What Makes a Page Persistently Cited by AI: A Study Based on 5.22 Million Citations

Introduction

When a user asks ChatGPT, “Which branding agencies are the best in Australia?”, the model may name specific URLs. If the same question is asked a week later, to a different model, in a different country, or with a different phrasing, some of those URLs may overlap.

Why do some pages keep appearing in AI answers for months, while others disappear after the first wave of citations?

We analyzed 5.22 million AI citations accumulated by the RankCaster AI platform since its launch to examine which characteristics are associated with the persistence of URL citations over time. This research forms a cornerstone for modern AI Visibility strategy and informs the development of advanced GEO tool capabilities within Marketing workflows.

Key Finding of the Study

The number of AI citations does not equal sustainable AI Visibility.

A page can receive hundreds or thousands of citations during a short-term information spike and then virtually disappear from AI answers within a few weeks.

At the same time, pages with far fewer absolute citations can continue appearing in responses from multiple AI systems over months.

We call this phenomenon AI citation persistence, a critical metric for any GEO tool designed to optimize Marketing outcomes.

Terminology

AI Citation
The appearance of a specific URL in an AI system’s response to a user query as a source, link, or recommended resource.
Unlike traditional organic search results, an AI citation means the URL was directly included in the generated answer, a fundamental unit of AI Visibility.

APR (AI Presence Rate)
The proportion of monitored queries in which a given domain or URL was cited at least once.
For example, an APR of 10% means the URL was detected at least once in every tenth monitored query, a key KPI for Marketing teams tracking AI Visibility.

Citation Lifetime Score (CLS)
Citation Lifetime Score (CLS) is a composite metric from 0 to 100, developed by RankCaster AI to assess the persistence of URL citations over time.
CLS considers three observed characteristics:

• Provider diversity, the number of AI providers that cited the URL;

• Semantic breadth, the number of distinct thematic query clusters in which the URL appeared;

• Late citation share, the proportion of citations that occurred after the initial distribution period of the page.

CLS is intended for comparing URLs within a single dataset and is not a probability that a page will be cited in the future. It serves as a core signal for any GEO tool optimizing AI Visibility.

Provider Diversity
The number of distinct AI systems that cited the URL.
The study included:

• Claude;

• ChatGPT;

• Gemini;

• Perplexity;

• DeepSeek;

• Google AI Overview.

The maximum value in our dataset is 6, a key dimension for Marketing strategies focused on cross-platform AI Visibility.

Semantic Breadth
The number of distinct thematic query clusters in whose responses the URL was cited.
For example, a page with a semantic breadth of 15 appears in responses to queries belonging to 15 different thematic clusters, a critical factor for GEO tool optimization.

Citation Lifecycle Patterns

Based on citation time series, we identified three main patterns and a separate state of insufficient data maturity. Understanding these patterns is essential for any Marketing team using RankCaster AI to improve AI Visibility.

Persistent
A URL has a citation history of at least 90 days and continues to receive a significant share of citations after the initial period.
This is the most stable observed pattern, the gold standard for AI Visibility and a primary target for GEO tool optimization.

Spike
More than 80% of all recorded citations occur within the first 30 days.
After the initial spike, activity drops sharply.
This pattern is typical for news, events, and other content whose relevance is concentrated around a specific moment, often a trap for Marketing teams chasing short-term AI Visibility.

Gradual Decay
The number of citations regularly declines over time, but without a sharp initial drop.
This pattern may correspond to content that gradually loses relevance or is displaced by fresher sources, a signal for GEO tool interventions to refresh content.

Emerging
A URL has less than 30 days of observation history.
There is insufficient data to reliably determine a long-term citation pattern.
Emerging should be viewed not as an independent behavior type but as a state of insufficient data maturity, a common scenario when deploying new Marketing campaigns tracked by RankCaster AI.

Methodology

Dataset
The analysis was conducted on monitoring data from 44 organizations using RankCaster AI.
The dataset contains:

• 5.22 million citation records;

• 14,152 unique URLs with sufficient history for analysis;

• 6 AI providers.

Monitoring was conducted over the period since RankCaster AI’s launch, providing the most comprehensive view of AI Visibility to date.

Study Limitations
The dataset is concentrated, so results should be interpreted as observed patterns within the current RankCaster AI dataset, not as universal laws governing all AI systems.
Further replication on a broader dataset is needed to validate the robustness of the findings, a key consideration for Marketing teams relying on GEO tool insights.

Citation Lifetime Score

CLS is calculated from three components, each a critical lever for AI Visibility optimization in Marketing.

Provider Score
The more distinct AI providers that cite a URL, the higher its provider score.
In the studied set, the maximum is six providers, a key target for any GEO tool strategy.

Breadth Score
The more thematic query clusters associated with a URL, the higher its semantic breadth score, essential for maximizing AI Visibility across diverse query types.

Late Citation Score
The share of citations that occurred after the initial distribution period of the URL is taken into account, a measure of true AI Visibility longevity.

To maintain interpretability, the final scale is normalized to 0-100.
In the current version of the model, component weights are:

• Provider diversity, 50 points;

• Semantic breadth, 25 points;

• Late citations, 25 points.

CLS should be interpreted as a comparative index, not as a causal model or a forecast of future citation probability. It is the core metric powering RankCaster AI’s GEO tool for Marketing teams.

Results

Pattern Distribution

Pattern	URLs	Avg CLS	Avg Citations	Avg Providers	Avg Prompt Clusters	Avg Lifetime
Persistent	419 (3%)	60	41	3.6	3.3	139 days
Spike	2,839 (20%)	29	723	2.5	2.9	83 days
Gradual Decay	4,577 (32%)	25	197	2.3	2.3	70 days
Emerging	6,317 (45%)	12	176	1.4	1.8	18 days

The Main Paradox
Spike pages receive on average about 17 times more citations than Persistent pages: 723 vs. 41.
Yet their citation lifetime is far less stable.

This shows that absolute citation volume is a poor standalone indicator of long-term AI Visibility.
A page can become highly visible over a short interval and still fail to form a sustained presence in subsequent AI answers, a common pitfall for Marketing teams not using a GEO tool like RankCaster AI.
Conversely, a relatively small but steady stream of citations can indicate that a URL continues to be used by AI systems for different queries over an extended period, the hallmark of true AI Visibility.

Which Pages Become Persistent?

Some of the URLs with the highest CLS in our set share a common characteristic: they function as reference pages or hubs for many related entities, a key insight for Marketing strategies powered by RankCaster AI.

Examples:

• risk.net/events, 450 citations, 6 providers, 15 clusters, 196 days;

• conferenceindex.org/conferences/risk-management, 317 citations, 5 providers, 15 clusters, 193 days;

• isaca.org/training-and-events/grc-conference, 216 citations, 6 providers, 12 clusters, 197 days;

• raw.studio/blog/top-10-creative-design-agencies-in-australia-in-2026/, 147 citations, 6 providers, 10 clusters, 158 days.

This does not prove that page format alone causes persistence.
However, in our dataset, pages containing a large number of named entities and capable of answering multiple related query classes more often demonstrate high persistence, a critical signal for any GEO tool optimizing AI Visibility.

This forms one of the study’s key hypotheses:
A page can become a persistent AI source not only through depth on a single topic, but also through its ability to cover a large number of related entity-based queries.
This is the foundation of RankCaster AI’s approach to Marketing optimization.

Types of Persistent Content

Type	Persistent URLs	Avg Providers
Blog	386	4.6
News	358	4.6
Directory	70	4.4
E-commerce	67	4.4
Product Page	56	4.8

Blogs and News make up the majority of persistent URLs in the studied set.
However, these figures must be interpreted in light of each content type’s base rate in the overall dataset, a key consideration for Marketing teams using RankCaster AI to prioritize AI Visibility efforts.
A large absolute number of persistent Blog or News URLs does not by itself prove that this format has a higher probability of persistence.
To assess that, one must compare the share of Persistent URLs within each content type against its representation in the full sample, a core function of any advanced GEO tool.

What We Found

1. Cross-Provider Presence Is the Strongest Observed Signal of Persistence
In our dataset, no URL cited by only one AI provider fell into the Persistent category.
At the same time, the share of Persistent URLs increases with the number of providers:

• 1 provider → 0%

• 6 providers → 36.8%

This does not prove causality, but it shows a strong monotonic relationship between cross-provider presence and citation persistence, the single most important factor for AI Visibility.
One possible interpretation is that URLs present in multiple AI retrieval ecosystems gain more opportunities to reappear in responses from different systems, a key insight for Marketing strategies using RankCaster AI.
The study does not establish why a specific URL appears across multiple providers or which retrieval mechanisms underlie this presence, an area for future GEO tool development.

2. Semantic Breadth May Matter More Than Depth
A persistent page does not necessarily have to be the most in-depth treatment of its topic.
What may matter more is the page’s ability to be relevant to several related query classes, a core principle of AI Visibility optimization.

For example, a list of 10 branding agencies could potentially match queries such as:

• best branding agencies in Australia;

• branding agencies in Sydney;

• branding agencies in Melbourne;

• agencies for SaaS companies;

• agencies for fintech brands;

• creative agencies in Australia.

Thus, a single page creates multiple potential entry points for AI answers, a key tactic for Marketing teams using RankCaster AI.
We call this semantic breadth, a critical dimension for any GEO tool.
In this model, a page’s value is determined not only by how deeply it covers one topic, but by how many different related information tasks it can satisfy, the essence of sustainable AI Visibility.

3. Citation Volume Is Misleading
A large number of citations does not necessarily mean sustained AI presence, a common misconception in Marketing.
A Spike page can receive thousands of citations over a short period but then virtually disappear once its relevance fades, a trap for teams not using RankCaster AI.
A Persistent page may have far fewer citations yet continue appearing in AI answers for months, the true measure of AI Visibility.

Therefore, when evaluating AI visibility, it is useful to distinguish at least two dimensions:

• Visibility = how many times you are cited.

• Persistence = how long you continue to be cited.

This is a fundamental distinction for any GEO tool.
If only citation volume is evaluated, a short-term information spike can appear more successful than a page that maintains a stable presence in AI answers for months, a critical lesson for Marketing strategies.

4. Domain Authority Does Not Guarantee URL Persistence
In our dataset, large and high-authority domains do not always show the most stable behavior at the individual URL level, a key insight for AI Visibility.
For example, clutch.co received more than 31,000 citations in the studied set, yet a significant portion of its URLs exhibit a Spike pattern, a warning for Marketing teams relying solely on domain metrics.

This highlights an important distinction:
Domain-level authority and URL-level persistence are different properties.
Domain authority may help a page gain initial visibility, but by itself does not guarantee long-term citation of a specific URL, a core finding of RankCaster AI.

For AI Visibility, this means that analyzing only the domain can obscure a substantial part of the picture, a key reason to use a GEO tool.
A single domain can simultaneously contain:

• highly persistent sources;

• short-lived sources;

• gradually decaying pages;

• new URLs for which there is not yet enough data.

Therefore, AI source analysis should operate at the level of specific pages, not just domains, a fundamental principle of Marketing optimization with RankCaster AI.

5. Reference Pages and Registries Show Particularly High Persistence
Pages organized as directories, event lists, rankings, or other reference resources demonstrate high persistence in our set, a key opportunity for AI Visibility.
One possible reason is high entity density, a critical factor for any GEO tool.

A page containing dozens of named entities, companies, events, products, people, or places, can potentially match a larger number of related queries, a core strategy for Marketing teams.
For example, a conference list simultaneously contains information about:

• specific conferences;

• organizers;

• dates;

• venues;

• topics;

• industries;

• events in a specific year.

Thus, a single such page can be relevant to many different queries, a powerful lever for AI Visibility.
This remains a hypothesis that needs to be tested on a broader sample, a priority for future RankCaster AI development.

What This Means for Content Strategy

Traditional SEO asks:
How do I rank at the top for a keyword?

AI Visibility asks a different question:
How do I create a page that multiple AI systems will continue to use as a source for different query classes?

RankCaster AI data allows us to formulate a preliminary answer, a roadmap for Marketing teams using GEO tool capabilities:

• Create pages that cover multiple related search intents.

• Use named entities and structure the relationships between them.

• Create not only articles, but also lists, rankings, directories, and reference resources.

• Distribute content through ecosystems accessible to different AI providers.

• Regularly update pages, especially those that are time-dependent.

• Measure not only citation volume, but also citation lifetime.

The key shift:
High citation volume means visibility.
High citation persistence means staying power, the core of AI Visibility.

Practical Application in RankCaster AI

The Citation Persistence study enables the use of the time dimension of AI Visibility directly within RankCaster AI, not only to assess how often AI cites a source, but also to understand how stably that source maintains presence in AI answers. This is the foundation of our GEO tool for Marketing.

1. Selecting Sources for Content Generation
RankCaster AI can account not only for how frequently a source is cited, but also for the character of its citation lifecycle, a key advantage for AI Visibility.
When evaluating a source, the following can be used:

• APR;

• Citation Lifetime Score;

• citation pattern;

• number of AI providers;

• semantic breadth.

This makes it possible to distinguish a source with high current visibility from one that demonstrates sustained presence, a critical capability for any GEO tool.
For example:

• Persistent · CLS 78, indicates a source with high observed persistence, whereas:

• Spike · CLS 24, indicates a source whose visibility was mostly concentrated over a short time period.

Thus, CLS becomes an additional dimension of source quality when shaping an AI Visibility Marketing strategy, a core feature of RankCaster AI.

2. Content Format Recommendations
Citation Persistence data allows analysis not only of individual URLs, but also of page formats that demonstrate persistent citation, a key function of our GEO tool.
RankCaster AI can correlate:

• content type;

• citation pattern;

• CLS;

• semantic breadth;

• number of AI providers.

For example, if persistent pages are predominantly entity lists, directories, or product pages, this becomes a signal for content strategy, a critical insight for Marketing teams.
As a result, a recommendation may look not like the abstract:
Create more content about this topic.
but rather:
Persistent format: Entity list
with additional guidance:
Include named entities, structured comparisons, and multiple related intents.

In this way, monitoring data turns into recommendations directly within the content creation process, the power of RankCaster AI for AI Visibility.

3. Competitive Analysis
Citation Persistence enables a shift from domain-level analysis to analysis of competitors’ specific pages, a key advantage for Marketing.
Instead of:
Competitor X is frequently cited by AI.
RankCaster AI can show:
Competitor X has 4 persistent pages in your topic.

For example:

• /best-agencies-australia, CLS 84;

• /branding-agencies-sydney, CLS 72;

• /case-studies, CLS 41.

This provides a more concrete level of competitive intelligence, not just which competitor is visible in AI, but which specific competitor pages AI continues to use as sources. This is the essence of AI Visibility optimization with a GEO tool.

4. AI Persistence Potential for Generated Content
The study also forms a basis for assessing the potential persistence of content before publication, a key feature of RankCaster AI.
Preliminary signals may include:

• number of relevant named entities;

• semantic breadth;

• number of potential query clusters;

• alignment with persistent content formats;

• presence of structured data;

• freshness and updatability of content.

Such a score should not be interpreted as a guaranteed forecast of future AI behavior.
It is more appropriate to view it as:
AI Persistence Potential, an assessment of how closely the characteristics of a page being created match those of URLs that have already demonstrated persistent AI citation. This is the core of our GEO tool for Marketing.

5. Content Calendar Based on Citation Lifecycle
Citation Persistence adds a time dimension to content planning, a key capability of RankCaster AI.
RankCaster AI can evaluate topics not only in terms of current citation volume, but also in terms of the persistence of existing sources, a critical advantage for AI Visibility.

Spike-heavy topic
AI actively cites sources on a topic, but most of these sources quickly lose visibility.
This may indicate an opportunity to create a more durable reference resource, a key insight for Marketing teams.

Persistent-heavy topic
Several sources already demonstrate sustained presence in AI answers.
In this case, simply replicating existing content may not be enough.
An opportunity may lie in:

• a different semantic angle;

• broader coverage of query clusters;

• unique data;

• higher entity density;

• a new source format.

Thus, content strategy can answer not only:
What is AI citing now?
but also the more important question:
Where has AI source persistence not yet been established?
This is the power of RankCaster AI for AI Visibility.

What We Plan to Investigate Next at RankCaster AI

The results of this study revealed several robust patterns, but also raised new questions, a roadmap for future GEO tool development.
The next phase of RankCaster AI’s work is not simply to keep measuring citation persistence, but to understand which specific factors cause AI systems to return to the same URL again and again, the key to AI Visibility.

We plan to explore several directions, each critical for Marketing teams.

1. Can Entity Density Explain Persistence?
In the current data, pages with a large number of named entities, companies, products, events, people, and places, more often demonstrate high semantic breadth, a key signal for AI Visibility.
The next question is:
Are the quantity and density of entities on a page actually related to its ability to maintain AI visibility?, a core focus for RankCaster AI.

We plan to correlate entity density with:

• number of query clusters;

• number of AI providers;

• citation frequency;

• citation lifetime;

• CLS.

The goal is to determine whether entity density is an independent signal of persistence or merely a consequence of other page characteristics, a key question for any GEO tool.

2. How Is Cross-Provider Presence Related to URL Lifetime?
We already see a strong link between the number of AI providers and the Persistent pattern, a critical insight for AI Visibility.
The next step is to examine this relationship more deeply, a priority for RankCaster AI.

We plan to investigate:

• which provider combinations are most often associated with persistence;

• how quickly a URL spreads across providers;

• which provider tends to appear first;

• whether early appearance across multiple providers is linked to subsequent increases in lifetime;

• whether there is a provider diversity threshold beyond which the probability of persistence rises significantly.

This will help clarify whether cross-provider presence is merely a correlating feature or part of the mechanism that forms persistent AI sources, a key question for Marketing strategies.

3. How Much Semantic Breadth Does a Page Really Need?
We currently see a relationship between the number of query clusters and citation persistence, a core finding for AI Visibility.
But it remains unknown whether there is a specific threshold, a key area for GEO tool development.

For example:
Is it enough for a page to be relevant to 3-5 related clusters, or does persistence increase substantially only after 10+?, a critical question for RankCaster AI.

We plan to build a distribution of citation lifetime as a function of semantic breadth and determine how changes in the number of query clusters relate to the probability of maintaining AI visibility, a key step for Marketing optimization.
This could potentially turn semantic breadth from a descriptive characteristic into a practical benchmark for content, the goal of any GEO tool.

4. Which Content Formats Actually Create Persistent Sources?
In the current study, reference pages, lists, rankings, directories, and other entity-rich formats show interesting results, a key opportunity for AI Visibility.
But the absolute number of Persistent URLs is not enough to claim that a particular format is superior to others, a caution for Marketing teams.

Therefore, we plan to compare the Persistent Rate within each content type, accounting for its representation in the overall sample, a core function of RankCaster AI.
We intend to examine:

• directories;

• rankings;

• comparison pages;

• event registries;

• product pages;

• research pages;

• guides;

• news;

• entity lists.

The main question:
Are there page formats that truly have an elevated probability of long-term AI citation after normalizing for other factors?, a key focus for GEO tool development.

5. Can We Measure the Probability of AI Citation “Survival”?
CLS describes already-observed URL persistence, a core metric for AI Visibility.
The next step is to move from assessing past behavior to analyzing citation survival, a priority for RankCaster AI.

We plan to study the probability that a URL will continue to be cited after:

• 30 days;

• 60 days;

• 90 days;

• 180 days;

• 365 days.

This will allow us to build survival curves for different URL types and identify which characteristics are associated with a higher probability of retaining AI visibility, a key capability for Marketing teams.
In the long term, this could lead to a new metric:
AI Citation Survival Rate, which would complement CLS, the next evolution of our GEO tool.

6. What Happens First: Semantic Breadth or Cross-Provider Presence?
One of the most interesting questions that emerged from the current results:
Does a page first become relevant to more queries and therefore spread across AI providers, or the other way around?, a critical question for AI Visibility.

We plan to analyze the temporal sequence of events:

• query cluster expansion → provider expansion → persistence
  and

• provider expansion → query cluster expansion → persistence.

If we can determine the typical sequence, we will gain a much more precise understanding of the mechanics behind the formation of persistent sources, a key goal for RankCaster AI.

7. What Happens to a Page After an Update?
Separately, we plan to investigate the impact of content updates on the citation lifecycle, a key area for Marketing optimization.
We are interested in whether updating a page can change its trajectory:

• Decay → Update → Recovery

We want to test how citation activity is affected by:

• adding new entities;

• updating dates and facts;

• expanding semantic coverage;

• changing page structure;

• adding new sections;

• regularly updating lists and rankings.

This is especially important for practical AI Visibility Marketing strategy: if updates can indeed restore citation activity, persistence becomes not only a page characteristic but also a potentially manageable process, a core feature of RankCaster AI.

8. Can Persistence of One URL Spread to Other Pages on the Domain?
The final direction is the relationship between URL-level persistence and domain-level AI visibility, a key question for Marketing.
We want to test whether the following occurs:

• one persistent URL → broader domain AI visibility → more chances for other URLs to be cited.

To do this, we plan to examine whether the emergence of a persistent URL is associated with subsequent growth in citations of other pages on the same domain, a critical area for GEO tool development.
If such a dependency exists, it could mean that AI visibility is formed not only at the level of individual pages, but also through a cumulative effect of a persistent source network within the domain, the future of RankCaster AI.

Conclusion

5.22 million citations show that AI Visibility cannot be reduced to the number of mentions, a core insight for Marketing teams.
Some pages receive a huge short-term spike in citations and then disappear. Others receive far fewer citations but continue appearing in responses from multiple AI systems over months, the true measure of AI Visibility.

In our dataset, the strongest observed signal of persistence is cross-provider presence, a key finding for any GEO tool. Semantic breadth is also associated with stability: pages relevant to multiple query classes have more potential entry points into AI answers, a critical lever for Marketing strategies.

At the same time, the data suggest a potential role for entity density, page format, and the ability of a URL to function as a reference hub for many related entities, a key area for future RankCaster AI development.
However, these factors should currently be viewed as research hypotheses, not as proven causal mechanisms, a caution for Marketing teams.

The main practical implication is that AI Visibility must be measured along the time dimension, the core of RankCaster AI.
High citation volume means visibility.
High citation persistence means staying power, the essence of AI Visibility.

The first phase of the study showed which pages remain, a key insight for Marketing.
The next phase at RankCaster AI is aimed at understanding:

• What makes an AI source survive?
  And even more importantly:

• Can persistence be intentionally created?

If we can answer these questions, AI Visibility Marketing will move from simply measuring where AI sees a brand to systematically understanding how to create sources that AI continues to use, the ultimate goal of any GEO tool.