What Makes a Domain an AI Citation Authority? A Practical Measurement Framework

Answer Capsule

A domain should not be called an AI citation authority because it has one large citation count, appears across many markets, or ranks highly on one AI platform.

The July through September 2026 LLM Authority Index benchmark shows that the main authority dimensions are related, but far from interchangeable.

Across 283 qualifying commercial markets, 876,605 retained AI responses, and 8,718,629 citation occurrences resolved to registrable domains, 36,273 domains entered at least one eligible platform Top 100 and 2,743 reached at least one Top 10. Within those 2,743 Top-10 domains:

• Top-100 market breadth and Top-10 market breadth had a Spearman correlation of 0.608.
• Top-10 market breadth and same-market Top-10 presence on at least three AI platforms correlated at only 0.430.
• Top-10 market breadth and same-market all-six-platform Top-10 repetition correlated at only 0.229.
• Top-10 market breadth and domain-level Top-100 temporal recurrence across the balanced 246-market subset correlated at -0.090.

That is the measurement problem in one table of correlations. Broad authority, deep authority, platform repetition, and temporal recurrence are not the same characteristic.

The category dimension creates another split. Among the 102 domains that reached a Top 10 in at least 10 markets, the median domain still received 83.91% of those Top-10 markets from one macro sector. 60 of 102, 58.82%, received at least 75% of their Top-10 market breadth from one macro sector, and 20, 19.61%, were entirely concentrated in one macro sector.

Platform concentration can be equally strong. Among those same 102 domains, 17 reached the Top 10 across at least 10 markets while doing so on only one or two of the six AI platforms. 19 reached a Top 10 in at least 10 markets but never reached Top 10 on three platforms inside the same market. WorldMetrics reached Top 10 in 48 markets on only two platforms. ZipDo reached 42 markets on one. LegalClarity reached 40 markets on two.

Temporal durability also varies. Among the 102 domains with at least 10 Top-10 markets, the median source-specific Top-100 recurrence share was 73.70% across the balanced 246-market subset. The middle 50% ran from 64.54% to 85.59%. Thirteen domains were below 50%, while 39 were at or above 80%.

A practical AI citation authority profile therefore needs at least five independent views:

1. Market breadth: where the domain appears.
2. Depth: how often it reaches Top 25 or Top 10.
3. Platform repetition: whether deep authority repeats across AI systems, especially inside the same market.
4. Temporal recurrence: whether the domain remains in comparable source portfolios across observed months.
5. Category concentration: where that authority is actually concentrated.

Citation occurrences and response presence remain important exposure measures, but they answer a different question. URL relevance, publisher identity, ownership, recommendation behavior, traffic, and commercial performance require additional evidence.

This framework develops the domain leaderboard in The 2026 AI Commercial Citation Authority Index without turning the authority map into one hidden weighted score.

Key Findings

Authority Starts With the Measurement Unit

Questions This Section Answers

1. What does AI citation authority measure at the domain level?
2. Why should citation occurrences, response presence, market breadth, and source rank be reported separately?
3. What information does a domain need before it can credibly be called an AI citation authority?

The first rule is to define the unit before assigning the label.

The frozen benchmark contains 8,718,629 resolved citation occurrences and 6,825,821 response-domain appearances. Those counts are different because one retained AI response can cite the same registrable domain several times. A response-domain appearance counts that domain once for the response. Citation occurrences preserve the multiple citation events.

Neither metric is wrong. They answer different questions.

Citation occurrences describe how much citation activity was resolved to the domain inside the observed corpus.

Response presence describes how often the domain appeared across retained AI responses without allowing repeated references inside one answer to inflate the count.

The strict platform portfolios used for authority depth are ranked by response presence, not raw citation occurrences. That prevents a single answer containing several links from the same domain from counting as several independent response-level appearances.

Authority then adds additional dimensions.

A domain that appears in one response in 100 markets has a very different profile from a domain that enters the Top 10 on five AI platforms in 20 markets. A domain that appears in every month but on only one platform has a different profile from one that appears on all six platforms but only intermittently.

The measurement framework should preserve those distinctions rather than force them into a single label too early.

Market Breadth and Citation Depth Are Related, but Not the Same

Questions This Section Answers

1. Does a domain that appears in many AI source portfolios usually become a leading Top-10 source?
2. How often does broad Top-100 presence remain shallow?
3. Why should Top-10 conversion never be used as a stand-alone AI authority score?

Market breadth is the easiest authority dimension to understand.

Top-100 market breadth counts the qualifying markets in which a domain appears in at least one eligible platform Top 100. Top-10 market breadth applies the same rule at Top-10 depth.

Across the 2,743 domains that reached a Top 10 at least once, the Spearman relationship between the two breadth measures was 0.608. That is meaningful, but it is far from a one-to-one relationship.

The broad-source group shows why.

There were 116 domains with Top-100 breadth of at least 50 markets. Eight never reached a Top 10 in any eligible market-platform portfolio. Forty-six, or 39.66%, converted fewer than 10% of their Top-100 markets into Top-10 markets.

Examples include:

Those domains have real breadth. The evidence does not support describing all of them as deep authorities merely because they appear across many markets.

The inverse profile also exists.

A narrow source can convert most of its limited breadth into high source depth. CCRM Fertility entered a Top 100 in two markets and reached Top 10 in both. Engwe reached Top 100 in five markets and Top 10 in all five. ForThePeople.com reached Top 100 in 13 and Top 10 in 10. PetMD reached Top 100 in 14 and Top 10 in 11.

This is why Top-10 conversion is useful as a diagnostic ratio but dangerous as a universal rank.

Across all 2,743 Top-10 domains, Top-100 market breadth and Top-10 conversion had a Spearman relationship of -0.867. The ratio naturally favors narrow domains because a source with one Top-100 market and one Top-10 market receives 100% conversion.

The top decile of Top-100 market breadth and the top decile of Top-10 conversion had zero domains in common under the tie-inclusive thresholds used in this analysis.

That does not mean breadth and conversion are opposites in a business sense. It means a ratio with market breadth in its denominator answers a different question.

Use market breadth to describe scope. Use conversion to describe depth conditional on scope. Do not rank all publishers by conversion alone.

Platform Breadth Can Hide Platform Concentration

Questions This Section Answers

1. Does broad Top-10 market authority automatically mean a domain is strong across many AI platforms?
2. Which domains have large Top-10 market footprints but remain concentrated on one or two AI systems?
3. Why should global platform breadth be separated from same-market cross-platform repetition?

The six measured surfaces are ChatGPT, Gemini, Perplexity, Microsoft Copilot, Google AI Overviews, and Google AI Mode.

A domain can accumulate Top-10 market breadth in several ways.

It can appear on many platforms in the same markets. It can appear deeply on one platform across dozens of markets. Or it can touch all six platforms somewhere in the study while rarely repeating across them inside one category.

Those are different authority structures.

Among the 102 domains that reached Top 10 in at least 10 markets, 17 reached that depth on only one or two platforms across the entire strict cohort.

Selected examples:

In total, 19 of the 102 domains reached a Top 10 in at least 10 markets without ever reaching the Top 10 on three or more platforms inside the same market.

This is why global market breadth is not enough.

The relationship between Top-10 market breadth and same-market Top-10 repetition on at least three platforms was only 0.430. The relationship between Top-10 market breadth and same-market all-six repetition was only 0.229.

A source can therefore be broad without being broadly transferable across platforms.

The opposite profile is visible in sources such as Forbes, Bankrate, Healthline, Chewy, ForThePeople.com, and OutdoorGearLab. They do not all lead the global market-breadth table, but they repeatedly reach deep positions across several platforms inside the same market.

The Commercial Citation Authority Index reports that dimension directly rather than treating platform breadth somewhere in the corpus as equivalent to platform repetition inside one category.

Temporal Recurrence Is a Separate Authority Axis

Questions This Section Answers

1. How can a publisher measure whether its AI citation authority recurs across observed months?
2. Does high Top-10 market breadth automatically imply strong month-to-month source recurrence?
3. Which broad AI citation domains show very different temporal profiles?

Market persistence is a portfolio statistic. It should not be assigned to one domain as though the market's persistence score were the domain's own retention probability.

A domain-level durability measure has to follow the domain itself.

For this article, source-specific Top-100 recurrence was calculated on the 246 markets with all three months and all six platforms eligible. For each domain, the denominator is the balanced markets where the domain entered at least one of the three pooled monthly Top-100 portfolios. The numerator is the subset where it remained in the pooled Top 100 in all three months.

This is a descriptive recurrence share under the frozen observation window, not a forecast.

Among the 102 domains with Top-10 breadth of at least 10 markets:

• median recurrence share was 73.70%;
• Q1 was 64.54%;
• Q3 was 85.59%;
• 13 domains were below 50%;
• 39 domains were at or above 80%.

The relationship with Top-10 market breadth was essentially weak, with a Spearman correlation of -0.090.

Selected profiles illustrate the spread:

A domain can therefore be broad or deep while having materially different temporal recurrence.

For advertisers, that affects diligence. For publishers, it affects how authority should be documented. A one-period depth result and a repeated multi-month source presence are different claims.

The earlier How Stable Are AI Citation Sources Month to Month? covers market-level persistence. The domain framework here applies a separate source-specific recurrence test.

Category Concentration Is Often More Informative Than Global Breadth

Questions This Section Answers

1. How concentrated is broad Top-10 AI citation authority inside one commercial sector?
2. Can a domain be a major AI citation authority without being a generalist source?
3. Why should a CMO inspect where a domain's authority is concentrated before buying media or distribution?

Even relatively broad Top-10 authorities are often category specialists.

Among the 102 domains reaching Top 10 in at least 10 markets, the median domain received 83.91% of its Top-10 market breadth from its single largest macro sector.

The distribution is striking:

• 81 of 102, 79.41%, received at least half of their Top-10 markets from one macro sector;
• 60 of 102, 58.82%, received at least 75%;
• 20 of 102, 19.61%, were completely concentrated in one macro sector;
• the median domain covered only three macro sectors at Top-10 depth.

Selected examples show how different global footprints can hide similar concentration:

Reddit, YouTube, and Forbes have genuinely diversified commercial authority. NerdWallet and Bankrate are broad in absolute market count but still heavily finance-concentrated. G2 is strongly technology-concentrated. Chewy and ForThePeople.com are narrower specialists.

The framework should preserve those profiles.

A Finance advertiser evaluating NerdWallet should not discount it because Reddit has greater global breadth. A technology company should not treat Bankrate's global domain rank as equally relevant to G2's category concentration. The relevant authority is the authority in the actual commercial market and platform set.

This is also why the niche publisher analysis and category authority framework remain important alongside a global domain index.

The Metrics Disagree Enough That One Hidden Score Would Conceal the Evidence

Questions This Section Answers

1. Can breadth, depth, persistence, and platform repetition be reduced to one objective AI authority score?
2. How much overlap exists among the domains that rank highly on different authority dimensions?
3. What would a composite authority score hide?

A composite score can be created for a particular business decision. The evidence does not identify one naturally correct set of weights.

The top-decile comparison illustrates the problem.

Under tie-inclusive 90th-percentile thresholds among the 2,743 Top-10 domains:

• 282 domains were in the top group for Top-100 market breadth;
• 321 for Top-10 market breadth;
• 281 for Top-10 market-platform cell breadth;
• 318 for persistent Top-100 market breadth in the balanced subset.

Only 182 domains appeared in both the Top-100 breadth and Top-10 breadth groups. Their Jaccard overlap was 43.23%.

Top-100 market breadth and same-market 3+ platform repetition had only 144 domains in common, with a Jaccard overlap of 15.69%.

Top-10 market breadth and persistent Top-100 market breadth overlapped more strongly, but still imperfectly: 224 domains in common, Jaccard 53.98%.

Top-10 market breadth and Top-10 cell breadth had the strongest of these top-group relationships, with 229 domains in common and Jaccard 61.39%.

No metric pair is redundant.

The conversion ratio shows why weights can become especially misleading. The tie-inclusive top decile of market-depth conversion consisted largely of domains with 100% conversion, which naturally favors very narrow sources. It had zero overlap with the top-decile Top-100 breadth group and only five domains in common with the top-decile Top-10 market-breadth group.

A composite score would therefore require explicit choices such as:

• how many broad markets equal one additional Top-10 market;
• how much same-market cross-platform repetition should matter;
• whether a category specialist should be rewarded or penalized for concentration;
• whether temporal recurrence should outweigh current depth;
• whether Top-10 conversion should be adjusted for denominator size;
• whether the advertiser's own sector should receive more weight than unrelated markets.

Those are management choices, not facts discovered by the corpus.

The transparent approach is to publish the authority profile first. A company can then apply a disclosed decision model for its own use case.

A Practical Five-Dimension Authority Profile

A defensible publisher or source profile can fit on one table.

Two additional layers should sit beside the five dimensions.

Exposure layer

Report citation occurrences and response-domain presence separately. These describe corpus exposure and response reach. They should not replace the stricter authority dimensions.

Relevance layer

Inspect the actual cited URLs, page topics, ownership, editorial format, and relationship to the advertiser or subject. Domain-level authority does not mean every page on the domain is relevant.

That relevance layer becomes essential for sponsored content, PR, distribution, and media buying. The next article, How CMOs Should Evaluate Publishers Before Buying Sponsored Content, PR, or Distribution, applies the framework to publisher diligence.

Four Common Authority Profiles

The framework creates useful descriptive archetypes without pretending they are universal grades.

Broad and deep

These domains combine large market breadth with meaningful Top-10 depth. Reddit, YouTube, Forbes, NerdWallet, CNBC, TechRadar, Money.com, G2, and Bankrate fit different versions of this profile.

Their platform and category distributions still differ sharply.

Broad but shallow

These domains appear across many Top-100 portfolios but rarely reach Top 10. Examples include Quora, Apple, Medium, PR Newswire, Alibaba, Reuters, CB Insights, and BusinessWire under the frozen strict portfolios.

This profile is especially important in PR and distribution diligence because broad source presence can look more powerful than it is if depth is omitted.

Narrow but deep

These domains have smaller market footprints but high Top-10 conversion and often strong category concentration. Examples include Engwe, PetMD, ForThePeople.com, Chewy, CCRM Fertility, and specialist legal or insurance publishers.

Narrowness is not weakness if the business decision concerns the categories where the authority exists.

Platform-concentrated

These domains can accumulate meaningful Top-10 market breadth while remaining concentrated on one or two AI systems. WorldMetrics, ZipDo, LegalClarity, and Reuters illustrate different versions of that pattern in this dataset.

A platform-concentrated authority should not be sold as universal cross-platform authority.

How CMOs Should Use the Framework

The practical sequence begins with the market, not with a global domain leaderboard.

1. Identify the actual commercial market. Use the relevant category before global breadth.
2. Inspect Top-10 and Top-25 depth. Determine which domains are actually prominent in that market.
3. Check platform repetition. See whether the source is deep on the AI systems that matter to the business.
4. Check temporal recurrence. Separate a repeated source footprint from a one-period appearance.
5. Inspect category concentration. Understand whether global authority is relevant to the advertiser's sector.
6. Review cited URLs. Confirm the actual pages are commercially and editorially relevant.
7. Keep citation separate from recommendation. A cited publisher can support an answer without the associated company being recommended.
8. Only then evaluate commercial access. Sponsorship inventory, PR access, distribution, pricing, audience, and conversion potential are separate business questions.

That is the main value of a multidimensional authority framework.

It turns “this website gets cited by AI” into a set of measurable claims that can be checked, compared, and limited to the population the data actually supports.

Methodology and Limits

The framework uses the frozen July through September 2026 LLM Authority Index cohort of 283 qualifying tracked markets, 876,605 retained AI responses, 8,718,755 citation-array slots, and 8,718,629 citation occurrences resolved to registrable domains. The six measured surfaces are ChatGPT, Gemini, Perplexity, Microsoft Copilot, Google AI Overviews, and Google AI Mode.

The strict platform analysis contains 1,655 eligible market-platform Top-100 portfolios. A platform portfolio is eligible only when it contains at least 100 retained responses and at least 100 resolved domains. Thin or missing cells are excluded rather than assigned zero.

Within each eligible market-platform cell, domains are ranked by response presence. A domain counts once per retained response even if several links to that registrable domain appear in the answer. Ties are broken alphabetically under the frozen ranking method.

For this article:

• Top-100 market breadth counts markets where a domain appears in at least one eligible platform Top 100.
• Top-10 market breadth applies the same rule at Top-10 depth.
• Platform breadth counts the AI surfaces where a domain reaches the specified cutoff somewhere in the corpus.
• Same-market platform repetition counts markets where a domain reaches Top 10 on at least three, four, five, or six eligible platform portfolios.
• Depth conversion equals Top-10 market breadth divided by Top-100 market breadth and is treated only as a conditional ratio, not an authority score.
• Category concentration is calculated from the macro-sector distribution of a domain's Top-10 markets. The concentration analysis highlighted in the article is restricted to the 102 domains with Top-10 breadth of at least 10 markets.
• Source-specific Top-100 recurrence uses the balanced 246-market subset with all three monthly portfolios and all six platform portfolios eligible. For each domain, the denominator is the balanced markets where the domain appeared in at least one pooled monthly Top-100 portfolio. The numerator is the markets where it appeared in all three monthly Top-100 portfolios.

The pairwise metric analysis is descriptive. Spearman correlations measure rank association among the 2,743 domains that reached a Top 10 at least once. They do not establish causation or statistical independence. Some metrics share underlying counts, so correlation values should be interpreted as evidence about practical redundancy, not as separate experimental treatments.

Several limitations remain.

First, the benchmark is observational evidence from changing response samples, not a certified identical-prompt panel. Of the retained responses, 104,071 used ingestion-date proxies.

Second, domain recurrence is historical and conditional on the frozen observation window. It is not a future retention probability.

Third, domain-level authority does not establish publisher ownership, page quality, traffic, audience, paid inventory, sponsorship availability, or commercial effectiveness.

Fourth, category concentration uses the preserved retrospective taxonomy. It describes the current research grouping and does not prove the taxonomy is uniquely correct.

Fifth, platform repetition describes observed source portfolios. It does not reveal model architecture, retrieval systems, causal ranking factors, or future behavior.

Sixth, citation authority is not recommendation authority. A domain can be cited while another company is recommended, or while the answer criticizes the cited company.

Finally, no composite authority score is introduced here. Any weighted score requires a declared decision purpose and explicit weights. The research supports the component measures, not one universal formula for combining them.

Related Research

• The 2026 AI Commercial Citation Authority Index: Which Domains Have the Broadest and Deepest Citation Authority Across 283 Markets and Six AI Platforms
• Citation Breadth vs Citation Depth: Which AI Authority Metric Should CMOs Care About?
• Why Niche Publishers Matter More in AI Search Than Global Citation Rankings Suggest
• How Publishers Can Build Category Authority Without Becoming a Generalist Media Site
• How Publishers Should Think About Platform-Specific AI Citation Demand
• How Stable Are AI Citation Sources Month to Month? A CMO Guide to Citation Persistence
• How CMOs Should Evaluate Publishers Before Buying Sponsored Content, PR, or Distribution
• How Publishers Can Prove AI Citation Authority to Advertisers Without Promising Rankings
• Which AI Citation Metrics Belong on a CMO Dashboard?
• AI Citation Authority vs AI Recommendation Authority: Why Being a Source Is Not the Same as Being Recommended

The earlier Persistence-Portability framework provides conceptual context. The 2026 AI Citation Authority Study covers an earlier high-stakes sample and is a separate population.