Top 25 vs Top 100 AI Citation Sources: Where Durable Authority Actually Lives

Answer Capsule

The most prominent AI citation sources formed a materially more durable core than the broader Top-100 source universe across the 283-market benchmark.

At the Top 25 level, median month-to-month citation persistence was 72.41%. At the Top 100 level, it fell to 63.93%. The difference was broad across markets: 246 of 283 markets, 86.93%, had higher persistence at Top 25 than Top 100, and 152 markets, 53.71%, gained at least 10 percentage points of persistence when the analysis narrowed to the Top 25.

The strongest sources were also somewhat more portable across platforms. Median portability rose from 22.70% at Top 100 to 28.21% at Top 25. But the increase in persistence was larger than the increase in portability. As a result, the median within-market Persistence-Portability Gap widened from 39.68 points at Top 100 to 46.99 points at Top 25.

That widening occurred in 207 of 283 markets, 73.14%. In another 75 markets the gap narrowed, and one market was unchanged. The pattern is therefore strong, not universal.

The depth effect becomes even clearer when looking at thresholds. At Top 25, 193 markets, 68.20%, had persistence of at least 70%. At Top 100, only 61 markets, 21.55%, reached 70%. Nearly half of markets, 137 of 283, reached at least 75% persistence in their Top-25 portfolios, compared with only 20 markets at Top 100.

Twenty-three markets showed an especially sharp core-versus-tail split: Top-25 persistence of at least 75% combined with Top-100 persistence of 60% or less. These included Mortgage Refinance Lenders, Payroll Software, Credit Cards for Building Credit, Online Stock Brokers, Home Warranties, Cloud Storage, SIEM Software, Home Builders, and others.

The strongest example was VPN Services. Its Top-25 monthly domain portfolios were identical across all three month pairs, producing 100% persistence, while its Top-100 persistence was only 63.93%. The durable core stayed fixed while the broader source set rotated substantially.

But there are important counterexamples. Natural Skincare Brand had only 56.25% Top-25 persistence and 72.41% Top-100 persistence. In that market, the highest-ranked portion churned more than the broader source universe. A Top-25 view can therefore reveal durable authority, but it should not replace the Top-100 view.

The practical conclusion for CMOs and publishers is to measure source depth as a layer, not choose one universal cutoff. Top 25 is useful for identifying the durable authority core. Top 100 is useful for understanding the broader opportunity set, emerging sources, and source turnover around that core.

This analysis is descriptive. It measures domain-level portfolio overlap in changing stored samples. It does not prove that Top-25 publishers are more influential, more valuable, more trustworthy, or more likely to cause company recommendations.

Key Findings

Portfolio Depth Changes the Question Being Asked

Questions This Section Answers

1. What is the difference between measuring the Top 25 and Top 100 AI citation sources in a commercial market?
2. Does a Top-25 AI citation portfolio measure the same type of authority as a Top-100 portfolio?
3. Which portfolio depth should a CMO use when evaluating durable third-party AI citation authority?

A Top-K source portfolio is not simply a bigger or smaller version of the same metric.

The cutoff changes the strategic question.

A Top-25 portfolio asks:

Which domains occupy the most prominent part of the observed source ecosystem, and how durable is that core?

A Top-100 portfolio asks:

How stable is the broader set of domains that meaningfully participates in the observed market source environment?

Both matter.

The study ranks registrable domains by the number of distinct retained responses in which the domain appeared. Repeated citations to the same domain inside one response count once for portfolio ranking. Ties are broken alphabetically.

Importantly, the Top-25 and Top-50 sensitivity analyses did not introduce weaker cells or a different group of markets. All three cutoffs use the same 283 qualifying markets and the same baseline cell requirement of at least 100 retained responses and at least 100 available resolved domains. The only change is the number of highest-ranked domains retained from each eligible portfolio.

That makes the depth comparison unusually clean for a descriptive sensitivity analysis. When Top-25 persistence differs from Top-100 persistence, the difference is not caused by a different market population entering the analysis.

It reflects how the observed source portfolio behaves as attention moves from the leading domains into the broader source set.

The Top-25 Source Core Was Much More Persistent Than the Top 100

Questions This Section Answers

1. How much more stable were Top-25 AI citation-source portfolios than Top-100 portfolios across 283 commercial markets?
2. How many markets had at least 70% or 75% month-to-month persistence among their Top-25 citation sources?
3. Does most AI citation turnover occur below the most prominent source tier?

The headline median rises by about 8.48 percentage points when the portfolio narrows from Top 100 to Top 25:

The full distribution shows how substantial that change is.

The Top-25 portfolio therefore creates a very different picture of source durability.

At Top 100, only about one in five markets reached 70% persistence. At Top 25, more than two-thirds did.

At Top 100, only two markets reached 80% persistence. At Top 25, 58 markets did.

Across individual markets, 246 of 283 had higher persistence at Top 25 than Top 100. The median increase was 10.50 percentage points.

The magnitude was often large:

• 204 markets, 72.08%, gained at least 5 points.
• 152 markets, 53.71%, gained at least 10 points.
• 91 markets, 32.16%, gained at least 15 points.
• 39 markets, 13.78%, gained at least 20 points.
• 15 markets, 5.30%, gained at least 25 points.

This supports a useful interpretation: much of the observed month-to-month source turnover sits outside the most prominent authority core.

That does not mean the Top 25 are fixed. It means the leading source set was usually more stable than the larger pool around it.

Twenty-Three Markets Had a Highly Stable Top 25 but a Much Less Stable Top 100

Questions This Section Answers

1. Which markets had at least 75% Top-25 citation persistence but 60% or less Top-100 persistence?
2. Where does the AI citation data show the clearest separation between a durable authority core and a rotating source tail?
3. Which commercial markets would be most misleading if a CMO looked only at Top-100 source stability?

Twenty-three markets combined Top-25 persistence of at least 75% with Top-100 persistence of 60% or less.

Selected examples are below.

These markets demonstrate why source-depth reporting matters.

A Top-100 dashboard could describe Mortgage Refinance Lenders as a market with moderate source stability. A Top-25 dashboard would show a much more durable leading core.

The two observations are not contradictory. They describe different layers of the source ecosystem.

For a CMO, that difference can change the order of operations. The durable core may deserve continued monitoring, publisher diligence, factual alignment, or relationship development. The broader set may be better treated as an opportunity and discovery layer that is expected to rotate more frequently.

VPN Services Had a Perfectly Stable Top 25 and a Much More Fluid Top 100

Questions This Section Answers

1. Can the same 25 AI citation sources remain identical across months while the broader Top-100 source portfolio changes substantially?
2. What does the VPN Services market reveal about a stable citation core and a rotating long tail?
3. How should marketers interpret 100% Top-25 persistence when Top-100 persistence is much lower?

VPN Services provides perhaps the cleanest example of a durable authority core.

Its market-level results were:

• Top-25 persistence: 100.00%
• Top-50 persistence: 88.68%
• Top-100 persistence: 63.93%
• Top-25 portability: 47.06%
• Top-100 portability: 29.87%

All three monthly Top-25 comparisons produced 100% Jaccard overlap.

The same 25 domains occupied the Top-25 source set in July, August, and September.

The Top-100 lists were far less stable:

For July versus August, the Top-25 portfolios shared all 25 domains.

At Top 100, the two months shared 77 domains, leaving 23 domains unique to July and 23 unique to August.

The changing portion contained a mix of specialist and general sources. July-only examples in that Top-100 comparison included domains such as Astrill, AppleInsider, Bitcatcha and GitHub. August-only examples included Kaspersky, IVPN, Check Point and Business Insider.

The point is not that one set was better. It is that the tail rotated around an unchanged core.

For an AI citation strategy, that is a fundamentally different condition from a market where even the top authority set is unstable.

Travel Insurance Shows How the Core Can Be Nearly Fixed While the Tail Rotates

Questions This Section Answers

1. How much more stable were the Top-25 AI citation sources than the Top-100 sources in the Travel Insurance market?
2. Can nearly identical leading source portfolios coexist with substantial turnover deeper in the same AI citation market?
3. Why should a CMO examine both shared core publishers and rotating secondary sources in travel insurance?

Travel Insurance produced another sharp depth effect.

Its market-level persistence was 92.31% at Top 25 and 61.29% at Top 100.

For July versus August:

• 24 of the 25 Top-25 domains were shared.
• The Jaccard overlap was 92.31%.
• At Top 100, only 76 domains were shared.
• The Top-100 Jaccard overlap was 61.29%.
• Twenty-four domains were unique to July and 24 were unique to August.

The Top-25 difference was literally one domain on each side. July included Haznos in the Top 25 while August included Facebook.

The Top-100 turnover was much broader. July-only examples included AAA, CBS News, ConsumersAdvocate.org, GEICO and Emergency Assistance Plus. August-only examples included Battleface, Blue Cross Blue Shield sources, Compare the Market, Go World Travel and other travel or insurance domains.

Again, the observation is not about source quality.

It shows that a market can have a highly durable authority core while still offering a large rotating outer ring of secondary sources.

That outer ring can matter for discovery, PR, sponsorship, content distribution, and competitive research even if it is less durable.

Top-25 Portability Also Increased, but Persistence Increased More

Questions This Section Answers

1. Are the Top-25 AI citation sources also more portable across platforms than the Top-100 sources?
2. Does narrowing an AI citation portfolio to the Top 25 eliminate cross-platform source fragmentation?
3. Why does the Persistence-Portability Gap widen at Top 25 even though portability also rises?

The Top-25 core was not only more persistent. It was usually more portable as well.

233 of 283 markets, 82.33%, had higher portability at Top 25 than Top 100.

The median increase in portability was 5.24 points.

This creates a more nuanced interpretation than “leading sources are platform-specific.”

The leading core is actually more shared across platforms than the broader Top-100 set in most markets.

But temporal persistence rises even more sharply than portability.

Across markets:

• 246 had higher persistence at Top 25.
• 233 had higher portability at Top 25.
• 211 had both higher persistence and higher portability at Top 25.
• 207 had a larger persistence-portability gap at Top 25.
• 75 had a smaller gap at Top 25.
• 1 was unchanged.

So the Top-25 core is, on average, both more durable and more portable, but durability gains faster than cross-platform transfer.

This distinction matters. A CMO should not interpret high Top-25 persistence as evidence that the same publishers dominate every AI platform.

The companion article Persistence vs Portability: Why a Source That Wins on ChatGPT May Not Transfer to Gemini or Google AI shows how large the remaining platform differences can be even inside durable markets.

Platform Pairs Responded Differently to Portfolio Depth

Questions This Section Answers

1. Which AI platform pairs become more similar when analysis narrows from the Top 100 citation sources to the Top 25?
2. Does Google AI Overviews and Google AI Mode overlap increase more at Top 25 than ChatGPT and Copilot overlap?
3. Is the effect of citation portfolio depth consistent across every AI platform pair?

Portfolio depth changed some platform relationships much more than others.

The Google AI Overviews and Google AI Mode relationship is the clearest depth-sensitive platform pair. Their median overlap rises from 53.85% at Top 100 to 66.67% at Top 25.

That suggests their strongest source cores are more similar than their broader source portfolios.

At the opposite extreme, ChatGPT and Microsoft Copilot had a 25.00% median overlap at all three cutoffs. Narrowing the portfolio did not change their median pair relationship at all.

This is another reason a single “platform similarity” number can hide useful structure. Platform relationships can look different depending on whether the analysis focuses on the leading authority core or the broader source environment.

The full platform comparison is covered in Six AI Engines, Six Different Webs.

The Depth Effect Also Changed Across Month Pairs

Questions This Section Answers

1. Was the Top-25 citation-source core more stable than the Top 100 in every July-to-September month comparison?
2. Which month pair showed the largest difference between Top-25 and Top-100 citation persistence?
3. Does portfolio depth change the interpretation of month-to-month AI citation stability?

Every calendar pair showed higher median overlap at Top 25 than Top 100.

The strongest depth effect appeared in July versus August.

The broader Top-100 portfolio had 62.60% median overlap, while the Top-25 core reached 78.57%.

August versus September was already relatively stable at Top 100, so narrowing to Top 25 produced a smaller increase.

This reinforces the conclusion from the separate month-to-month persistence study: citation stability depends both on when the portfolio is measured and how deep into the source ecosystem the measurement goes.

A single month pair at one arbitrary cutoff is therefore an incomplete description of source durability.

The Portfolio-Depth Effect Was Strongest in Some Industries

Questions This Section Answers

1. Which industries showed the largest increase in the Persistence-Portability Gap when analysis narrowed from Top 100 to Top 25 sources?
2. Did legal, finance, health, home, consumer, technology, and travel markets respond differently to source-depth cutoffs?
3. Should CMOs use the same Top-K citation benchmark across every industry?

Every named macro sector had a wider median gap at Top 25 than Top 100, but the magnitude varied.

Legal & Compliance is particularly notable. Its Top-100 median gap was the lowest of the seven named macro sectors at 30.35 points, but its Top-25 gap rose to 43.66 points.

The change does not mean legal sources become universally stable. It means the leading legal source core behaves differently from the broader legal source universe.

At a more detailed sector level, the median Top-25 versus Top-100 gap increase was especially large in:

• Automotive, Mobility, Travel & Transport
• Legal Services
• Medical, Dental & Clinical Services
• Outdoor, Cycling & Fitness
• Mortgage, Business & Specialty Lending
• Cybersecurity, IT & Web Infrastructure

The industry-specific research in There Is No Universal GEO Strategy develops the broader case for market-specific benchmarks.

Top 25 Was Not More Persistent in Every Market

Questions This Section Answers

1. Are there commercial markets where Top-100 AI citation sources are more stable than the Top 25?
2. Which markets are counterexamples to the claim that the most prominent citation sources are always the most durable?
3. Why should CMOs keep a Top-100 view even when the Top-25 source core is usually more persistent?

A strong aggregate pattern should not be converted into a universal rule.

31 of 283 markets had lower persistence at Top 25 than Top 100. Six were unchanged.

The largest reverse examples included:

Natural Skincare Brand illustrates why.

Its August-September Top-25 portfolios shared 18 domains, producing 56.25% Jaccard overlap. The Top-100 portfolios shared 84 domains, producing 72.41% overlap.

The leading ranks rotated between domains such as Acure, 100% Pure, Sephora and other sources on one side, and Forbes, Wikipedia, Pacifica Beauty and other sources on the other.

Deeper in the portfolio, however, the overall source universe remained much more similar.

This can happen when domains remain broadly important but move above or below the Top-25 cutoff between periods. A rank threshold can create apparent churn even when the larger source environment remains stable.

That is exactly why Top-K membership should be paired with response-presence data rather than treated as an absolute on/off indicator of authority.

Home Equity Loans and Certificates of Deposits Show Why Cutoff Sensitivity Matters

Questions This Section Answers

1. Does the direction of the Persistence-Portability Gap ever change between Top 25 and Top 100 source portfolios?
2. Which markets were exceptions to the positive Persistence-Portability Gap at smaller portfolio cutoffs?
3. Can a market move from zero or negative gap at Top 25 to a positive gap at Top 100?

The broad persistence-portability finding survived all three cutoffs, but the exceptions were not identical.

At Top 25:

• 281 markets had positive gaps.
• 1 market had a zero gap.
• 1 market had a negative gap.

At Top 50 and Top 100:

• 282 markets had positive gaps.
• 1 market had a negative gap.

Home Equity Loans was negative at all three depths:

The market becomes less negative as the portfolio broadens, but portability still exceeds persistence at every cutoff.

Certificates of Deposits behaved differently. At Top 25, persistence and portability were both 31.58%, producing a zero gap. At Top 50 the gap became +7.12 points, and at Top 100 it was +7.75 points.

These examples show why the Top-25 result should not be presented as a mathematical law. Portfolio depth can change not only the magnitude of a market's score but, in rare cases, the interpretation of the relationship.

What the Findings Mean for a CMO Dashboard

Questions This Section Answers

1. Should a CMO track Top-25, Top-50, and Top-100 AI citation-source portfolios at the same time?
2. Which AI citation portfolio depth is most useful for publisher prioritization versus market discovery?
3. How can marketing teams distinguish durable core sources from rotating secondary AI citation opportunities?

The evidence argues against choosing one cutoff and discarding the rest.

A practical dashboard can use the three depth levels for different decisions.

Top 25: durable authority core

Use the Top 25 to identify the most prominent source layer.

This is the group most likely to show strong month-to-month persistence and, in many markets, somewhat stronger cross-platform portability.

It is the logical starting point for deeper diligence on:

• publisher authority,
• competitive source presence,
• cited URL types,
• factual consistency,
• PR targets,
• sponsorship opportunities,
• and durable third-party brand coverage.

Top 50: transition layer

Top 50 often captures the point where the durable core begins to blend into a broader source ecosystem.

Its median persistence of 69.49% sits between Top 25 and Top 100, while portability is 26.58%.

This layer can help identify sources that are meaningful but not as entrenched as the leading core.

Top 100: opportunity and turnover layer

Top 100 captures more of the market's source diversity.

It is more useful for:

• finding emerging publishers,
• monitoring category specialists,
• identifying new source entrants,
• tracking competitive distribution,
• and understanding how quickly the broader citation ecosystem changes.

The key is not to treat Top 100 as “worse” because it is less persistent.

The broader list answers a different question.

A durable core plus a rotating tail can be strategically healthy because it gives a company both established authority targets and new opportunities to enter the source ecosystem.

The related Citation Breadth vs Citation Depth article expands the measurement framework beyond Top-K persistence.

What the Findings Mean for Publishers

Questions This Section Answers

1. How can a publisher prove that its AI citation authority belongs to the durable Top-25 source core rather than the broader rotating Top 100?
2. Should publishers sell advertisers on Top-25 AI citation prominence or broader Top-100 category coverage?
3. What evidence should a publisher show before claiming durable AI citation authority?

For publishers, portfolio depth creates a more precise way to describe authority.

A publisher that repeatedly appears in a market's Top 25 has a different evidence story from one that periodically appears in the Top 100.

Neither is automatically more valuable to an advertiser.

But they represent different assets.

A durable Top-25 position can support claims about recurring prominence in the observed citation ecosystem. Broader Top-100 coverage can support claims about category participation or reach across a larger source set.

The responsible publisher evidence package should include:

• the market being measured,
• the AI platform or pooled market context,
• the portfolio depth,
• the observation period,
• response-presence counts,
• persistence across months,
• portability across platforms,
• the cited URLs,
• and clear disclosure that citation is not recommendation.

A publisher should not convert Top-25 recurrence into a claim that sponsored content will be cited or that advertiser brands will be recommended.

The commercial application is explored further in How Publishers Can Turn Persistent AI Citation Authority Into a Sponsorship and Distribution Asset.

Methodology

The underlying study covers completed observations from July, August, and September 2026 across six consumer AI surfaces:

• ChatGPT
• Gemini
• Perplexity
• Microsoft Copilot
• Google AI Overviews
• Google AI Mode

The strict flagship cohort contains 283 qualifying tracked markets, 876,605 retained AI responses, and 8,718,629 citation occurrences resolved to registrable domains.

Within each eligible market-month or market-platform cell, domains were ranked by the number of distinct retained responses in which the registrable domain appeared. Repeated citations to the same domain within one response counted once for ranking. Ties were broken alphabetically.

The baseline cell eligibility rule required:

• at least 100 retained responses;
• at least 100 resolved registrable domains;
• and an actual Top-100 list of exactly 100 domains.

The Top-25 and Top-50 sensitivity analyses retain the same 283 qualifying markets and the same baseline eligible cells. They then truncate those ranked lists to exactly 25 or 50 domains.

This is important because smaller cutoffs do not gain additional thin cells. The comparison isolates portfolio depth rather than mixing depth with sample-eligibility changes.

Persistence compares eligible monthly portfolios using Jaccard intersection divided by union, then takes the median of the eligible month-pair values within each market.

Portability compares eligible platform portfolios using the same method, then takes the median of eligible platform-pair values within each market.

The Persistence-Portability Gap is calculated within each market as persistence minus portability.

The headline medians are unweighted medians across the same 283 markets.

Limitations

Portfolio-depth sensitivity reveals structure, but it does not identify a universally correct cutoff.

Top-K membership is threshold-sensitive

A domain can move from rank 24 to rank 26 with little change in response presence. That can alter Top-25 membership even when its underlying authority changes only modestly.

Persistence does not mean fixed rank

Two portfolios can contain many of the same domains while those domains change rank substantially. The metric measures set overlap, not rank-order stability.

Domain persistence is not URL persistence

A publisher can remain in the portfolio while different pages from that domain are cited over time.

Higher persistence is not automatically better

A highly stable source ecosystem can be concentrated and difficult to enter. A more fluid ecosystem can create new publisher and brand opportunities.

Higher portability is not automatically better

Cross-platform similarity can make a source relevant across several AI surfaces, but the study does not measure media cost, audience quality, editorial fit, traffic, or conversion value.

The study is observational

The historical collection does not establish an identical fixed panel of prompts, prompt versions, model versions, countries, retrieval settings, and collection configurations across every month and platform.

The results describe observed source portfolios rather than controlled causal effects.

Citation is not recommendation

A durable Top-25 source can be repeatedly used as evidence without causing an AI system to recommend the publisher, advertiser, or company discussed in the cited content.

Recommendation authority must be measured separately.

Conclusion

The Top-25 versus Top-100 comparison reveals that AI citation authority has layers.

Across 283 commercial markets, the leading Top-25 source portfolios were substantially more persistent than the broader Top 100.

Median persistence rose from 63.93% at Top 100 to 72.41% at Top 25. More than half of markets gained at least 10 percentage points of persistence, and 23 markets combined a Top-25 persistence rate of at least 75% with Top-100 persistence of 60% or less.

The Top-25 source core was also somewhat more portable across platforms, with median portability rising from 22.70% to 28.21%. But persistence increased more than portability, pushing the median within-market gap from 39.68 points to 46.99 points.

The market examples show what that means in practice.

VPN Services had 100% Top-25 persistence across all three month pairs, while its Top-100 persistence was only 63.93%. Travel Insurance shared 24 of 25 leading domains between July and August but only 76 of its Top 100. Those are clear examples of a durable authority core surrounded by a more fluid source tail.

Natural Skincare Brand shows the opposite can happen. Its Top-100 portfolio was more stable than its Top 25, reminding us that cutoff sensitivity can reflect rank movement around the threshold.

For CMOs, the best measurement approach is therefore not Top 25 or Top 100.

It is Top 25 and Top 100, with each answering a different question.

The Top 25 reveals the durable authority core.

The Top 100 reveals the broader competitive and publisher opportunity set.

Together, they provide a much more useful map of AI citation authority than a single undifferentiated visibility score.

Related LLM Authority Index Research

• The Persistence-Portability Gap at Scale: 282 of 283 Markets Show Higher AI Citation Persistence Than Cross-Platform Portability Across 8.72 Million Citation Occurrences is the parent 283-market benchmark.
• How Stable Are AI Citation Sources Month to Month? A CMO Guide to Citation Persistence examines temporal source durability across the full Top-100 benchmark.
• Persistence vs Portability: Why a Source That Wins on ChatGPT May Not Transfer to Gemini or Google AI examines cross-platform transfer in detail.
• How Publishers Can Turn Persistent AI Citation Authority Into a Sponsorship and Distribution Asset focuses on publisher commercialization.
• Citation Breadth vs Citation Depth: Which AI Authority Metric Should CMOs Care About? expands the source-depth measurement framework.
• The 2026 AI Commercial Citation Authority Index examines breadth and depth across markets and platforms.
• Six AI Engines, Six Different Webs provides the broader six-platform overlap matrix.
• There Is No Universal GEO Strategy examines industry-level variation.
• AI Citation Authority vs AI Recommendation Authority separates source authority from company recommendation behavior.

About LLM Authority Index

LLM Authority Index measures how AI systems cite sources, surface publishers, and structure authority across commercial markets. Its research separates citation persistence, platform portability, market breadth, platform breadth, source depth, citation authority, and recommendation authority so that a recurring source is not mistaken for a universally influential one.