Google AI Overviews and AI Mode Are Siblings, Not Twins. Gemini Is a Second Cousin. Perplexity Is the Family Friend Who Looks Suspiciously Related.

Answer Capsule

Google AI Overviews and Google AI Mode looked like siblings in their citation-source portfolios, but not twins. Gemini looked like a more distant relative. Perplexity looked close enough to the same family pattern that the joke in this article's title is supported by the data more often than a simple corporate-family assumption would suggest.

Across the 283-market benchmark, the median Top-100 Jaccard overlap between Google AI Overviews and Google AI Mode was 53.85%. With two equal Top-100 domain portfolios, that corresponds to 70 shared domains and roughly 30 domains unique to each surface.

Gemini's median overlap with each Google search surface was 30.72%, equivalent to about 47 shared domains. Perplexity's median overlap with each was 29.03%, equivalent to about 45 shared domains. Gemini and Perplexity overlapped with each other at a lower 23.46%, equivalent to about 38 shared domains.

The deeper four-platform analysis makes the metaphor more useful.

Among the 265 markets where Gemini, Perplexity, Google AI Overviews and Google AI Mode all had eligible Top-100 portfolios:

• the median Google AI Overviews/AI Mode intersection contained 70 domains;
• a median 40 of those sibling-shared domains also appeared in Gemini;
• a median 40 also appeared in Perplexity;
• a median 28 domains appeared in all four portfolios;
• the median union across all four surfaces contained 221 distinct domains;
• the median four-way Jaccard intersection-over-union was only 12.67%.

Gemini was usually closer to the two Google search surfaces than Perplexity was, but not overwhelmingly so. On the common 265-market cohort, Gemini had the higher average overlap with AI Overviews and AI Mode in 160 markets, 60.38%. Perplexity was closer in 102 markets, 38.49%, with 3 ties. The median Gemini advantage was only 1.64 percentage points.

That is why “second cousin” and “family friend who looks suspiciously related” are useful editorial shorthand. Gemini usually resembled the Google search surfaces more. Perplexity was close enough to beat Gemini in more than one-third of comparable markets.

The metaphor stops there. This study measures observed registrable-domain citation portfolios. It does not establish model lineage, training data, retrieval systems, source-selection logic, copying, corporate integration, answer similarity, recommendation similarity, or technical architecture.

Key Findings

The “Family” Is a Citation-Source Family, Not a Technical Family

Questions This Section Answers

1. How similar were the citation sources used by Google AI Overviews, Google AI Mode, Gemini and Perplexity?
2. Does citation-source resemblance prove that Google AI products and Perplexity share technical infrastructure?
3. Why is Perplexity described as a “family friend who looks suspiciously related” in this study?

The title is intentionally memorable, but the underlying measure is narrow and explicit.

The study compares the registrable domains appearing in eligible Top-100 citation portfolios. It asks how much the source sets overlap. It does not inspect source-selection code, retrieval stacks, model weights, hidden prompts, training corpora, product architecture, or internal Google or Perplexity systems.

Within that narrow frame, the family resemblance is visible.

The two Google search surfaces are clearly the closest pair. Their median source resemblance is about 23 points higher than Gemini's median relationship with either surface.

But the gap between Gemini and Perplexity is much smaller. Gemini's 30.72% median is only 1.69 points above Perplexity's 29.03% median.

That is the statistical basis for the title.

Gemini looks more related to the Google search surfaces than Perplexity does, but Perplexity is close enough that branding and corporate ownership alone would not accurately predict the observed source portfolios.

The dedicated platform study, Six AI Engines, Six Different Webs, shows where these relationships sit inside the complete 15-pair platform matrix.

Google AI Overviews and AI Mode Are Siblings, but They Still Have Separate Source Portfolios

Questions This Section Answers

1. How many Top-100 citation sources do Google AI Overviews and Google AI Mode typically share?
2. Can a CMO use AI Overview citation visibility as a proxy for AI Mode visibility?
3. How much of each Google search surface's Top-100 source portfolio is usually absent from the other?

A median Jaccard overlap of 53.85% is strong relative to every other platform pair in this benchmark, but it is far from identity.

With two equal Top-100 lists, a 53.85% Jaccard corresponds to:

• 70 domains in both lists;
• 30 domains unique to AI Overviews;
• 30 domains unique to AI Mode;
• 130 total distinct domains across the two lists.

The common four-platform cohort produces almost exactly that structure. Across 265 comparable markets, the median AI Overviews/AI Mode intersection was 70 domains, and the median union was 130.

That is a useful way to interpret “siblings, not twins.”

The shared 70-domain core is large enough to justify coordinated source research. A publisher repeatedly visible on both search surfaces belongs in a shared diligence lane.

The remaining 30 domains on each side are large enough that the surfaces should still be measured separately.

A single “Google AI citation score” would hide those differences. A brand could improve on one surface and remain unchanged on the other. A publisher could be prominent on AI Overviews and peripheral on AI Mode, or the reverse. A source can also appear on both surfaces while ranking very differently or while the two systems cite different URLs from the same domain.

The parent study found an even stronger market-level fact: among the 260 markets where all six platform portfolios were eligible, the AI Overviews/AI Mode pair was the unique highest-overlap pair in every market. That consistency supports the sibling analogy.

It does not support combining the two products into one measurement cell.

The Google Sibling Core Is Where Gemini and Perplexity Look Most Related

Questions This Section Answers

1. How much of the source core shared by Google AI Overviews and AI Mode also appears in Gemini or Perplexity?
2. Are Gemini and Perplexity mainly similar to the Google search surfaces because they share the same central sources?
3. How large is the source set common to all four surfaces?

The four-platform intersection makes the family pattern more concrete.

Across the 265 markets where all four portfolios were eligible, the median Google sibling core contained 70 domains.

A median 40 of those 70 domains also appeared in Gemini.

A median 40 also appeared in Perplexity.

Expressed as a share of the Google sibling core:

• median 58.33% of the sibling core also appeared in Gemini;
• median 56.96% also appeared in Perplexity;
• median 40.58% appeared in both Gemini and Perplexity.

The median number of domains present in all four Top-100 portfolios was 28.

That 28-domain common core sat inside a much larger median union of 221 distinct domains across the four surfaces. The resulting median four-way Jaccard was only 12.67%.

This distinction matters.

Pairwise resemblance can look meaningful while the all-platform reusable core remains small.

A CMO may see 53.85% sibling overlap, roughly 30% Gemini-to-Google overlap, and roughly 29% Perplexity-to-Google overlap and conclude that one publisher list should cover the family. The four-way intersection shows why that shortcut is risky.

The middle market had about 28 domains common to all four systems, not 70, 47, or 45.

The source relationships are nested rather than interchangeable:

1. AI Overviews and AI Mode share a large core.
2. Gemini and Perplexity each pick up a little more than half of that sibling core in the median market.
3. Only about two-fifths of the sibling core appears in both outsiders simultaneously.
4. Hundreds of additional domains populate the nonshared portions of the four portfolios.

That is a much better description of the observed “family” than a single pairwise percentage.

Gemini Usually Looks More Related, but Perplexity Wins Far Too Often to Ignore

Questions This Section Answers

1. Was Gemini consistently closer to Google AI Overviews and AI Mode than Perplexity was?
2. How often did Perplexity's citation portfolio resemble the Google search surfaces more than Gemini's did?
3. How large was Gemini's typical citation-overlap advantage over Perplexity?

A corporate-family assumption would predict Gemini to be decisively closer to the two Google search surfaces.

The observed citation portfolios support the direction of that prediction, but not the word “decisively.”

For each of the 265 common markets, this analysis averaged:

• Gemini's overlap with AI Overviews and AI Mode; and
• Perplexity's overlap with AI Overviews and AI Mode.

Gemini had the higher two-surface average in 160 markets, 60.38%.

Perplexity had the higher average in 102 markets, 38.49%.

The averages tied in 3 markets, 1.13%.

The median difference was only 1.64 percentage points in Gemini's favor.

The middle half of markets ranged from a 1.82-point Perplexity advantage to a 5.01-point Gemini advantage.

The tails were much wider:

• Gemini led by at least 5 points in 67 markets;
• Perplexity led by at least 5 points in 19 markets;
• Gemini led by at least 10 points in 20 markets;
• Perplexity led by at least 10 points in 4 markets.

The same pattern appears when each Google search surface is examined separately on the common cohort.

Against AI Overviews, Gemini had the higher overlap in 158 markets, Perplexity in 97, with 10 ties.

Against AI Mode, Gemini led in 152 markets, Perplexity in 91, with 22 ties.

So “Gemini is the second cousin” describes the typical ordering.

“Perplexity is the family friend who looks suspiciously related” describes the frequency with which Perplexity reaches or exceeds that same level of source resemblance.

Neither statement should be converted into a technical claim about why the resemblance exists.

The Relationship to Both Google Search Surfaces Is Usually Symmetrical

Questions This Section Answers

1. Does Gemini tend to resemble Google AI Overviews much more than Google AI Mode, or vice versa?
2. Is Perplexity's citation relationship with the two Google search surfaces balanced?

Another useful test is whether Gemini or Perplexity strongly favors one Google sibling over the other.

In most markets, the difference is small.

Across the common 265-market cohort, the median absolute difference between Gemini's overlap with AI Overviews and its overlap with AI Mode was only 1.90 percentage points.

For Perplexity, the median absolute difference was 1.80 points.

That symmetry is consistent with the pair medians:

• Gemini / AI Overviews: 30.72%
• Gemini / AI Mode: 30.72%
• Perplexity / AI Overviews: 29.03%
• Perplexity / AI Mode: 29.03%

The equal medians do not mean the individual market values are identical. They are not.

But the overall pattern says that Gemini and Perplexity are generally relating to a Google search-surface cluster, not strongly to one sibling while ignoring the other.

There are exceptions. Gemini's largest observed sibling asymmetry in the common cohort was 8.87 points. Perplexity's largest was 9.79 points.

Those exceptions are another reason to retain separate platform measurement even when the aggregate relationships look balanced.

The Family Resemblance Is Strongest in the Shared Core, Not the Sibling-Specific Fringe

Questions This Section Answers

1. Do Gemini and Perplexity mostly reproduce the sources shared by AI Overviews and AI Mode, or do they also pick up the sources unique to one Google surface?
2. How much of the Google sibling-specific source fringe transfers to Gemini or Perplexity?

The median Google sibling portfolio contained about 30 domains unique to AI Overviews and 30 unique to AI Mode.

Gemini picked up only a minority of those sibling-specific sources:

• median 6 AI-Overview-only domains also appeared in Gemini;
• median 6 AI-Mode-only domains also appeared in Gemini.

That is roughly 19% to 20% of the sibling-specific fringe in the median market.

Perplexity picked up even fewer:

• median 5 AI-Overview-only domains also appeared in Perplexity;
• median 5 AI-Mode-only domains also appeared in Perplexity.

That is roughly 15% to 17% of the sibling-specific fringe.

Compare those rates with the shared sibling core:

• 58.33% of the sibling core appeared in Gemini at the median;
• 56.96% appeared in Perplexity.

The resemblance is therefore concentrated in the sources the two Google search surfaces already agree on.

That finding is strategically useful.

If a publisher appears on both AI Overviews and AI Mode, it has a much better observed chance of also appearing in Gemini or Perplexity than a source that appears on only one of the two Google search surfaces.

This is still a descriptive association. It does not establish that presence on the two Google surfaces causes visibility elsewhere, and it does not predict future citations.

Some Markets Look Like a Family Reunion. Others Barely Look Related.

Questions This Section Answers

1. Which commercial markets had the strongest four-way citation-source core across Gemini, Perplexity, AI Overviews and AI Mode?
2. Which markets showed weak family resemblance even when AI Overviews and AI Mode were relatively similar?
3. Can Perplexity be more Google-like than Gemini in specific markets?

The family metaphor varies dramatically by market.

Hearing Aids: a strong four-way family core

Hearing Aids was one of the strongest examples.

• AI Overviews / AI Mode: 69.49%
• Gemini average overlap with the two Google search surfaces: 47.62%
• Perplexity average: 40.35%
• Google sibling shared domains: 82
• sibling-shared domains also in Gemini: 59
• sibling-shared domains also in Perplexity: 53
• domains in all four: 43
• four-way Jaccard: 23.50%

The four-way source core included domains such as ncoa.org, hearingtracker.com, youtube.com, seniorliving.org, hearadvisor.com, reddit.com, hearusa.com, forbes.com, fda.gov, soundly.com, audicus.com, and audienhearing.com.

That is a market where a shared source-research workstream has obvious value, even though the four platforms still have substantial nonshared inventories.

eSignature Software: Gemini and Perplexity are nearly equally close

eSignature Software produced one of the cleanest “family friend” examples.

• AI Overviews / AI Mode: 65.29%
• Gemini family average: 44.41%
• Perplexity family average: 43.39%
• difference: only 1.02 points
• domains in all four: 44
• four-way Jaccard: 23.53%, the highest observed in the common four-platform cohort

The four-way core included reddit.com, pandadoc.com, docusign.com, zapier.com, signeasy.com, adobe.com, getaccept.com, gartner.com, signwell.com, onespan.com, proposify.com, boldsign.com, and jotform.com.

In this market, corporate ownership would tell a poor story about citation-source resemblance. Perplexity sits almost exactly where Gemini does relative to the two Google search surfaces.

Dental Implants: siblings, but a very small extended family

Dental Implants shows the opposite pattern.

• AI Overviews / AI Mode: 45.99%
• Gemini family average: 13.96%
• Perplexity family average: 5.27%
• Google sibling shared domains: 63
• sibling core also in Gemini: 19
• sibling core also in Perplexity: 9
• domains in all four: only 5
• union across all four: 288 domains
• four-way Jaccard: 1.74%

The five four-way domains were aspendental.com, affordabledentures.com, youtube.com, reddit.com, and healthline.com.

That is an extreme warning against converting pairwise sibling similarity into a four-platform strategy assumption.

Online Betting Sites: Perplexity looks more related than Gemini

Online Betting Sites reverses the normal cousin/friend ordering.

• AI Overviews / AI Mode: 49.25%
• Gemini family average: 20.17%
• Perplexity family average: 32.01%
• Perplexity advantage: 11.84 points
• four-way shared domains: 26

The four-way core included foxsports.com, cbssports.com, youtube.com, rotowire.com, reddit.com, si.com, fanduel.com, sportshandle.com, hardrock.bet, bettingusa.com, actionnetwork.com, espn.com, wikipedia.org, and covers.com.

In this market, the “family friend” looks considerably more related to the Google search surfaces than the “second cousin.”

Invisible Braces: Gemini looks much more related

Invisible Braces produces the reverse.

• AI Overviews / AI Mode: 51.52%
• Gemini family average: 37.54%
• Perplexity family average: 19.05%
• Gemini advantage: 18.48 points, the largest in the common cohort
• domains in all four: 23

This is why the analogy should never replace the market-specific source map.

Final Expense Insurance Shows Why “Shared Google Source” Does Not Mean “Shared Everywhere”

Questions This Section Answers

1. What does the Google-family citation pattern look like inside a specific insurance market?
2. Can a source shared by AI Overviews and AI Mode still be absent from Gemini or Perplexity?

Final Expense Insurance had the strongest Google sibling overlap in the benchmark at 76.99%.

The two Google search surfaces shared 87 of their Top 100 domains.

Yet the wider family core was much smaller:

• 49 of those sibling-shared domains also appeared in Gemini;
• 41 also appeared in Perplexity;
• only 27 appeared in all four.

The four-way core included aflac.com, choicemutual.com, ethos.com, nerdwallet.com, fidelitylife.com, funeraladvantage.com, reddit.com, youtube.com, progressive.com, gerberlife.com, guardianlife.com, policygenius.com, mutualofomaha.com, seniorliving.org, and forbes.com.

But many domains in the Google sibling core did not transfer to Gemini. Examples included colonialpenn.com, cnbc.com, usnews.com, wsj.com, insurancebrokersusa.com, mintcofinancial.com, nylaarp.com, marketwatch.com, statefarm.com, and jrcinsurancegroup.com.

Likewise, sibling-shared domains absent from Perplexity included colonialpenn.com, selectquote.com, trustage.com, westernsouthern.com, protective.com, aaa.com, pinnaclequote.com, insurancebrokersusa.com, mintcofinancial.com, and nylaarp.com.

Gemini also brought in sources outside both Google search Top-100 portfolios, including finalexpensebenefits.org, blakeinsurancegroup.com, finalexpensedirect.com, aafmaa.com, betterwealth.com, elcomutual.com, goodrx.com, and meetfabric.com.

Perplexity brought its own outside-Google set, including consumeraffairs.com, trustpilot.com, insurance.com, and other domains not present in either Google search Top 100 for this market.

The commercial lesson is not that one platform's sources are better. It is that even an unusually tight Google sibling market can require separate publisher and source investigation once Gemini and Perplexity are added.

Credit Repair Shows What a Broadly Reusable Family Core Looks Like

Credit Repair is a more portable example.

• AI Overviews / AI Mode: 56.25%
• Gemini family average: 38.42%
• Perplexity family average: 35.14%
• sibling-shared domains: 72
• sibling core also in Gemini: 49
• sibling core also in Perplexity: 50
• domains in all four: 39
• four-way Jaccard: 18.66%

The four-way core included money.com, reddit.com, thecreditpeople.com, consumerfinance.gov, experian.com, consumeraffairs.com, incharge.org, firstcard.app, youtube.com, badcredit.org, businessinsider.com, bestcompany.com, ftc.gov, equifax.com, and creditrepair.com.

For a CMO, that is the kind of market where a substantial shared publisher-research lane can be rational.

But the nonshared source sets still matter. Gemini included outside-Google sources such as goodfinancialcents.com, creditfirm.net, creditlawcenter.com, tomocredit.com, creditrepairreview.com, and greenpath.com. Perplexity included outside-Google sources such as financebuzz.com, credlocity.com, thecreditreview.com, creditrepairease.com, getoutofdebt.org, and creditsecrets.com.

A shared core can reduce duplicated work. It does not eliminate platform-specific work.

What the Family Metaphor Means for CMOs

Questions This Section Answers

1. Should a CMO combine AI Overviews, AI Mode, Gemini and Perplexity into one GEO channel?
2. How should marketing teams use the shared Google-family source core without missing platform-specific publishers?
3. Which source relationships should be measured together and which should remain separate?

The practical model has three layers.

1. Coordinate AI Overviews and AI Mode source research

They are the most consistently similar pair in the benchmark. Shared publishers and recurring source domains deserve coordinated diligence.

But keep separate platform results, ranks, cited URLs, recommendation observations, and trend lines.

2. Treat Gemini as usually closer, not automatically equivalent

Gemini is the more Google-like outsider in the typical market, but it wins the family-closeness comparison in only about 60% of the common cohort.

That is too weak to justify using Gemini as a proxy for the two Google search surfaces or vice versa.

3. Do not dismiss Perplexity as an unrelated source ecosystem

Perplexity was closer to the two Google search surfaces than Gemini in 102 of 265 comparable markets.

In markets such as Online Betting Sites, Flight Bookings, Certificates of Deposits, Dermatologist Recommended Skin Care Brands, Vitamins and Dietary Supplements, and Job Posting Sites, Perplexity's family resemblance exceeded Gemini's by meaningful margins.

That makes Perplexity a useful comparison surface when identifying sources that may have cross-platform relevance.

4. Build the source plan from the four-way core outward

The median market had only 28 domains shared by all four platforms.

Those domains are natural candidates for the shared source-research lane.

Then layer on:

• Google-sibling-only sources;
• Gemini-specific additions;
• Perplexity-specific additions;
• sources shared by only selected pairs;
• and platform-specific cited URLs within shared domains.

This is consistent with the operating model in Can One GEO Strategy Work Across ChatGPT, Gemini, Perplexity, Copilot, Google AI Overviews, and Google AI Mode?: one coordinated program, not one copied source list.

What Publishers Should and Should Not Claim

A publisher can make bounded evidence statements such as:

• its domain appeared in the Top 100 on both AI Overviews and AI Mode in a defined market and period;
• it also appeared on Gemini or Perplexity in the same market;
• it belonged to a documented four-platform source core;
• or its citation presence differed materially by surface.

A publisher should not turn those observations into unsupported claims such as:

• “Google's AI stack prefers us”;
• “Perplexity uses the same sources as Google”;
• “being cited by AI Overviews causes Gemini citations”;
• “a sponsorship will transfer across all Google AI products”;
• or “the same article will be cited by every platform.”

The dataset does not establish any of those statements.

For advertiser-facing use, the strongest asset is a transparent market-by-platform evidence table with dates, source ranks, cited URLs where available, and explicit definitions.

The companion research, How Publishers Should Think About Platform-Specific AI Citation Demand, develops that commercial reporting framework.

Methodology and Limits

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

The six benchmark surfaces are ChatGPT, Gemini, Perplexity, Microsoft Copilot, Google AI Overviews and Google AI Mode. API copies and experimental surfaces are excluded.

Platform portfolios pool completed observations across the July through September window. Portfolio rank counts each domain once per retained response, sorts primarily by response presence and breaks ties alphabetically. Eligible Top-100 platform cells require at least 100 retained responses and at least 100 resolved domains. Thin or missing cells are excluded rather than assigned zero.

Pairwise overlap uses Jaccard intersection divided by union. The full pair medians use all markets eligible for each individual pair, so support differs slightly:

• AI Overviews / AI Mode: 283 markets;
• Gemini / AI Overviews and Gemini / AI Mode: 271;
• Perplexity / AI Overviews and Perplexity / AI Mode: 272;
• Gemini / Perplexity: 265.

The deeper family analysis uses the common 265-market subset where all four relevant Top-100 portfolios were eligible. This avoids comparing Gemini and Perplexity using different market populations.

The four-way Jaccard is calculated as the number of domains present in all four portfolios divided by the number of distinct domains present in at least one of the four portfolios. It is not the average of the six pairwise overlaps.

This is observational evidence from changing samples, not a certified identical-prompt panel. Of the primary responses, 104,071 used ingestion-date proxies. Shared source ecosystems, retrospective taxonomy, prompt variation, changing answer conditions and candidate cross-job repetitions limit population inference.

Most importantly, source overlap does not reveal why the overlap exists.

The study does not identify:

• platform retrieval architecture;
• model lineage;
• training data;
• source-selection algorithms;
• hidden search integrations;
• copying between platforms;
• causal effects of content or PR;
• source quality;
• recommendation influence;
• traffic;
• conversions;
• or future citation performance.

“Sibling,” “second cousin,” and “family friend” are editorial metaphors for observed source-set resemblance only.

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?
• Which AI Platforms Share the Most Citation Sources, and Which Diverge the Most?
• How Publishers Should Think About Platform-Specific AI Citation Demand
• The Persistence-Portability Gap at Scale
• Persistence vs Portability: Why a Source That Wins on ChatGPT May Not Transfer to Gemini or Google AI
• How to Find the Market-Specific Publishers AI Systems Actually Cite
• AI Citation Authority vs AI Recommendation Authority: Why Being a Source Is Not the Same as Being Recommended

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.