Persistence vs Portability: Why a Source That Wins on ChatGPT May Not Transfer to Gemini or Google AI

Answer Capsule

A source can be durable in an AI citation ecosystem without being portable across AI platforms.

Across the 283-market benchmark, median month-to-month citation persistence was 63.93%, while median cross-platform portability was only 22.70%. The difference was not abstract. When ChatGPT's Top-100 cited-domain portfolio was compared directly with other AI surfaces, the median overlap was:

• 19.76% with Gemini
• 21.21% with Google AI Overviews
• 21.21% with Google AI Mode
• 25.00% with Perplexity
• 25.00% with Microsoft Copilot

Because every eligible platform portfolio contained exactly 100 domains, those Jaccard values can be translated into approximate shared-domain counts. A 19.76% Jaccard overlap corresponds to 33 shared domains between two Top-100 lists. A 21.21% overlap corresponds to 35 shared domains. A 25.00% overlap corresponds to 40 shared domains.

The low transfer was widespread. 145 of 265 eligible markets, 54.72%, had ChatGPT-Gemini overlap below 20%. 229 of 265, 86.42%, were below 25%, and only 4 markets, 1.51%, reached 30% or higher.

The pattern was similar for Google. 115 of 274 markets, 41.97%, had ChatGPT-Google AI Overviews overlap below 20%, and only 6 markets, 2.19%, reached 30% or more. ChatGPT versus Google AI Mode was slightly less fragmented, but 200 of 274 markets, 72.99%, still remained below 25% overlap.

High persistence did not solve the transfer problem. Life Insurance Companies had 72.41% citation persistence, but ChatGPT overlapped only 8.70% with Gemini, 11.73% with Google AI Overviews, and 16.28% with Google AI Mode. Project Management Software had 73.91% persistence, but only 16.96% ChatGPT-Gemini overlap. Human Resources Software for Small Businesses had 73.91% persistence and only 15.61% ChatGPT-Gemini overlap.

The strategic implication for CMOs is straightforward: a durable source map is not automatically a portable source map. If a company identifies the publishers and domains that matter to ChatGPT and simply copies that list into a Gemini or Google AI strategy, the data suggests it will miss a large portion of the relevant source ecosystem in most markets.

This is a descriptive source-overlap analysis. It does not establish why one platform selects a source, whether platforms use shared retrieval infrastructure, whether a citation causes a recommendation, or whether citation visibility produces traffic or revenue.

Key Findings

Persistence and Portability Answer Different Business Questions

Questions This Section Answers

1. What is the difference between durable AI citation authority and cross-platform AI citation portability?
2. Can a source remain important to AI systems over time without appearing consistently across ChatGPT, Gemini, and Google AI?
3. Which citation metrics should a CMO track separately when evaluating third-party source authority?

Persistence measures recurrence over time. Portability measures transfer across platforms.

Those sound similar because both use source overlap, but they answer very different strategic questions.

Citation persistence asks whether the important cited domains in a market recur across completed months. In the parent benchmark, eligible monthly Top-100 domain portfolios are compared with Jaccard similarity, and the market's persistence score is the median of its eligible month-pair overlaps.

Citation portability asks whether the important cited domains in a market recur across AI platforms. Eligible platform-specific Top-100 portfolios are compared using the same Jaccard method, and the market's portability score is the median of its eligible platform-pair overlaps.

The parent study found 63.93% median persistence and 22.70% median portability, with persistence exceeding portability in 282 of 283 markets.

That means a source environment can have a durable core without having a universal core.

For a CMO, the difference is commercially meaningful. Persistence answers:

Which source relationships appear durable enough to keep monitoring or investing against?

Portability answers:

Which parts of that source strategy can reasonably be expected to matter on another AI platform?

A domain that repeatedly appears in ChatGPT can be valuable even if Gemini rarely cites it. A source that appears across multiple platforms can be valuable even if its month-to-month prominence is unstable. Those are different forms of authority and should not be collapsed into one visibility score.

The parent research, The Persistence-Portability Gap at Scale, establishes the broader market-level relationship. The companion study, How Stable Are AI Citation Sources Month to Month? A CMO Guide to Citation Persistence, focuses on the temporal side. This article isolates the transfer problem.

ChatGPT Was Not a Universal Source Map for Any Other Major AI Platform

Questions This Section Answers

1. How much did ChatGPT's Top-100 citation sources overlap with Gemini, Perplexity, Copilot, Google AI Overviews, and Google AI Mode?
2. Which AI platform had the highest citation-source overlap with ChatGPT across commercial markets?
3. How often did ChatGPT share at least 30% Top-100 source overlap with another AI platform?

ChatGPT's closest median relationships in this benchmark were Perplexity and Microsoft Copilot, both at 25.00% Jaccard overlap.

That still represents substantial source separation.

The upper quartiles matter as much as the medians.

Even at the 75th percentile, ChatGPT-Gemini overlap was only 22.70%. ChatGPT-Google AI Overviews and ChatGPT-Google AI Mode reached only 25.00% at the 75th percentile.

In other words, low overlap was not limited to the bottom of the market distribution.

The threshold counts make the fragmentation even clearer.

Only two ChatGPT-Perplexity market comparisons reached 40% or more. None of the ChatGPT-Gemini, ChatGPT-Google, or ChatGPT-Copilot comparisons reached 40%.

The data does not say that one platform is better. It says that the platforms often rely on different source portfolios inside the same commercial market.

The broader six-platform matrix is analyzed in Six AI Engines, Six Different Webs.

What a 19.76% Jaccard Overlap Actually Means

Questions This Section Answers

1. How many shared domains does a 19.76% Jaccard overlap represent when two AI platforms each have a Top-100 source list?
2. Why is Jaccard overlap different from simply saying two AI platforms share 20 of their Top 100 sources?
3. How should a CMO translate AI citation-overlap percentages into an understandable source-count comparison?

Jaccard overlap uses the intersection divided by the union.

For two lists that each contain exactly 100 domains:

Jaccard = shared domains / all unique domains across both lists

That means the headline pair medians correspond to more shared domains than a casual reading of the percentage might suggest.

So the ChatGPT-Gemini median of 19.76% corresponds to roughly 33 shared domains, not 20.

But the strategic conclusion is still strong. If only 33 domains are shared between two Top-100 portfolios, then 67 domains on ChatGPT's list are absent from Gemini's list and 67 domains on Gemini's list are absent from ChatGPT's list.

At the ChatGPT-Google AI Overviews median of 21.21%, two Top-100 lists share 35 domains. That leaves 65 domains unique to each list.

This is why a source strategy copied from one platform can miss so much of another platform's citation environment even when the overlap percentage does not look close to zero.

For context, Google AI Overviews and Google AI Mode were the closest pair in the entire six-platform study at 53.85% median overlap, corresponding to 70 shared domains in two equal Top-100 lists. That relationship is examined separately in Google AI Overviews and AI Mode Are Siblings, Not Twins. Gemini Is a Second Cousin. Perplexity Is the Family Friend Who Looks Suspiciously Related..

High Citation Persistence Did Not Guarantee ChatGPT-to-Gemini Transfer

Questions This Section Answers

1. Can a commercial market have highly persistent AI citation sources while ChatGPT and Gemini use very different Top-100 source portfolios?
2. Which markets combined at least 60% month-to-month citation persistence with less than 20% ChatGPT-Gemini overlap?
3. Does durable source authority on ChatGPT imply comparable source authority on Gemini?

The strongest evidence against a universal source strategy comes from markets that are stable over time but fragmented across platforms.

Among eligible ChatGPT-Gemini comparisons, 264 of 265 markets had market-level persistence greater than their ChatGPT-Gemini overlap. The median difference between market persistence and ChatGPT-Gemini overlap was 43.21 percentage points.

There were 91 markets with at least 60% market persistence and less than 20% ChatGPT-Gemini overlap.

Examples include:

This is the central commercial pattern.

A CMO could look at a market with 70% to 80% persistence and reasonably conclude that the source environment contains a durable core. But if that same market has 10% to 20% ChatGPT-Gemini overlap, the durable core is not transferring cleanly from one platform to the other.

The correct strategic conclusion is therefore not “persistent publishers matter everywhere.”

It is:

Persistent publishers matter, but the relevant set has to be measured separately by platform.

The Same Transfer Problem Appeared Between ChatGPT and Google's AI Search Surfaces

Questions This Section Answers

1. Does strong month-to-month AI citation persistence transfer from ChatGPT to Google AI Overviews or Google AI Mode?
2. How often did highly persistent markets still have less than 20% ChatGPT-Google citation overlap?
3. Should a CMO use ChatGPT source research as a substitute for Google AI source research?

The ChatGPT-to-Google comparisons show the same broad pattern.

For ChatGPT versus Google AI Overviews, 273 of 274 eligible markets had persistence greater than the pairwise platform overlap. The median persistence-minus-overlap difference was 41.67 points.

For ChatGPT versus Google AI Mode, 272 of 274 markets had persistence greater than the pairwise overlap, with a median difference of 41.39 points.

The high-persistence, low-transfer group was substantial:

• 68 markets had persistence of at least 60% and ChatGPT-Google AI Overviews overlap below 20%.
• 66 markets had persistence of at least 60% and ChatGPT-Google AI Mode overlap below 20%.

Several markets showed low transfer to all three Google-associated surfaces measured in this article.

This is why “optimize for AI” is too broad as an operational instruction.

A company can maintain one factual source of truth and one coordinated content strategy, but the source-acquisition and publisher-opportunity map should be validated on each priority platform.

Life Insurance Companies: 72.41% Persistence, but Only 16 Shared ChatGPT-Gemini Top-100 Domains

Questions This Section Answers

1. How different were ChatGPT and Gemini citation-source portfolios in the Life Insurance Companies market?
2. Which life-insurance sources appeared in ChatGPT's Top 100 but not Gemini's, and which appeared in Gemini but not ChatGPT?
3. Can a highly persistent life-insurance citation ecosystem still fragment sharply across AI platforms?

Life Insurance Companies is one of the clearest examples in the benchmark.

The market had:

• 72.41% persistence
• 14.94% overall portability
• 8.70% ChatGPT-Gemini overlap
• 11.73% ChatGPT-Google AI Overviews overlap
• 16.28% ChatGPT-Google AI Mode overlap

For the ChatGPT-Gemini comparison, the two Top-100 portfolios shared only 16 domains. Their union contained 184 distinct domains.

The shared group included recognizable sources such as:

• Reddit
• Forbes
• MoneyGeek
• Policygenius
• Insure.com
• Wikipedia
• Ethos
• Money.com
• Ladder Life
• Insurify
• Experian
• SelectQuote

But the nonshared portfolios were much larger.

Examples appearing in ChatGPT's Top 100 but not Gemini's included NerdWallet, The Wall Street Journal, Trustpilot, ConsumerAffairs, and Bankrate.

Examples appearing in Gemini's Top 100 but not ChatGPT's included Aflac, Guardian Life, Mutual of Omaha, Ramsey Solutions, LifeInsure, Insurance and Estates, and Ogletree Financial.

The Google comparisons were broader than Gemini but still highly fragmented. ChatGPT shared 21 of 100 domains with Google AI Overviews and 28 with Google AI Mode in the Life Insurance Companies market.

That is a commercially useful distinction. A life-insurance marketing team that identifies high-value source opportunities from ChatGPT alone could miss insurer-owned, specialist, or publisher sources that appear prominently in Gemini or Google AI.

The data does not prove that placing content on any of those domains will cause a company recommendation. It demonstrates that the observed source portfolios differ materially.

Job Posting Sites Shows the Same Problem in a Different Commercial Market

Questions This Section Answers

1. How much did ChatGPT's job-posting citation sources overlap with Gemini and Google AI?
2. Which job-search publishers appeared on one AI platform's Top-100 source list but not another's?
3. Does cross-platform source fragmentation also appear in employment and recruiting markets?

The Job Posting Sites market had lower overall persistence than Life Insurance Companies, but it provides a clean example of platform-specific source selection.

Its market-level persistence was 36.99%. ChatGPT overlap was:

• 14.29% with Gemini
• 19.05% with Google AI Overviews
• 15.61% with Google AI Mode
• 20.48% with Perplexity
• 14.29% with Microsoft Copilot

ChatGPT and Gemini shared 25 domains across their two Top-100 portfolios.

Shared sources included Indeed, Reddit, LinkedIn, Money.com, Forbes, ZipRecruiter, G2, Wikipedia, Coursera, Upwork, YouTube, Betterteam, and Capterra.

Examples in ChatGPT's Top 100 but not Gemini's included Glassdoor, Built In, TechRadar, Trustpilot, Chamber of Commerce, TechRepublic, and Salary.com.

Examples in Gemini's Top 100 but not ChatGPT's included FlexJobs, SimplyHired, Recruitee, Manatal, Scion Staffing, ResumeSpice, and Big Interview.

The Google AI Overviews portfolio shared 32 domains with ChatGPT, while Google AI Mode shared 27.

This is precisely the kind of market where a universal publisher list can mislead a CMO. A company might see strong presence around Indeed, LinkedIn, Forbes, and ZipRecruiter across several systems and conclude that the source ecosystem is mostly the same. The long tail around that shared core says otherwise.

Some Markets Were More Portable Than Others

Questions This Section Answers

1. Are there commercial markets where ChatGPT shares substantially more citation sources with Gemini or Google AI?
2. Which markets provide counterexamples to extremely low ChatGPT cross-platform overlap?
3. Should a CMO expect the same level of source fragmentation in every industry?

The transfer problem was broad, but it was not identical everywhere.

Mortgage and Credit Repair provide useful counterexamples.

Those are still far from identical source portfolios, but the transfer is meaningfully higher than in Life Insurance Companies or Job Posting Sites.

In Mortgage, ChatGPT and Google AI Overviews shared 49 domains, corresponding to 32.45% Jaccard overlap. ChatGPT and Microsoft Copilot also shared 49.

In Credit Repair, ChatGPT and Perplexity reached 38.89% overlap, one of the stronger ChatGPT pair relationships in the benchmark.

These cases reinforce a core measurement principle: platform portability has to be benchmarked inside the market.

The overall median is useful for context. It should not replace market-specific evidence.

Sector-Level Transfer Was Especially Weak in Legal and Health Markets

Questions This Section Answers

1. Which broad industries had the lowest ChatGPT-Gemini and ChatGPT-Google AI citation-source overlap?
2. Is ChatGPT source portability higher in finance than in legal or health markets?
3. Should cross-platform AI source benchmarks be adjusted for industry context?

The market-level pair data also reveals meaningful sector differences.

Finance & Insurance had the highest median ChatGPT-Gemini transfer at 22.70%. Legal & Compliance had the lowest at 11.73%.

The same general pattern appeared against Google AI Overviews and Google AI Mode. Legal & Compliance remained the least aligned macro sector in both comparisons.

This does not mean legal source strategies are inherently worse or that finance is universally easier. The sector populations differ in market count, coverage, source structure, and data composition. The result is descriptive.

It does mean that a single cross-industry portability benchmark can hide large strategic differences. The industry-specific pillar, There Is No Universal GEO Strategy, develops that point further.

There Were Rare Markets Where Platform Overlap Exceeded Temporal Persistence

Questions This Section Answers

1. Are there markets where ChatGPT and another AI platform share more citation sources than the market retains month to month?
2. Which markets are exceptions to the pattern that persistence exceeds ChatGPT cross-platform overlap?
3. Why should AI citation research preserve low-persistence exceptions instead of forcing a universal rule?

The broad result should not be turned into an absolute rule.

For ChatGPT versus Gemini, 264 of 265 eligible markets had persistence greater than the pairwise platform overlap. The exception was Home Equity Loans, where persistence was 18.34% and ChatGPT-Gemini overlap was 21.21%.

For ChatGPT versus Google AI Overviews, 273 of 274 markets had higher persistence. Again, Home Equity Loans was the exception, with 22.70% ChatGPT-Google AI Overviews overlap.

For ChatGPT versus Google AI Mode, there were two exceptions:

These exceptions matter because they reveal different failure modes.

Home Equity Loans had unusually low temporal persistence. Its problem was not unusually high platform transfer. Its monthly source portfolios were simply far less stable than most markets.

Financial Technology and Banking Software is more interesting. ChatGPT-Google AI Mode overlap reached 38.89%, while market persistence was only 31.58%. In that market, the ChatGPT-AI Mode relationship was more similar than the market's own month-to-month source recurrence.

A research program should preserve those cases rather than smoothing them away. Exceptions help define where a benchmark is useful and where local measurement has to override the global pattern.

What This Means for a CMO Building a Publisher or Source Strategy

Questions This Section Answers

1. How should a CMO change publisher targeting when ChatGPT citation sources do not transfer reliably to Gemini or Google AI?
2. Should companies maintain separate source maps for ChatGPT, Gemini, Perplexity, Copilot, and Google AI?
3. How can a marketing team use shared and platform-specific AI citation sources without duplicating its entire content strategy?

The data does not imply that a company needs six disconnected GEO programs.

It does imply that one source map is not enough.

A practical strategy can separate the source ecosystem into three layers.

1. Shared core sources

These are domains that recur across multiple priority platforms in the same market.

They deserve special attention because work that improves factual coverage, publisher relationships, or brand presence on those domains may be relevant to several AI surfaces at once.

2. Platform-specific authority sources

These are domains that matter strongly on one platform but are absent or weak on another.

The Life Insurance Companies example shows why this layer matters. A source set inferred from ChatGPT would miss several Gemini-specific and Google-specific domains. The same pattern appears in Job Posting Sites.

3. Durable but nonportable sources

These are domains that recur over time within a market or platform but do not transfer widely.

They can still be valuable. A source does not need universal platform coverage to matter commercially. It needs to matter on a platform that matters to the company's customers and business objectives.

The operational workflow should therefore be:

1. define the commercial market;
2. build a source portfolio for each priority platform;
3. identify the shared core;
4. identify high-value platform-specific gaps;
5. track persistence over time;
6. inspect the actual cited URLs and context;
7. separately measure company recommendations and sentiment.

That final step is essential. Citation authority and recommendation authority are different. The distinction is developed in AI Citation Authority vs AI Recommendation Authority: Why Being a Source Is Not the Same as Being Recommended.

What This Means for Publishers

Questions This Section Answers

1. Can a publisher have strong AI citation authority on ChatGPT without having comparable authority on Gemini or Google AI?
2. How should publishers describe platform-specific AI citation authority to advertisers?
3. What evidence should a publisher provide before claiming broad cross-platform AI visibility?

For publishers, low portability is not necessarily bad news.

A platform-specific citation footprint can be commercially meaningful if the publisher can document it accurately.

The problem is overclaiming.

A publisher that performs strongly on ChatGPT should not automatically describe itself as broadly authoritative across “AI search.” The benchmark shows why that claim can be misleading.

A better advertiser-facing evidence package would show:

• the specific commercial markets where the publisher appears;
• the AI platforms where it appears;
• its source rank or response presence on each platform;
• whether the footprint persists over time;
• the cited URLs and article types;
• and the degree of cross-platform portability actually observed.

A publisher with concentrated strength on one AI surface can present that as a platform-specific asset. A publisher with broad transfer can demonstrate that separately.

The supporting article How Publishers Should Think About Platform-Specific AI Citation Demand focuses on the publisher side of this problem.

Methodology

The underlying benchmark covers July, August, and September 2026 across six consumer AI surfaces:

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

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

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

A platform portfolio required at least:

• 100 retained responses;
• 100 resolved registrable domains;
• and an actual ranked list of exactly 100 domains.

Thin or missing cells were excluded rather than scored as zero.

Platform portability comparisons use Jaccard similarity:

intersection / union

This article uses the frozen Top-100 platform-pair values from the publication-grade evidence package. The ChatGPT threshold counts and market examples are descriptive calculations from those same pair values. No new responses or external data were added.

Market-level persistence uses eligible monthly Top-100 portfolios pooled across the six consumer surfaces, as defined in the parent study.

Limitations

Several limitations are important when interpreting cross-platform transfer.

Source overlap does not identify platform architecture

Two platforms can cite similar domains for many reasons. Two platforms can cite different domains despite using related technologies. Citation overlap does not reveal backend retrieval systems, model training data, licensing relationships, ranking algorithms, or copying.

Domain overlap does not imply URL overlap

The study normalizes citations to registrable domains. Two platforms can both cite the same domain while citing entirely different pages.

The study does not establish identical prompts across platforms

The historical collection does not certify a fixed panel of identical prompts, prompt versions, model versions, countries, retrieval settings, or collection conditions across all platforms and months.

The results therefore describe observed source portfolios, not controlled platform effects.

Citation is not recommendation

A platform can cite a source while recommending a different company. A company can be recommended without its own domain being cited. Citation authority and recommendation authority require separate measurement.

Market samples are not independent randomized populations

Markets can share domains, companies, source ecosystems, and prompt concepts. The pair distributions are descriptive benchmarks, not classical independent-sample causal estimates.

High or low portability is not inherently good or bad

A fragmented market can create platform-specific opportunities. A highly portable market can create a concentrated source landscape. The business value depends on audience, market, source quality, competition, and platform importance.

Conclusion

The 283-market benchmark supports a simple but consequential distinction:

A source can win repeatedly without winning everywhere.

Month-to-month source persistence was much higher than cross-platform source portability. That gap becomes especially concrete when ChatGPT is compared directly with Gemini and Google AI.

The median ChatGPT-Gemini overlap was 19.76%, equivalent to 33 shared domains between two equal Top-100 lists. More than half of eligible markets were below 20% overlap, and only four reached 30% or higher.

ChatGPT versus Google AI Overviews and Google AI Mode was only modestly more aligned at 21.21% median overlap. Roughly three quarters of eligible markets remained below 25% overlap with each Google search surface.

The most important evidence is not the aggregate median. It is the coexistence of high persistence and low platform transfer inside the same markets.

Life Insurance Companies had 72.41% persistence but only 8.70% ChatGPT-Gemini overlap. Long-Term Care Insurance reached 85.19% persistence but 19.76% ChatGPT-Gemini overlap. Project Management Software reached 73.91% persistence but only 16.96% ChatGPT-Gemini overlap.

Those are not examples of weak authority. They are examples of durable authority that is platform-specific.

For CMOs, the practical conclusion is that a mature AI search program should preserve a shared factual and content foundation while measuring the source ecosystem separately for each priority platform.

For publishers, the same finding creates a more precise commercial language: durable authority, platform-specific authority, and cross-platform authority are different assets.

And for researchers, the result argues against collapsing AI citation performance into a single global 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 analyzes temporal source durability.
• Top 25 vs Top 100 AI Citation Sources: Where Durable Authority Actually Lives examines how source depth changes the persistence-portability relationship.
• Six AI Engines, Six Different Webs: Comparing Citation Source Overlap Across ChatGPT, Gemini, Perplexity, Copilot, Google AI Overviews, and Google AI Mode provides the full six-platform matrix.
• Can One GEO Strategy Work Across ChatGPT, Gemini, Perplexity, Copilot, Google AI Overviews, and Google AI Mode? translates platform differences into program design.
• Google AI Overviews and AI Mode Are Siblings, Not Twins. Gemini Is a Second Cousin. Perplexity Is the Family Friend Who Looks Suspiciously Related. examines Google-family and Perplexity citation resemblance.
• Microsoft Copilot Lives on a Different Web examines Copilot-specific source divergence.
• There Is No Universal GEO Strategy analyzes sector and industry variation.
• AI Citation Authority vs AI Recommendation Authority separates being used as a source from being recommended as a company.

About LLM Authority Index

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