Which AI Platforms Share the Most Citation Sources, and Which Diverge the Most?

Answer Capsule

Google AI Overviews and Google AI Mode were not merely the highest-overlap pair on an aggregate leaderboard. They were the highest-overlap Top-100 citation-source pair in every one of the 260 markets where all 15 platform comparisons were available.

At the other end, Microsoft Copilot dominated the low-overlap side of the matrix. A Copilot pair occupied the bottom position in 227 of 260 complete markets, or 87.31%. The single most common lowest-overlap relationship was Gemini vs Microsoft Copilot, which finished last in 120 markets, or 46.15%.

The full six-platform hierarchy was also more stable market by market than a simple table of global medians might suggest. When each complete market's 15 pairwise overlaps were ranked and compared with the aggregate pair ordering, the median Spearman rank correlation was 0.876. Of the 260 complete markets:

• 186, 71.54%, had a correlation of at least 0.80 with the overall ordering;
• 110, 42.31%, were at or above 0.90;
• 247, 95.00%, were at or above 0.60;
• only 13 markets, 5.00%, fell below 0.60;
• and only one, DUI Lawyers, fell below 0.40, at 0.265.

That means the aggregate platform hierarchy is useful, but it is not a substitute for market-specific source measurement.

The most stable conclusion is the top of the matrix: Google AI Overviews and Google AI Mode are consistently the closest citation-source ecosystems. The least stable part is the middle. After the Google sibling pair, no single runner-up dominated. In the 260 complete markets, the second-highest pair was:

• Gemini vs Google AI Mode in 79 markets, 30.38%;
• Gemini vs Google AI Overviews in 61, 23.46%;
• Perplexity vs Google AI Mode in 42, 16.15%;
• Perplexity vs Google AI Overviews in 29, 11.15%;
• ChatGPT vs Perplexity in 25, 9.62%;
• ChatGPT vs Microsoft Copilot in 14, 5.38%;
• and another pair in the remaining 10 markets.

The threshold results make the separation even clearer. Google AI Overviews vs AI Mode reached at least 40% overlap in 255 of 283 markets, 90.11%, and at least 50% in 190, 67.14%. By contrast, Gemini vs Copilot was below 20% overlap in 234 of 265 eligible markets, 88.30%, and never reached 30%.

The practical conclusion for CMOs, agencies, and publishers is:

Use the aggregate pair ranking to understand the normal structure of the six-platform citation ecosystem, but use market-level pair data before assuming that the normal relationship applies to your category.

This study measures overlap among cited registrable-domain portfolios. It does not establish answer similarity, model architecture, retrieval systems, source quality, recommendation influence, traffic, or future rankings.

Key Findings

The Aggregate Ranking Is Clear, but the Market-Level Ranking Is More Informative

Questions This Section Answers

1. Which AI platform pairs share the most and least Top-100 citation sources across commercial markets?
2. How stable is the global AI platform-overlap ranking when each commercial market is examined separately?
3. Can a CMO use the aggregate platform-pair ranking as a substitute for market-specific source analysis?

The global ranking provides a useful first approximation of how the six AI citation ecosystems relate to one another.

The Six AI Engines, Six Different Webs pillar establishes the complete matrix and the broad source-fragmentation result.

The more specific question here is whether this ordering survives when the 15 pair values are ranked separately inside each market.

It usually does.

For each of the 260 complete markets, this analysis compared that market's 15 pair values with the aggregate ordering above using a descriptive Spearman rank correlation.

The median correlation was 0.876.

The result is important because it separates two ideas that can otherwise be confused.

First, the pair hierarchy is not random. Most markets broadly resemble the global ordering.

Second, a high aggregate rank does not guarantee the same relative rank in every market.

The hierarchy therefore works well as a prior for platform planning. It should not be treated as a category-specific fact until the market is measured.

Google AI Overviews and AI Mode Were the Only Universally Stable Top Pair

Questions This Section Answers

1. Which AI platform pair was most consistently similar across individual markets?
2. Did any pair other than Google AI Overviews and Google AI Mode repeatedly rank first in complete markets?
3. How much more stable was the Google search-surface relationship than the rest of the platform matrix?

One part of the ranking was completely stable.

In all 260 complete markets, the highest-overlap pair was:

Google AI Overviews vs Google AI Mode

The pair ranked first 260 times out of 260, with no ties.

No other pair recorded a first-place finish in those complete markets.

That is a much stronger statement than saying its aggregate median was 53.85%.

The threshold distribution reinforces the result.

The observed range still matters.

The pair's minimum market overlap was 25.00% in Law Firms. Its maximum was 76.99% in Final Expense Insurance.

So the Google search surfaces were consistently closer to each other than any other pair, but the absolute degree of overlap varied substantially by market.

This is why the dedicated Google siblings analysis uses “siblings, not twins” as the editorial framing.

The source-overlap evidence supports a uniquely close relationship. It does not support treating the two surfaces as identical.

The Runner-Up Position Was Much More Competitive

Questions This Section Answers

1. Which AI platform pair most often ranked second behind Google AI Overviews and AI Mode?
2. Was Gemini consistently the next-closest platform to Google's AI search surfaces?
3. How often did Perplexity, ChatGPT, or Copilot become the second-most-similar source pair inside a market?

After the universal Google sibling pair, the ranking becomes much less deterministic.

The second-place pair across the 260 complete markets was distributed as follows:

The two Gemini-Google relationships combined for 140 runner-up finishes, or 53.85% of complete markets.

The two Perplexity-Google relationships combined for 71 runner-up finishes, or 27.31%.

That supports the aggregate finding that Gemini generally sits slightly closer to Google's search surfaces than Perplexity does.

But it also shows why the hierarchy should not be oversimplified.

In almost half of complete markets, the second-closest pair was not a Gemini-Google pair.

ChatGPT and Perplexity finished second in 25 markets. ChatGPT and Copilot did so in 14. Even a Copilot-Google relationship finished second twice.

A CMO looking only at aggregate medians would miss that market-specific flexibility.

Pair Rank Stability Separates the Strong Relationships From the Volatile Ones

Questions This Section Answers

1. Which AI platform-pair relationships stayed near the top of the citation-overlap ranking across markets?
2. Which platform pairs moved most between high and low positions depending on the commercial category?
3. How often did ChatGPT-Perplexity or ChatGPT-Copilot become a Top-3 source-overlap relationship?

A useful way to distinguish stable from volatile platform relationships is to look at each pair's rank distribution across the 260 complete markets.

Several patterns stand out.

The two Gemini-Google relationships are high-ranking but not fixed at ranks two and three. Their median market ranks are 3.0 and 3.5 because Perplexity-Google and occasionally ChatGPT relationships move ahead of them.

ChatGPT vs Perplexity is more volatile. Its aggregate median ranks sixth, but it entered the Top 3 in 43 markets and the Top 5 in 97.

ChatGPT vs Copilot has the same displayed aggregate median, 25.00%, but it entered the Top 3 in only 27 markets and the Top 5 in 76.

That illustrates a recurring problem with aggregate medians: equal medians do not imply equal rank behavior across markets.

The same applies to the two 30.72% Gemini-Google medians and the two 29.03% Perplexity-Google medians.

Equal Aggregate Medians Can Hide Different Market Behavior

Questions This Section Answers

1. If two AI platform pairs have the same median citation overlap, do they behave similarly across individual markets?
2. How did ChatGPT-Perplexity differ from ChatGPT-Copilot despite both having 25% median overlap?
3. Why should agencies retain the full pair distribution instead of reporting only one median percentage?

Several pair relationships share the same displayed aggregate median:

• Gemini vs Google AI Mode and Gemini vs Google AI Overviews: 30.72%;
• Perplexity vs Google AI Mode and Perplexity vs Google AI Overviews: 29.03%;
• ChatGPT vs Perplexity and ChatGPT vs Copilot: 25.00%;
• ChatGPT vs Gemini and Perplexity vs Copilot: 19.76%.

Those equal medians do not make the relationships interchangeable.

Take ChatGPT's two 25% relationships.

ChatGPT vs Perplexity:

• ranked in the Top 3 in 43 complete markets;
• ranked in the Top 5 in 97;
• fell into the bottom three in only 3;
• reached at least 30% overlap in 39 of 270 eligible markets, 14.44%;
• reached at least 40% in 2 markets.

ChatGPT vs Microsoft Copilot:

• ranked in the Top 3 in 27 complete markets;
• ranked in the Top 5 in 76;
• fell into the bottom three in 11;
• reached at least 30% overlap in 38 of 268 eligible markets, 14.18%;
• never reached 40%.

The difference is not enormous, but it is real.

Similarly, the two Gemini-Google pairs share a 30.72% median, yet Gemini vs Google AI Mode finished second 79 times while Gemini vs Google AI Overviews finished second 61 times.

For reporting, the lesson is straightforward:

A single median is useful for orientation. Pair rank frequencies and threshold distributions are needed to understand how often that relationship actually holds.

The Bottom of the Matrix Was Dominated by Copilot

Questions This Section Answers

1. Which AI platform pairs most often had the lowest citation-source overlap inside a market?
2. How often did Microsoft Copilot appear in the least-similar platform relationship?
3. Was Gemini-Copilot consistently the most divergent AI platform pair?

The bottom position was less universal than the top, but it was highly concentrated around Microsoft Copilot.

A Copilot pair occupied the bottom position in 227 of 260 complete markets, 87.31%.

Gemini-Copilot alone accounted for almost half of all last-place finishes.

The threshold data is even more one-sided.

This is why the separate Microsoft Copilot Lives on a Different Web research cluster is warranted.

The current result does not imply that Copilot is inferior. It says its observed source portfolio was frequently less similar to the other platforms' portfolios.

Thresholds Reveal More Than the Median Ranking Alone

Questions This Section Answers

1. Which AI platform pairs regularly exceeded 30%, 40%, or 50% Top-100 citation-source overlap?
2. Which pair relationships almost never achieved high source overlap?
3. How should a CMO distinguish a consistently close platform pair from one with only an attractive median?

The pair medians show the center of each distribution. Threshold frequencies show how often the relationship reaches strategically meaningful levels of overlap.

Only the Google sibling pair exceeded 40% overlap in anything close to a large majority of markets.

The next-best Gemini-Google relationships crossed 40% in fewer than 8% of eligible markets.

Most of the matrix therefore lives in a much lower-overlap regime than the top pair.

This matters for planning because “second closest” can sound more similar than the numbers justify. A pair can rank second relative to the alternatives and still share fewer than half of their Top-100 domains.

Market Extremes Show How Far Individual Pair Relationships Can Move

Questions This Section Answers

1. Which commercial markets produced the highest and lowest citation-source overlap for major AI platform pairs?
2. Can a usually weak platform relationship become relatively strong in a particular market?
3. How large is the observed market range for ChatGPT, Gemini, Perplexity, Copilot, and Google pair overlaps?

The pair distributions contain substantial market-level ranges.

The market ranges make an important point.

A pair that is generally weak can become comparatively strong in a particular category. A generally strong relationship can also compress sharply.

That is why the aggregate hierarchy should be used to structure research, not to eliminate market-specific measurement.

DUI Lawyers Was the Strongest Warning Against Blindly Applying the Global Ordering

Questions This Section Answers

1. Which market differed most from the overall AI platform-overlap ranking?
2. What does the DUI Lawyers source-overlap pattern reveal about using aggregate platform benchmarks?
3. Can a low-overlap platform pair move far above its normal global position in one market?

DUI Lawyers was the clearest ordering exception among the 260 complete markets.

Its 15 pair values had a Spearman correlation of only 0.265 with the global pair ordering, the lowest in the complete sample.

The top pair remained the same:

• Google AI Overviews vs Google AI Mode: 34.23%.

But the rest of the ranking looked very different.

Two things happened at once.

First, many normally higher-overlap relationships collapsed. The Perplexity-Google relationships, normally near the top of the aggregate matrix, fell to 9.29%.

Second, Copilot vs Google AI Mode became the second-highest pair in the market even though that relationship ranks near the bottom globally.

The lesson is not that DUI Lawyers should define the cross-platform strategy for legal markets in general.

It is that market-level pair ordering can deviate enough to change source-prioritization decisions.

The global ranking is a benchmark. The local ranking is the operating evidence.

Adjustable Beds Shows a Different Kind of Ranking Inversion

Adjustable Beds had a market-order correlation of 0.430, also among the lowest in the complete sample.

Google AI Overviews and AI Mode remained first at 58.73%.

But several Copilot relationships rose unusually high:

• ChatGPT vs Copilot: 26.58%;
• Copilot vs Google AI Overviews: 26.58%;
• Copilot vs Google AI Mode: 25.79%.

The Copilot-Google pairs normally rank 14th and 13th in the aggregate table. In Adjustable Beds, they moved into the upper half of the market's pair ordering.

At the same time, Perplexity-Google relationships, normally around ranks four and five globally, fell below 20%.

This is exactly the kind of category-specific inversion that a single platform-overlap leaderboard cannot reveal.

The Global Ordering Was Especially Consistent in Many Technology and Finance Markets

The exceptions matter, but they should not obscure the broader result.

Many markets followed the aggregate pair hierarchy extremely closely.

Examples of the highest Spearman correlations with the global ordering included:

• VPN Services: 0.985;
• Web Hosting: 0.980;
• IVF Clinics: 0.978;
• Prenatal Vitamins: 0.978;
• Online Estate Planning and Wills: 0.975;
• Fertility Clinics: 0.974;
• Online Car Buying Sites: 0.974;
• Trademark Registration Services: 0.971;
• Human Resources Software: 0.971;
• Online Therapy: 0.971.

The fact that 71.54% of complete markets reached a rank correlation of at least 0.80 means the aggregate table has real descriptive value.

It is not merely an average produced by wildly unrelated market structures.

The correct interpretation is therefore balanced:

The six-platform source-overlap hierarchy is usually recognizable at the market level, but individual categories can contain large enough inversions to matter operationally.

How CMOs and Agencies Should Use the Pair Ranking

Questions This Section Answers

1. How should a CMO use a cross-platform citation-overlap ranking when deciding which AI source maps can share work?
2. When should an agency override the global AI platform-overlap benchmark with market-specific evidence?
3. What pair-level metrics belong in an AI visibility or GEO reporting system?

The aggregate pair ranking is most useful as a planning prior.

It can guide the order in which a team checks for reusable source work.

A reasonable process is:

1. Treat Google AI Overviews and Google AI Mode as the first reuse candidate. Their source overlap was uniquely and consistently high, but still not complete.
2. Check Gemini-Google and Perplexity-Google relationships next. They usually occupy the next tier, but the runner-up varies by market.
3. Measure ChatGPT relationships separately. ChatGPT's strongest aggregate pair relationships were only 25% Jaccard.
4. Give Copilot an explicit independent check. Copilot relationships dominate the low-overlap side of the matrix.
5. Override the prior when the market data disagrees. DUI Lawyers and Adjustable Beds show that pair rankings can invert materially.

For reporting, a useful pair-level dashboard should preserve at least:

• platform pair;
• market;
• eligible response support;
• Top-100 Jaccard overlap;
• shared-domain count;
• each platform's nonshared domains;
• pair rank within the market;
• change over time when a comparable monitoring panel is available.

Do not turn the pair table into a causal claim.

A low-overlap pair does not prove that two platforms use different retrieval architecture. A high-overlap pair does not prove that one platform copies another. The study observes cited-domain sets only.

The companion article, Can One GEO Strategy Work Across ChatGPT, Gemini, Perplexity, Copilot, Google AI Overviews, and Google AI Mode?, translates the overlap structure into a broader operating model.

Methodology

The study uses the frozen July through September 2026 benchmark of 283 qualifying tracked markets, 876,605 retained AI responses, and 8,718,629 citation occurrences resolved to registrable domains.

The six consumer AI surfaces are:

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

For each eligible 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 portfolio ranking.

A strict Top-100 platform portfolio required:

• at least 100 retained responses;
• at least 100 resolved registrable domains;
• and an actual 100-domain ranked portfolio.

Thin or missing cells were excluded rather than assigned zero.

Pair overlap uses Jaccard similarity:

intersection / union

Because both eligible lists contain exactly 100 domains, a pair's Jaccard value can also be translated into a shared-domain count. For example, 40 shared domains produce a union of 160 and therefore 25% Jaccard overlap.

The aggregate pair table summarizes eligible markets separately for each pair. The market-rank analysis in this article is restricted to the 260 markets with all 15 platform-pair comparisons available.

Within each complete market, the 15 pair values were ranked from highest to lowest. Descriptive Spearman rank correlation compares that market-specific ordering with the ordering implied by the aggregate pair medians. It is used as a descriptive concordance measure, not a causal or inferential test.

Limitations

Pair overlap measures source sets, not answers

Two platforms can share domains while producing different answers, recommendations, rankings, or factual claims. Conversely, two platforms can produce similar answers while citing different sources.

Registrable-domain overlap is not URL overlap

Two systems can cite the same publisher domain but use different pages. Publisher-level similarity can therefore overstate page-level similarity.

Equal medians do not imply equal distributions

Several pair relationships have equal displayed medians. Their market-rank frequencies and threshold distributions still differ.

Market samples are observational

The historical benchmark does not certify an identical fixed panel of prompts, model versions, geographies, retrieval settings, or collection conditions across all markets and platforms.

Rank correlation does not validate one universal ordering

The descriptive Spearman statistic shows how closely a market's pair ordering resembles the aggregate table. It does not prove that the aggregate ordering is a stable law or predict future market behavior.

Pair support varies

Different platform pairs have different numbers of eligible markets because thin cells are excluded. The 260-market rank analysis therefore uses only markets where all 15 comparisons were available.

Citation overlap does not establish commercial value

Neither high nor low overlap proves traffic, audience quality, conversion value, sponsorship performance, or recommendation influence.

Conclusion

The six AI platforms do have a recognizable citation-source hierarchy.

Google AI Overviews and Google AI Mode sit at the top with unusual consistency. They ranked first in all 260 complete markets and exceeded 40% overlap in 90.11% of eligible markets.

Gemini and Perplexity usually form the next tier around Google's search surfaces, but the runner-up position changes frequently by market.

ChatGPT relationships occupy the middle of the matrix and can move substantially depending on category.

Copilot dominates the low-overlap side, with a Copilot pair finishing last in 87.31% of complete markets. Gemini-Copilot alone finished last in 46.15%.

Most importantly, the global hierarchy usually survives at the market level, but not perfectly. The median market's pair ranking correlated 0.876 with the global ordering, while rare markets such as DUI Lawyers showed major inversions.

That leads to the right operating rule:

Use the global platform-pair hierarchy to decide where to look first. Use the market-specific hierarchy to decide what to do.

Related LLM Authority Index Research

• Six AI Engines, Six Different Webs: Comparing Citation Source Overlap Across ChatGPT, Gemini, Perplexity, Copilot, Google AI Overviews, and Google AI Mode is the parent platform-overlap pillar.
• Can One GEO Strategy Work Across ChatGPT, Gemini, Perplexity, Copilot, Google AI Overviews, and Google AI Mode? translates overlap into program architecture.
• 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 the Google-family relationship in detail.
• How Publishers Should Think About Platform-Specific AI Citation Demand focuses on publisher implications.
• Microsoft Copilot Lives on a Different Web develops Copilot's distinctive source ecosystem.
• Persistence vs Portability: Why a Source That Wins on ChatGPT May Not Transfer to Gemini or Google AI distinguishes temporal durability from platform transfer.
• The Persistence-Portability Gap at Scale provides the broader 283-market benchmark.

About LLM Authority Index

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