Microsoft Copilot Lives on a Different Web: How Copilot Citation Sources Diverge From ChatGPT, Gemini, Perplexity, and Google AI Across 283 Markets

Answer Capsule

Microsoft Copilot did not behave like a simple extension of ChatGPT, Gemini, Perplexity, Google AI Overviews, or Google AI Mode in the citation-source data.

Across the 283-market benchmark, Copilot's median Top-100 Jaccard overlap with the other five surfaces was:

• 25.00% with ChatGPT
• 19.76% with Perplexity
• 17.65% with Google AI Mode
• 16.96% with Google AI Overviews
• 16.28% with Gemini

That already made Copilot one of the most source-distinct surfaces in the six-platform study. The stricter 260-market complete-platform subset makes the result stronger.

In those 260 markets, Copilot had the lowest average Top-100 overlap with the other five platforms in 215 markets, or 82.69%. It was also the most isolated platform in 219 markets at Top 50, 193 markets at Top 25 when ties were excluded, and 172 markets at Top 10 when ties were excluded.

The divergence was visible not just in pairwise percentages but in the source lists themselves. In the median complete market:

• 45 of Copilot's Top 100 domains appeared in none of the other five Top-100 lists;
• 21 of its Top 50 were unique to Copilot;
• 10 of its Top 25 were unique;
• and 4 of its Top 10 were unique.

Copilot also had the single largest platform-exclusive Top-100 set in 188 of 260 markets, or 72.31%. At Top 50 it led in 186 markets, at Top 25 in 166, and at Top 10 in 146.

ChatGPT was usually Copilot's nearest neighbor, but "nearest" should not be confused with "similar." ChatGPT was the unique closest platform to Copilot in 206 of 260 complete markets, 79.23%, yet their median Top-100 Jaccard overlap remained only 25.00%, equivalent to about 40 shared domains between two equal 100-domain lists.

Some markets were much more fragmented. In Dental Implants, 67 of Copilot's Top 100 domains appeared in none of the other five Top-100 portfolios. Copilot's Jaccard overlap was only 12.99% with ChatGPT, 9.89% with Gemini, 5.26% with Perplexity, 3.63% with Google AI Overviews, and 5.82% with Google AI Mode. Three Copilot Top-10 domains in that market, ThePricer.org, HowMuchBlog.com, and 4MDentalImplants.com, did not appear in any of the other five Top-100 lists.

The pattern was not universal. In Home Builders, Copilot's Top-100 overlap with ChatGPT reached 39.86%, the highest Copilot nearest-neighbor overlap in the complete-market cohort. Even there, 33 of Copilot's Top 100 domains were unique to Copilot, and MSN.com ranked eighth on Copilot while missing the other five Top-100 lists.

The commercial conclusion is narrow but important:

Microsoft Copilot should be measured as a distinct citation channel inside a coordinated AI visibility program. ChatGPT is often the best starting comparison, but it is not a reliable proxy for Copilot's source portfolio.

"Different web" is editorial shorthand for observed citation-source separation. This research does not establish different technical architecture, retrieval systems, training data, source-selection policy, answer quality, audience value, or future ranking behavior.

Key Findings

Copilot Was Usually the Most Isolated Platform in the Market

Questions This Section Answers

1. Was Microsoft Copilot the most source-divergent AI platform in most commercial markets?
2. Does Copilot remain isolated after comparing all five of its platform relationships inside the same market?
3. How often can ChatGPT, Gemini, Perplexity, or Google AI substitute for a Copilot source map?

The full six-platform matrix already showed that several Copilot relationships sat near the bottom of the cross-platform overlap ranking.

But pair medians alone cannot answer the more useful market-level question: which platform is most isolated inside a specific category?

For each of the 260 complete markets, this analysis calculated every platform's average Jaccard overlap with the other five platforms. The platform with the lowest average overlap was the most source-isolated platform for that market.

At the Top-100 level, Copilot was the most isolated in 215 of 260 markets, 82.69%. There were no Copilot ties at the minimum in this Top-100 calculation.

The median average overlap to the other five platforms was:

Copilot's median connectedness was therefore more than 3 points below ChatGPT and more than 11 points below either Google search surface.

This is not a measure of product quality. It means the domains in Copilot's market-level source portfolios overlapped less with the rest of the six-platform system.

The same result appeared at shallower cutoffs.

The finding matters because a platform can look different at rank 80 while sharing most of its genuinely important sources near the top. That is not what the Copilot data showed.

Even at Top 10, where each platform contributes only its most prominent observed domains, Copilot was still the uniquely most isolated platform in roughly two-thirds of complete markets.

Copilot's Nearest Neighbor Was Usually ChatGPT, but the Distance Was Still Large

Questions This Section Answers

1. Which AI platform had the most similar citation sources to Microsoft Copilot?
2. Is ChatGPT close enough to Copilot that marketers can reuse the same publisher list?
3. Which platform was most often Copilot's furthest source neighbor?

ChatGPT was the obvious nearest neighbor to Copilot in the aggregate matrix.

Across all eligible markets, Copilot's Top-100 median overlap was:

The complete-market comparison makes the relationship more concrete.

At Top 100, ChatGPT was Copilot's unique nearest neighbor in 206 of 260 markets, 79.23%. Perplexity was the unique nearest neighbor in 27 markets. Google AI Mode was nearest in 8, Google AI Overviews in 4, and Gemini in only 3. The remaining markets involved ties.

Gemini was Copilot's unique furthest neighbor in 135 of 260 markets, 51.92%. Google AI Overviews was furthest in 34, Google AI Mode in 30, Perplexity in 18, and ChatGPT in only 3. Additional markets contained ties for the minimum.

This creates a useful but bounded hierarchy:

ChatGPT is usually the best platform to compare against Copilot, but not a substitute for Copilot measurement.

A 25% median Jaccard overlap between equal Top-100 lists corresponds to roughly 40 shared domains and 60 domains in each list that are not shared with the other list.

The threshold distribution reinforces the point.

ChatGPT clearly had the strongest relationship to Copilot. It was still uncommon for the two surfaces to exceed 30% Jaccard overlap.

The Divergence Survives at Top 25 and Top 10

Questions This Section Answers

1. Does Copilot only diverge in the long tail of its Top-100 source list?
2. Are Copilot's highest-ranked citation sources more portable to ChatGPT, Gemini, Perplexity, and Google AI?
3. How much does platform overlap change when analysis is restricted to Top 10, Top 25, or Top 50 sources?

Shrinking the portfolio does not erase the Copilot gap.

Within the 260 complete markets, Copilot's median overlap with the other five platforms was:

The top of the portfolio can be somewhat more portable for some pairs, especially Perplexity. But the magnitude remains modest.

Copilot and ChatGPT are especially notable. Their median overlap is 25.00% at Top 10, Top 25, Top 50, and Top 100 in the complete-market subset.

That does not mean the same domains are shared at every cutoff. It means the median Jaccard value remains the same as the list expands.

The practical implication is that Copilot divergence cannot be dismissed as noise in the bottom half of a Top-100 list. The prominent source layer remains materially distinct.

A Typical Copilot Portfolio Contains More Platform-Exclusive Sources

Questions This Section Answers

1. How many Copilot Top-100 sources typically appear on no other AI platform in the same market?
2. Is Copilot's unique-source share larger than ChatGPT, Gemini, Perplexity, Google AI Overviews, or Google AI Mode?
3. Does Copilot platform specificity increase or decrease deeper in the source list?

Pairwise overlap asks how much Copilot resembles one other platform at a time.

A different question is more directly useful for source planning:

How many Copilot sources appear nowhere in the corresponding source lists of the other five platforms?

For each complete market, this analysis compared Copilot's ranked portfolio with the union of the other five platform portfolios at the same depth.

The median Copilot result was:

The middle half of markets had 41 to 51 Copilot-only domains in the Top 100.

At the same Top-100 depth, the median exclusive shares for the other platforms were:

Copilot therefore did not merely have lower pairwise overlap. Its actual market portfolios contained more domains that none of the other five platforms surfaced at the same cutoff.

The market-by-market leadership result is similarly clear.

Copilot had the sole largest platform-exclusive source set in:

• 146 of 260 markets at Top 10, 56.15%;
• 166 at Top 25, 63.85%;
• 186 at Top 50, 71.54%;
• 188 at Top 100, 72.31%.

The deeper the portfolio becomes, the more often Copilot contributes the largest unique source set.

Copilot's Source Separation Was Largest in Legal and Health Markets

Questions This Section Answers

1. In which industries was Copilot's citation-source portfolio most platform-specific?
2. Are legal, health, finance, software, and consumer markets equally divergent on Copilot?
3. Which sectors deserve the most caution when reusing a non-Copilot source list?

The 260 complete markets were grouped into the study's macro sectors. Copilot's median Top-100 exclusive share varied meaningfully.

The one unassigned market is excluded from this interpretation.

Legal markets stand out. In the median complete Legal & Compliance market, more than half of Copilot's Top-100 domains did not appear in any of the other five Top-100 source portfolios.

The Top-10 layer showed the same broad pattern. Legal, Health & Wellness, and Technology & Business each had a median Copilot-exclusive Top-10 share of 50%, while Finance & Insurance was lower at 30%.

The sector result does not establish why the source ecosystems differ. Legal and health markets may vary in prompt composition, domain supply, sampling structure, source-type mix, or other unmeasured characteristics.

It does show why an aggregate platform average should not substitute for category-specific diligence.

Copilot Separation Varied Sharply by Market

The median Copilot Top-100 exclusive count was 45, but the market range was much wider.

Across the 260 complete markets, the middle 50% ran from 41 to 51 Copilot-only Top-100 domains. The observed minimum was 28 and the maximum was 67.

The markets with the largest Copilot-exclusive Top-100 sets were:

Several relatively connected markets sat at the other end of the distribution.

This spread is commercially important. A single statement such as “Copilot is 45% unique” is useful for orientation, but it is too coarse for market planning. The observed Copilot-only share ranged from less than one-third of the Top 100 in several markets to roughly two-thirds in the most fragmented markets.

The source research burden therefore varies materially by category. A mortgage or help-desk team can begin with more reusable source intelligence. A dental-implant, legal, rehab, or PR team should expect a much larger Copilot-specific source layer.

Dental Implants Shows How Far Copilot Can Separate From the Other Five

Dental Implants was the most extreme complete market by Copilot Top-100 exclusivity.

Copilot's Top-100 list contained 67 domains that appeared in none of the other five Top-100 lists.

Its pairwise Top-100 Jaccard overlap was:

The Top-10 lists make the fragmentation tangible.

Copilot's Top 10 were:

1. AffordableDentures.com
2. NuviaSmiles.com
3. AspenDental.com
4. RealDentalCosts.com
5. ClearChoice.com
6. ThePricer.org
7. HowMuchBlog.com
8. MySpecialtyDentist.com
9. 4MDentalImplants.com
10. BBB.org

Three of those domains were especially platform-specific in this observed market:

• ThePricer.org ranked #6 on Copilot and did not appear in any other platform's Top 100;
• HowMuchBlog.com ranked #7 on Copilot and did not appear in any other Top 100;
• 4MDentalImplants.com ranked #9 on Copilot and did not appear in any other Top 100.

This is a much stronger separation than simply saying the platforms ranked common sources differently.

It means some of Copilot's leading observed domains were entirely absent from the other five 100-domain source portfolios in the same market.

Credit Repair Shows a More Shared Copilot Market

Not every market looks like Dental Implants.

In Credit Repair, Copilot's Top-100 overlap was:

• 33.33% with ChatGPT;
• 29.03% with Perplexity;
• 25.79% with Gemini;
• 24.22% with Google AI Mode;
• 23.46% with Google AI Overviews.

Copilot still had 36 Top-100 domains unique to Copilot, but the overall source ecosystem was much more connected than Dental Implants.

The Copilot Top 10 included ConsumerAffairs, CNBC, FinanceBuzz, ConsumerFinance.gov, MoneyLion, CreditSaint, AllAboutLawyer, TopConsumerReviews, Business Insider, and FirstCard.

AllAboutLawyer.com was a deep-exclusive case: it ranked #7 on Copilot and did not appear in the Top 100 on any of the other five surfaces.

Credit Repair demonstrates why the P4 conclusion should not be overstated. Copilot is usually distinct, but the size and composition of that difference vary materially by market.

Home Builders Is a Counterexample to the Most Extreme Isolation Story

Home Builders had the highest Copilot nearest-neighbor Top-100 overlap in the strict complete-market cohort.

Copilot and ChatGPT reached 39.86% Jaccard overlap.

That is still far from identical, but it is substantially above Copilot's 25.00% median relationship with ChatGPT.

Copilot had only 33 Top-100 domains unique to itself, one of the smaller exclusive sets in the cohort.

Yet even this relatively connected market contained a clear Copilot-specific leading source. MSN.com ranked #8 on Copilot, appearing in 31 of 500 retained Copilot responses, while missing the Top 100 on all five other surfaces.

This is the right way to use the outlier:

• it shows that Copilot can align much more closely with ChatGPT in some categories;
• it also shows that meaningful Copilot-only source opportunities can remain inside an otherwise shared market.

Some Domains Formed Distinctive Copilot Authority Clusters

The most useful source-level examples are not necessarily the domains with the most Copilot appearances. They are domains whose depth on Copilot materially exceeded their same-market depth elsewhere.

The strict 260-market analysis identified 235 domain-market cases where a domain ranked in Copilot's Top 10 but did not appear in the Top 100 on any of the other five platforms.

Several domains repeated that pattern across multiple markets.

LegalTechTrend.com

LegalTechTrend.com produced one of the clearest category clusters.

It entered Copilot's Top 10 in nine legal markets, while missing every other platform's Top 100 in those same markets:

This is not evidence that the domain is a better legal publisher or that Copilot has a preference for it. It is evidence of a Copilot-specific observed citation footprint across a coherent market cluster.

MSN.com

MSN.com reached Copilot's Top 10 in 11 complete markets. In eight of those markets it was absent from every other platform's Top 100.

Those deep-exclusive markets included:

• Certificates of Deposits
• Cybersecurity Services
• Equipment Financing
• EV Charging Networks
• Home Builders
• Money Transfer Fintech
• Online Betting Sites
• Real Estate Investment Trusts

In Money Transfer Fintech, MSN ranked #2 on Copilot with 124 response presences out of 600 retained Copilot responses and missed the other five Top-100 lists.

In Real Estate Investment Trusts, it ranked #2 with 91 of 400 and again missed the other five Top 100s.

LendEDU.com

LendEDU demonstrates a different pattern.

It reached Copilot's Top 10 in eight finance markets. It did not reach the Top 10 on another platform in any of those eight same-market cases, but it did appear somewhere in the other platforms' Top 100s.

That distinction matters.

LendEDU was Copilot-concentrated at the top of the ranking, not completely absent elsewhere.

Examples included:

• Student Loan Refinance: Copilot #3
• Student Loans: Copilot #4
• Personal Finance Tools: Copilot #6
• Budgeting Apps: Copilot #7
• Bad Credit Loans: Copilot #9
• Debt Management: Copilot #10
• Peer to Peer Lending: Copilot #10
• Personal Loans and Online Lenders: Copilot #10

A publisher or CMO should treat these two patterns differently. Deep exclusivity identifies a source that is missing entirely from the other platforms' Top 100. Top-rank concentration identifies a source that exists elsewhere but is much more prominent on Copilot.

What “Copilot Lives on a Different Web” Actually Means

The title is intentionally memorable, but the evidence supports a narrower technical statement than the headline might imply.

The study can say that:

• Copilot's citation-source portfolios overlap less with the other five platforms in most markets;
• Copilot frequently contributes the largest set of platform-exclusive domains;
• ChatGPT is usually its nearest source neighbor;
• Copilot's leading sources remain materially distinct even at Top 10 and Top 25;
• the degree of separation varies substantially by industry;
• and some domains develop deep Copilot-specific footprints across related markets.

The study cannot say that:

• Copilot uses a unique technical web index;
• Copilot has a specific source-suppression or preference policy;
• Microsoft intentionally favors particular publisher categories;
• Bing, Microsoft, OpenAI, or another provider caused the observed pattern;
• a domain's Copilot citation history will continue in the future;
• or a publisher placement will cause a company to be cited or recommended.

The phrase "different web" refers to the observed domain portfolio, not the infrastructure behind it.

What CMOs Should Measure Separately for Copilot

A coordinated AI search program does not require six disconnected strategies.

It does require a Copilot-specific measurement layer.

For each priority market, the minimum useful Copilot view should include:

1. Copilot Top-100 and Top-25 source portfolios
   Identify which domains recur and which are absent from other platforms.
2. Company recommendation coverage
   Keep recommendation measurement separate from citation-source measurement.
3. Competitor coverage
   Track which competitors are recommended and which third-party domains support the surrounding answer ecosystem.
4. Copilot-exclusive and Copilot-concentrated publishers
   Separate sources that exist only on Copilot from sources that are simply ranked much higher there.
5. Actual cited URLs
   Domain-level authority is only a starting point. A publisher may have hundreds or thousands of pages with different commercial relevance.
6. Longitudinal support
   A single collection period can identify a source candidate. Persistent authority requires repeated observation.

The dedicated companion article, Should CMOs Treat Microsoft Copilot as a Separate AI Search Channel?, develops the channel-design implications.

Publisher and PR Implications

Copilot's divergence creates a specific evidence opportunity for publishers.

A publisher should not sell a broad "AI authority" claim merely because it performs well on Copilot. But it can document a bounded Copilot-specific footprint.

A defensible media-kit statement would identify:

• the commercial market;
• the platform;
• the observation period;
• the publisher's Top-100, Top-25, or Top-10 position;
• response-presence support;
• whether the domain also appears on other platforms;
• and the pages actually cited where available.

That is especially useful for a publisher with Copilot-concentrated depth. LendEDU's finance examples illustrate one version. LegalTechTrend.com's legal-market footprint illustrates a more extreme version.

PR teams should similarly avoid assuming that a placement strategy built around ChatGPT or Google AI automatically covers Copilot. The candidate publisher universe can be materially different.

The companion analyses, Which Source Types Overperform in Microsoft Copilot Compared With Other AI Engines? and Why Microsoft Copilot's Citation Mix Creates Different Publisher and PR Opportunities, examine that question more narrowly.

Methodology

The primary benchmark uses the frozen July through September 2026 cohort of 283 qualifying tracked commercial 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

API copies and experimental surfaces are excluded from the primary consumer benchmark.

For each market-platform cell, domains are ranked by the number of retained responses containing that registrable domain. A domain is counted once per response for response-presence ranking. Ties are broken alphabetically.

An eligible Top-100 platform portfolio requires at least 100 retained responses and at least 100 resolved domains. Thin or missing cells are excluded rather than assigned zero.

Pair overlap is measured with Jaccard similarity:

intersection of domain sets / union of domain sets

The full 283-market pair summaries use all markets eligible for each particular pair, so support varies by pair from 265 to 272 markets for the five Copilot relationships.

The complete-market analyses in this article use the stricter subset of 260 canonical markets with eligible portfolios on all six platforms. Those analyses include:

• average overlap from each platform to the other five;
• nearest and furthest Copilot neighbor by market;
• platform-exclusive domain counts at Top 10, 25, 50, and 100;
• macro-sector summaries;
• and domain-market deep-exclusivity checks.

A platform-exclusive domain at depth K appears in one platform's Top-K portfolio and in none of the other five Top-K portfolios in the same market.

A deep-exclusive Top-10 case is stricter: the domain appears in Copilot's Top 10 and does not appear in the Top 100 on any of the other five platforms in that market.

These measures describe source-set structure. They do not measure recommendation authority, traffic, audience size, publisher quality, conversions, or commercial value.

Limitations

This is observational research from changing collected samples, not a certified identical-prompt matched panel across every platform and period.

Of the retained responses, 104,071 used ingestion-date proxies. The study maintains a 183-market no-date-proxy subset for robustness work, but the Copilot market examples in this article use the primary cohort unless otherwise stated.

Only 260 markets have all six eligible platform portfolios for the complete-market analyses. Pair-specific medians use larger or smaller eligible sets depending on the two platforms involved.

The taxonomy was reconciled retrospectively. Candidate repetitions across jobs and shared source ecosystems limit population-level inference.

Domain presence does not verify:

• who owns or controls the domain;
• which exact URL drove the broader domain rank unless URL-level evidence is separately inspected;
• editorial quality;
• whether a source is commercially available;
• whether a citation affected the answer;
• whether a cited company was recommended;
• or whether observed authority will persist.

The deep-exclusive source examples are especially useful for identifying research candidates, but they should not be treated as endorsements. Some domains may be unfamiliar, mixed-purpose, thin, syndicated, transactional, or otherwise unsuitable for marketing activity after human review.

No causal claim is made about why Copilot's source mix differs.

Conclusion

Microsoft Copilot was not simply a lower-overlap outlier in one aggregate table.

Across the strict 260-market complete-platform cohort, Copilot was usually the most isolated source ecosystem, usually contributed the largest unique domain set, and remained materially distinct even in the Top 10 and Top 25 source layers.

ChatGPT was Copilot's closest neighbor in most markets, but the median relationship still left a majority of each Top-100 list unshared.

The most extreme markets were striking. Dental Implants contained 67 Copilot-only Top-100 domains. Legal markets showed a 54% median Copilot-exclusive Top-100 share. A single domain, LegalTechTrend.com, reached Copilot's Top 10 across nine legal categories while missing the other five Top-100 lists in each of those same markets.

Other markets were more connected. Home Builders reached 39.86% Copilot-ChatGPT overlap, and Credit Repair showed meaningful sharing across all six surfaces.

That variation is the point.

A global Copilot average can establish that the platform is different. It cannot tell a CMO or publisher how different Copilot is in the market that matters to them.

The defensible strategy is therefore not six unrelated programs and not one universal source map. It is one coordinated AI visibility system with a separately measured Copilot source layer.

Related Research

• Six AI Engines, Six Different Webs: Comparing Citation Source Overlap Across ChatGPT, Gemini, Perplexity, Copilot, Google AI Overviews, and Google AI Mode
• Can One GEO Strategy Work Across ChatGPT, Gemini, Perplexity, Copilot, Google AI Overviews, and Google AI Mode?
• How Publishers Should Think About Platform-Specific AI Citation Demand
• Should CMOs Treat Microsoft Copilot as a Separate AI Search Channel?
• Which Source Types Overperform in Microsoft Copilot Compared With Other AI Engines?
• Why Microsoft Copilot's Citation Mix Creates Different Publisher and PR Opportunities
• Copilot vs ChatGPT: Why Similar Answers Can Rely on Different Citation Sources
• The Persistence-Portability Gap at Scale
• How to Find the Market-Specific Publishers AI Systems Actually Cite

The earlier Persistence-Portability framework provides conceptual context. The 2026 AI Citation Authority Study covers an earlier high-stakes sample and is a separate population.

About LLM Authority Index

LLM Authority Index tracks how AI systems cite sources, mention companies, and structure commercial answer ecosystems across hundreds of markets. The research program separates citation authority from recommendation authority and focuses on observable source behavior, cross-platform portability, persistence, market breadth, platform breadth, and source depth.