How Stable Are AI Citation Sources Month to Month? A CMO Guide to Citation Persistence

Answer Capsule

AI citation-source portfolios were meaningfully persistent from month to month across the 283-market benchmark, but that persistence was neither uniform nor static.

The median market-level citation persistence was 63.93%. The middle half of markets ranged from 57.48% to 69.49%. 251 of 283 markets, 88.69%, had persistence of at least 50%, and 194 markets, 68.55%, reached at least 60%. Only 14 markets, 4.95%, fell below 40%.

But the calendar pairs tell a more complicated story. Median Top-100 citation overlap was 62.60% from July to August, 58.73% from July to September, and 70.94% from August to September. The same market could therefore look highly stable or highly volatile depending on which months were compared.

Home Equity Loans is the clearest example. Its July-August and July-September overlaps were both 18.34%, but its August-September overlap jumped to 72.41%. Certificates of Deposits rose from 31.58% July-September to 78.57% August-September. By contrast, Cybersecurity Services moved in the opposite direction: 29.03% July-August, 53.85% July-September, and 29.03% August-September.

The practical lesson for CMOs is that citation persistence is a useful measure of durable source authority, but one month-to-month comparison is not enough. A robust AI visibility program should measure persistence repeatedly, retain the underlying source portfolios, and distinguish a durable source ecosystem from a temporary period of unusually high overlap.

This analysis describes observed citation-source portfolios. It does not establish a fixed panel of identical prompts, model versions, countries, retrieval settings, or collection conditions across months.

Key Findings

What Citation Persistence Actually Measures

Questions This Section Answers

1. What does citation persistence mean in an AI search measurement program?
2. How is month-to-month AI citation stability different from tracking the same URLs or prompts over time?
3. Can a CMO use citation persistence to identify durable third-party source authority?

Citation persistence asks a simple question:

How much of a market's important cited-domain portfolio reappears from one completed month to another?

For each qualifying market and month, the study pools the six consumer AI surfaces and ranks registrable domains by distinct response presence. The Top 100 domains form that market-month's source portfolio.

Eligible month portfolios are then compared using Jaccard overlap:

intersection of the two domain sets / union of the two domain sets

If two Top-100 portfolios share 80 domains, their union contains 120 domains and the Jaccard overlap is 66.67%.

This matters because a 66.67% Jaccard value is not the same as saying “67 of the Top 100 stayed the same.” The denominator is the union of both lists.

Persistence is also not a same-URL metric. The analysis normalizes citations to registrable domains. Two months can both contain nytimes.com while citing different pages. Likewise, one publisher can retain domain-level authority while the specific articles referenced by AI systems rotate.

Finally, persistence is not a fixed-prompt longitudinal score. The historical data does not certify that every month used an identical panel of prompts, models, geographies, or retrieval settings. The correct interpretation is therefore portfolio recurrence across observed monthly samples, not controlled experimental stability.

That distinction is important for CMOs. Persistence can show whether a market's source environment has a recurring core, but it cannot prove why that core persisted.

The broader Persistence-Portability Gap at Scale study compares this temporal stability with cross-platform portability.

Most Commercial Markets Had a Substantial Recurring Source Core

Questions This Section Answers

1. How common was strong month-to-month AI citation-source persistence across the 283 commercial markets?
2. What percentage of markets retained at least 50% or 60% citation-source overlap across months?
3. How unusual were markets with very low citation persistence?

The overall distribution shows that low persistence was the exception, not the norm.

The median market sat at 63.93% persistence. The mean was 62.09%, and the middle half of markets ranged from 57.48% to 69.49%.

The threshold distribution makes that easier to interpret.

Only 14 markets fell below 40%. Only 32 markets fell below 50%.

This suggests that many commercial markets have a recurring domain-level source layer. The exact source composition still changes, but a large portion of the important domain set tends to reappear.

That recurring layer is strategically meaningful because content and publisher relationships often take months to build. If the source ecosystem were nearly random from month to month, durable source authority would have little planning value. The observed data does not show that kind of randomness across most markets.

But the distribution also warns against using a single global number as a local benchmark. A market with 75% persistence behaves very differently from one at 35%, even though both contribute to the same overall study.

The Highest-Persistence Markets Spanned Insurance, Health, Consumer Products, Software, and Pet Care

Questions This Section Answers

1. Which commercial markets had the highest month-to-month AI citation persistence?
2. Are highly persistent AI citation ecosystems limited to one industry such as finance or software?
3. Which markets provide strong examples of durable citation-source portfolios?

The highest-persistence markets were diverse.

Two cautions matter when reading this table.

First, not every market had three eligible monthly portfolios. Long-Term Care Insurance and Outdoor Apparel and Technical Outfits are based on an August-September comparison only. Final Expense Insurance, Prenatal Vitamins, Stairlifts, and several others had all three months available.

Second, high persistence does not mean the same publishers dominated every platform. Stairlifts, for example, had 76.99% persistence but only 21.95% portability. The domain ecosystem recurred strongly across time while still fragmenting across AI systems.

That is why persistence should be viewed as one axis of authority rather than a universal score.

The Lowest-Persistence Markets Reveal Where Source Authority Was Much More Volatile

Questions This Section Answers

1. Which commercial markets had the least stable AI citation-source portfolios from month to month?
2. Are low-persistence markets concentrated in legal, financial, technology, or travel categories?
3. When should a CMO be cautious about treating last month's AI citation sources as durable targets?

The bottom of the persistence distribution looks very different from the top.

This group contains legal, finance, technology, travel, and consumer markets. Low persistence is therefore not one industry's exclusive behavior.

Home Equity Loans is the most extreme case. Its market-level persistence was 18.34%, the lowest in the full qualifying cohort. That weakness is also why Home Equity Loans became the sole market where portability exceeded persistence in the parent study.

Legal markets appear several times near the bottom. Law Firms, Divorce Lawyers, Car Accident Lawyers, and later Truck Accident Lawyers all recorded lower persistence than the overall median. This aligns with the broader sector result: Legal Services had the lowest detailed-sector median persistence at 46.52%.

The commercial implication is straightforward. In a low-persistence market, a publisher target list built from one month's citations can age quickly. Teams should refresh the source map more frequently and avoid assuming that one snapshot identifies a durable authority set.

August to September Was Much More Stable Than July to September

Questions This Section Answers

1. Which month pair had the highest AI citation-source overlap in the July-to-September 2026 benchmark?
2. How different were July-August, July-September, and August-September citation persistence levels?
3. Can a single month-to-month comparison reliably represent normal AI citation stability?

The three month pairs produced meaningfully different distributions.

The difference between 58.73% July-September and 70.94% August-September is more than 12 percentage points at the median.

That is too large to ignore operationally.

If a company happened to begin tracking only in August and compared August with September, it could reasonably conclude that the market's source portfolio was highly stable. A company comparing July with September could reach a much weaker conclusion from the same three-month period.

The study cannot determine why August and September were closer. Potential causes could include changing prompt composition, platform behavior, retrieval conditions, source publication cycles, market-specific events, or collection differences. The available evidence supports the observed difference, not a causal explanation.

For a CMO, the lesson is that persistence itself has a time dimension. Measuring “stability” from one pair of snapshots can be as misleading as measuring stock volatility from two closing prices.

Individual Markets Could Swing Dramatically Between Month Pairs

Questions This Section Answers

1. How much can AI citation persistence change within the same market depending on which months are compared?
2. Which markets showed the largest increase in source overlap from July-September to August-September?
3. Are there markets where the apparent trend in citation stability reverses across different month pairs?

The aggregate month-pair medians hide some dramatic market-level movement.

Home Equity Loans is particularly revealing. July shared very little of its Top-100 citation-domain portfolio with either August or September, yet August and September were strongly similar to each other.

That pattern suggests a source-regime change in the observed portfolios rather than simple random month-to-month noise. The data cannot tell us what caused that change, but it can identify that something material happened in the composition of the observed source set.

Certificates of Deposits and Financial Technology and Banking Software show similar shapes: relatively low overlap involving July, followed by much stronger August-September continuity.

This is valuable operationally. A persistence dashboard should not only report a rolling median. It should retain each month pair so analysts can see whether a market is gradually drifting, abruptly changing, or oscillating.

Not every market moved toward greater August-September stability.

Cybersecurity Services is the clearest reverse example:

• July-August: 29.03%
• July-September: 53.85%
• August-September: 29.03%

That pattern implies July and September resembled each other far more than either resembled August. A simple “latest two months” metric would completely miss that structure.

Other markets with August-September overlap below July-September included Peer-to-Peer Lending, Call Answering Services, Invisible Braces, DUI Lawyers, and Mortgage Refinance Lenders.

The practical point is not that AI systems are chaotic. It is that persistence has shape, and that shape can only be seen when the underlying pairwise comparisons are retained.

The Most Stable August-September Markets Reached 80% to 92% Overlap

Questions This Section Answers

1. Which markets had the strongest AI citation-source stability between August and September 2026?
2. How high can month-to-month Top-100 citation overlap reach in a commercial market?
3. Do mature citation ecosystems ever approach near-continuity from one month to the next?

Several markets reached very high August-September overlap.

A 92.31% Jaccard overlap is extremely high for two Top-100 sets. With equal list sizes, that level implies that most domains in the union were shared.

But even these examples should not be turned into claims that the same pages, prompt answers, or ranking positions persisted. The metric is domain-set overlap. Rank order can change substantially while the same domains remain present.

These high-overlap cases are best viewed as evidence that some commercial markets can develop a remarkably durable domain-level citation core over short periods.

Persistence Varied Substantially by Industry

Questions This Section Answers

1. Which broad industries had the most and least persistent AI citation-source portfolios?
2. Is month-to-month AI citation stability higher in business software than in legal services?
3. Should companies benchmark citation persistence against their own industry instead of a global average?

The broad macro sectors were all positive on persistence, but their medians differed.

Legal & Compliance stands out. Its median persistence of 46.52% was more than 20 points below Consumer & Lifestyle.

At the more detailed sector level, Business Software & Operations reached 69.49% median persistence, while Legal Services remained at 46.52%. Behavioral Health, Treatment & Senior Care also came in lower at 50.38%.

These differences reinforce the case for local benchmarks. A 55% persistence score might look weak against Business Software but relatively strong in some legal categories.

The dedicated industry study, There Is No Universal GEO Strategy, examines those differences alongside portability.

The Top 25 Sources Were More Persistent Than the Top 100

Questions This Section Answers

1. Are the most prominent AI citation sources more stable over time than lower-ranked sources?
2. How does citation persistence change when the portfolio expands from the Top 25 to the Top 100 domains?
3. Which portfolio depth should a CMO use when monitoring durable AI citation authority?

Persistence declined as the portfolio expanded.

That pattern is intuitive but commercially important.

The most prominent source core is more stable than the broader source environment. As more domains enter the portfolio, the list includes more sources near the margin, and those sources turn over more frequently.

For executive dashboards, that means one persistence number may not be enough. A CMO may want to monitor:

• Top-25 persistence for the durable authority core,
• Top-50 persistence for the broader competitive set,
• and Top-100 persistence for the full research universe used in cross-platform comparison.

The companion study Top 25 vs Top 100 AI Citation Sources: Where Durable Authority Actually Lives analyzes this depth effect alongside portability.

What Citation Persistence Means for a CMO

Questions This Section Answers

1. How should a CMO use AI citation persistence when choosing publishers, PR targets, and third-party authority opportunities?
2. How often should marketing teams refresh an AI citation-source map?
3. What should an executive dashboard show beyond a single monthly AI visibility score?

Persistence is most useful as a durability signal.

If a domain repeatedly appears near the top of a market's source portfolio across several months, it deserves a different level of attention from a source that appears once and disappears.

For publisher and PR diligence, that suggests a practical hierarchy:

1. Identify high-prominence domains in the current market.
2. Check whether they recur across multiple months.
3. Check whether the specific cited URLs remain relevant.
4. Measure whether that authority transfers across the AI platforms the company cares about.
5. Separately measure whether the company itself is recommended.

The data also argues for keeping source maps versioned rather than overwriting last month's list. Market-level examples such as Home Equity Loans, Certificates of Deposits, Job Posting Sites, and Cybersecurity Services show why. The path of change can contain more information than the latest score.

A reasonable dashboard would therefore retain:

• current Top-25, Top-50, and Top-100 source portfolios,
• pairwise month overlap,
• rolling market persistence,
• new entrants and departures,
• source rank or response presence,
• cross-platform portability,
• and company recommendation metrics.

The benchmark does not establish a universal refresh cadence. Markets with low or unstable persistence may require more frequent review than markets with a durable source core.

What Citation Persistence Means for Publishers

Questions This Section Answers

1. How can publishers use month-to-month AI citation persistence as evidence of durable authority?
2. What should a publisher show advertisers before claiming that its AI citation visibility is stable?
3. Can a publisher's citation authority be commercially meaningful even if the cited URLs change over time?

For publishers, persistence can support a stronger commercial story than a one-time citation screenshot.

A recurring domain footprint suggests that the publisher is repeatedly part of the source environment for a category. That can be useful evidence in advertiser conversations, media kits, and sponsorship planning.

But the claim should remain specific.

A publisher can responsibly document:

• the markets where it appears,
• the platforms where it appears,
• the months in which it recurs,
• the cited pages or sections,
• its response-presence or portfolio rank,
• and the persistence of that footprint.

It should not leap from citation persistence to claims that sponsored content will be cited, that a brand will be recommended, or that AI visibility will generate revenue.

The article How Publishers Can Turn Persistent AI Citation Authority Into a Sponsorship and Distribution Asset focuses specifically on that commercial translation.

Persistence Is Not Portability

Questions This Section Answers

1. Does a source that remains stable month to month also tend to appear across multiple AI platforms?
2. Why can durable AI citation authority remain concentrated on one platform?
3. Should CMOs report citation persistence and portability as separate KPIs?

High persistence does not mean high portability.

That is the central finding of the parent study. Across the same 283 markets, median persistence was 63.93%, while median portability was only 22.70%.

The most persistent markets often demonstrate the distinction clearly. Long-Term Care Insurance reached 85.19% persistence but only 23.46% portability. Outdoor Apparel and Technical Outfits reached 81.82% persistence and 19.76% portability. Final Expense Insurance reached 78.57% persistence and 20.48% portability.

A source ecosystem can therefore be durable without being universal.

The supporting article Persistence vs Portability: Why a Source That Wins on ChatGPT May Not Transfer to Gemini or Google AI focuses on that distinction.

Methodology

The underlying benchmark covers July, August, and September 2026 and six consumer AI surfaces: ChatGPT, Gemini, Perplexity, Microsoft Copilot, Google AI Overviews, and Google AI Mode.

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

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

A month portfolio required at least 100 retained responses and at least 100 resolved domains. The ranked list then used exactly the Top 100 domains.

Persistence compares eligible monthly Top-100 portfolios using Jaccard intersection divided by union. A market's headline persistence score is the median of its eligible month-pair overlaps.

The Top-25 and Top-50 sensitivity calculations retain the same baseline support requirement and the same 283-market cohort, then rank exactly 25 or 50 domains.

Missing or thin cells are excluded rather than scored as zero.

Limitations

Citation persistence is descriptive, not causal.

The dataset does not establish a certified fixed panel of identical prompts, prompt versions, model versions, countries, retrieval settings, or collection configurations across every month.

A market may therefore show higher or lower overlap because the observed sample changed, because AI source selection changed, because the underlying web changed, or because multiple conditions changed together.

Other important limitations include:

• Date provenance: 104,071 responses in the strict cohort use ingestion-date proxies.
• Market dependence: markets can share domains, companies, prompts, and source ecosystems.
• Taxonomy: historical observations are mapped into the current market taxonomy.
• Domain-level analysis: persistence is measured at registrable-domain level, not exact URL or passage level.
• No recommendation claim: recurring source authority does not establish recurring company recommendations.
• No future guarantee: past persistence does not guarantee that the same domains will remain important next month.

Among the 183 markets with no ingestion-date proxies, all 183 still had a positive Persistence-Portability Gap and the median persistence remained 63.93%. Among the 246 markets with all three months and all six platforms represented in the marginal portfolios, median persistence was 62.60%. Those checks support the broad descriptive pattern without eliminating the limitations above.

Conclusion

AI citation-source portfolios were not static between July and September 2026, but they were far from random.

The median commercial market retained 63.93% Jaccard overlap across its eligible monthly Top-100 source portfolios. Nearly 89% of markets had persistence of at least 50%, and more than two-thirds reached 60% or higher.

The most important complication is that persistence itself moved through time. August and September were much closer at 70.94% median overlap than July and September at 58.73%. Some individual markets changed by more than 30 or even 50 points depending on the month pair.

That makes citation persistence useful, but only when it is measured as a time series rather than a single comparison.

For CMOs, the signal answers a commercially valuable question: which third-party source relationships look durable enough to deserve continued attention?

For publishers, it provides a defensible way to distinguish recurring source authority from one-time visibility.

And for researchers, it reinforces the broader persistence-portability finding: AI source ecosystems can have a durable temporal core even while remaining highly fragmented across platforms.

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 benchmark.
• Persistence vs Portability: Why a Source That Wins on ChatGPT May Not Transfer to Gemini or Google AI examines why temporal durability does not guarantee cross-platform transfer.
• Top 25 vs Top 100 AI Citation Sources: Where Durable Authority Actually Lives analyzes portfolio-depth sensitivity.
• How Publishers Can Turn Persistent AI Citation Authority Into a Sponsorship and Distribution Asset focuses on publisher commercialization.
• Six AI Engines, Six Different Webs compares the source ecosystems of the six major AI surfaces.
• There Is No Universal GEO Strategy examines industry-level differences in both persistence and portability.
• How Stable Are AI Citations for High-Stakes Consumer Decisions? provides earlier repeated-query citation-drift research from a separate study population.

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, portability, market breadth, source depth, and recommendation authority so that recurring source visibility is not mistaken for universal AI influence.