

That means strong citation visibility in another frontier model does not establish strong Claude visibility.

A domain can be important in one model's evidence environment and largely absent from another.

Cross-model citation authority has to be measured directly.

What Does Claude's Independent-Source Pattern Mean for AI Optimization?

Answer Capsule

Claude's citation data suggests that third-party evidence can be extremely important for some commercial categories, but first-party optimization remains essential in others. The correct strategy depends on the buyer-intent cluster and industry. A universal rule such as "get more third-party mentions" is not supported by the data.

Questions This Section Answers

• What should brands do if Claude cites independent sources heavily?
• Should companies focus on third-party content for Claude optimization?
• Is first-party website optimization still important for Claude?

A simplistic interpretation of the aggregate result would be:

Claude likes third-party reviews, so brands need more third-party reviews.

The category data shows why that conclusion is inadequate.

For medical alert systems:

69.9% of fit-stage citations were independent.

For personal and debt consolidation loans:

84.9% were company-owned.

Those require very different evidence strategies.

A more defensible Claude optimization process begins with measurement.

Step 1: Identify Commercial Prompt Clusters

Determine the buyer questions that matter economically.

These might involve:

• best provider
• product fit
• pricing
• comparisons
• specific use cases
• limitations
• eligibility
• features
• buyer circumstances

Step 2: Measure Recommendation Visibility

Determine whether Claude:

• mentions the company
• considers the company
• recommends the company
• ranks the company prominently

Step 3: Map Claude's Citation Environment

Identify which sources appear around those exact prompt clusters.

Separate:

• first-party company sources
• independent review sources
• journalism
• government sources
• directories
• other publishers

Step 4: Compare Source Claims

Determine whether the information surfaced by Claude is:

• current
• accurate
• complete
• internally consistent
• aligned with the company's actual products
• aligned across first-party and independent sources

Step 5: Identify Evidence Gaps

Compare the company's evidence environment with competitors that Claude recommends more frequently.

The objective is not to create artificial consensus.

It is to make accurate product and company information easier to retrieve from the evidence environments Claude actually surfaces.

Why Claude's Citation Pattern Makes Third-Party Accuracy Important

Answer Capsule

In categories where Claude heavily cited independent sources, a company's own website represented only part of the observable evidence environment. Outdated pricing, incorrect product descriptions or missing differentiators on third-party sites could therefore coexist with perfectly accurate first-party content.

Questions This Section Answers

• Why do third-party sources matter for Claude AI visibility?
• Can accurate company content be undermined by inaccurate external information?
• What should brands audit outside their own websites?

Consider a hypothetical brand whose website accurately states:

• current pricing
• available products
• cancellation terms
• service areas
• features
• eligibility

Now suppose several independent sources say something different.

They may contain:

• obsolete pricing
• discontinued products
• incorrect fees
• missing capabilities
• outdated comparisons
• inaccurate geographic availability

If Claude is heavily surfacing independent evidence for that category, correcting only the company website leaves the external inconsistency unresolved.

The practical task becomes an evidence consistency audit.

Brands can compare:

What the company says

with:

What independent sources say

with:

What Claude says

That creates an observable chain that can be researched and corrected when factual discrepancies exist.

Is Being Cited by Claude the Same as Being Recommended by Claude?

Answer Capsule

No. Citation authority and recommendation authority are separate. A third-party publisher can have substantial Claude citation authority without selling the product being recommended. A company can also be recommended even when other domains provide much of the supporting evidence.

Questions This Section Answers

• Does a Claude citation mean a company is recommended?
• What is Claude citation authority?
• How is citation visibility different from recommendation visibility?

An AI buying answer can contain at least three distinct entities.

Recommended Company

The brand or product Claude presents as a suitable choice.

First-Party Evidence Source

The company's own website providing product facts.

Independent Evidence Source

An external publisher supporting or contextualizing those facts.

Those entities should not be collapsed into one metric.

Claude Citation Authority

How frequently does a domain appear as evidence in Claude responses?

Claude Prompt-Specific Citation Authority

Does the domain appear around commercially important prompt clusters?

Claude Recommendation Authority

How frequently does Claude actually recommend the company?

Claude Recommendation Position

Where does the company appear within Claude's recommended options?

Claude Consensus Recommendation Authority

Do other frontier models independently make the same recommendation for the same buyer scenario?

The distinction matters because:

a source can influence the visible evidence environment without being the recommended entity.

Does Claude Use One Universal Set of Authoritative Sources?

Answer Capsule

The data does not support the idea of one universal Claude authority list. Claude surfaced 1,505 normalized domains, and the dominant evidence type changed substantially by industry. Some publishers appeared broadly, but many citation sources were relevant only to particular buyer scenarios or markets.

Questions This Section Answers

• Is there a universal list of websites that Claude considers authoritative?
• Can brands simply target the most frequently cited Claude domains?
• Does citation authority depend on the buyer question?

The most commonly cited domains are useful.

They reveal publishers with broad observable citation presence.

But a list of top domains cannot answer:

Which sources matter for this specific company trying to win this specific buying decision?

That requires prompt-level analysis.

A domain with broad citation authority might have no meaningful role in a particular product category.

Conversely, a highly specialized publisher might appear repeatedly for one commercially valuable prompt cluster even though it is rarely cited elsewhere.

This suggests at least three different forms of authority:

Broad Citation Authority

The domain appears frequently across many Claude responses.

Category Citation Authority

The domain appears repeatedly within a particular commercial market.

Intent-Specific Citation Authority

The domain appears around a particular buyer need or purchase decision.

For commercial AI optimization, the third may ultimately be the most actionable.

Does This Research Reveal How Claude Internally Chooses Sources?

Answer Capsule

No. The study measures Claude's observable outputs and citations. It does not reveal Anthropic's proprietary retrieval architecture, source weighting, training data, trust systems or hidden reasoning. Citation frequency describes what appeared in the research responses, not why Claude internally selected it.

Questions This Section Answers

• Does this study reveal Claude's retrieval algorithm?
• Does a citation prove Claude trusts a source?
• Can citation frequency establish why Claude recommended a company?

The study does not provide access to:

• proprietary retrieval systems
• internal ranking mechanisms
• source trust scores
• model reasoning
• complete training datasets
• hidden weighting factors

We therefore avoid claims such as:

Claude trusts review sites more than company websites.

The observable claim is:

Review sources represented 47.9% of Claude citation events in this dataset.

Likewise:

Independent sources represented 57.7% of Claude's fit-stage citation events.

Those are measured results.

The internal mechanism producing those results remains unknown.

What This Claude Citation Study Does Not Prove

Answer Capsule

This research describes Claude citation behavior across 150 high-intent buying scenarios in 10 consumer categories. It does not prove universal behavior across all Claude products, industries or query types. It also does not establish that citations caused recommendations or that traditional SEO metrics predict Claude visibility.

Questions This Section Answers

• What are the limitations of the Claude citation study?
• Does this prove backlinks affect Claude citations?
• Can these findings be generalized to every Claude response?

Several limitations are important.

The Study Focuses on Commercial Buying Questions

The dataset does not represent all possible Claude interactions.

It should not automatically be generalized to:

• educational questions
• coding
• news
• general knowledge
• navigational questions
• local search
• every B2B category

Citation Does Not Equal Causation

A source appearing in a response does not prove that it caused the recommendation.

The Study Does Not Test Traditional SEO Metrics

This analysis does not establish relationships between Claude citations and:

• Domain Rating
• backlinks
• referring domains
• Google rankings
• organic traffic

Those require separate joined datasets.

Multiple Claude Configurations Were Present

The ranking dataset consisted overwhelmingly of Claude Haiku 4.5, with two Sonnet 5 configurations.

The results are therefore presented at the Claude-family level rather than implying every observation came from one identical model configuration.

Source Classification Is an Analytical Layer

Source type and ownership classifications are part of the structured research dataset.

They support aggregate analysis but should not be interpreted as an independent manual audit of every publisher.

How Was the Claude Citation Dataset Normalized?

Answer Capsule

LLM Authority Index separated Claude's 2,269 ranking-stage citations from 10,326 company-fit citations. Ranking responses were used for matched prompt and cross-model comparisons. Fit-stage citations were used for first-party versus independent analysis. Domains were normalized for domain-level concentration and breadth measurements.

Questions This Section Answers

• How were Claude citations analyzed?
• Why are ranking citations and fit-stage citations separated?
• How were domains normalized?

The study contains two related but different evidence layers.

Ranking Stage

Claude independently answered all 150 standardized buying scenarios.

This produced:

• 150 ranking responses
• 1,354 recommendations
• 2,269 citation events

Ranking-stage data is particularly useful for:

• matched-prompt comparison
• cross-model source overlap
• domain breadth
• prompt-specific citations
• recommendation evidence

Company-Fit Stage

Companies surfaced through the research process were evaluated in greater detail.

This produced:

• 1,069 company-fit evaluations
• 10,326 citation events

The fit stage is particularly useful for:

• first-party versus independent evidence
• detailed source composition
• product claims
• pricing evidence
• limitations
• use-case fit

Repeated citation events were retained because citation frequency is itself a measurable behavior.

For domain analysis, common domain variants were normalized to the underlying root domain where appropriate.

What Is the Main Finding About Claude Citation Sources?

Answer Capsule

Claude's aggregate citation environment leaned toward reviews and independent evidence, but that description is incomplete. Independent citations dominated aging and home-related categories at 64.6%, while company-owned citations dominated consumer finance at 74.7%. Claude citation behavior therefore appears strongly dependent on commercial context.

Questions This Section Answers

• What is the main conclusion from the Claude citation study?
• Does Claude prefer independent or first-party sources?
• What should brands learn from Claude's industry differences?

If we looked only at the aggregate dataset, the headline would be:

57.7% independent

versus:

34.8% company-owned

But the cohort data tells a more useful story.

Aging, Safety, Mobility and Home

64.6% independent

Consumer Credit and Financial Services

74.7% company-owned

That reversal is difficult to reconcile with a universal statement such as:

Claude prefers third-party sources.

The evidence instead points toward a more specific framework:

Claude's observable citation environment varies materially according to the commercial market and buyer question.

For brands, the workflow becomes:

1. Define the high-intent buyer scenarios that matter commercially.
2. Test Claude against those scenarios.
3. Measure recommendation presence and position.
4. Identify the sources Claude actually surfaces.
5. Separate first-party from independent evidence.
6. Measure category-specific citation patterns.
7. Compare the company's evidence environment with better-performing competitors.
8. Correct factual inconsistencies and fill legitimate information gaps.
9. Re-run the same prompt set over time.

The question is no longer simply:

How do we optimize our website for Claude?

It becomes:

What evidence environment does Claude surface when a buyer asks the questions that generate revenue for our company?

That is a narrower question.

It is also measurable.

Study Methodology

Answer Capsule

LLM Authority Index analyzed Anthropic Claude across 150 standardized high-commercial-intent buyer scenarios in 10 consumer categories. The dataset contains 150 ranking responses, 1,354 recommendations, 1,069 detailed company-fit evaluations and 12,595 observable citation events collected between July 27 and September 9, 2026.

Questions This Section Answers

• What is the sample size of the Claude citation study?
• Which Claude models were included?
• How many citations and company evaluations were analyzed?
• When was the research collected?

Research Scope

• Model family: Anthropic Claude
• Primary ranking configuration: Claude Haiku 4.5
• Haiku ranking responses: 148
• Sonnet 5 ranking responses: 2
• High-intent buyer scenarios: 150
• Consumer categories: 10
• Standardized ranking responses: 150
• Ranking recommendations: 1,354
• Detailed company-fit evaluations: 1,069
• Ranking-stage citation events: 2,269
• Fit-stage citation events: 10,326
• Total citation events: 12,595
• Normalized domains observed: 1,505
• Collection period: July 27 through September 9, 2026

Primary Research Question

What sources does Claude surface when answering narrowly defined, high-commercial-intent buying questions?

Secondary Research Questions

• What source types appear most frequently?
• How much evidence is independent versus company-owned?
• Does source ownership change by industry?
• Which domains appear most often?
• Which domains appear across the widest range of buyer scenarios?
• How does Claude's citation environment compare with other frontier models?
• How should citation authority be distinguished from recommendation authority?

Important Measurement Definitions

Citation event: One recorded citation occurrence within a model response.

Normalized domain: A citation source normalized to the domain level for aggregate analysis.

Company-owned source: A citation classified as controlled by the company being evaluated.

Independent source: A citation classified as external to the company being evaluated.

Ranking response: Claude's response to a standardized high-intent buyer scenario.

Company-fit evaluation: A second-stage analysis of a specific company's suitability for the defined buyer need.

Prompt breadth: The number of distinct buyer scenarios in which a domain appears.

The research measures observable model outputs and citation behavior.

It does not claim access to Anthropic's proprietary retrieval architecture, ranking systems or hidden reasoning.

About LLM Authority Index

LLM Authority Index measures how companies, sources and domains appear across high-intent AI recommendation environments.

Our research separates signals that are often incorrectly treated as interchangeable:

• mentions
• citations
• consideration
• recommendations
• recommendation position
• source ownership
• prompt coverage
• cross-model consensus

The objective is to measure what frontier AI systems actually surface and recommend for commercially meaningful buyer questions, then track how those environments differ by model, industry and time.

For brands, agencies and researchers, LLM Authority Index provides prompt-cluster benchmarking, citation analysis and cross-model recommendation measurement across major AI platforms.