How to Find the Market-Specific Publishers AI Systems Actually Cite

Answer Capsule

The fastest way to find the publishers and third-party sources AI systems actually use in a commercial market is to work from the market outward, not from a global citation leaderboard inward.

Across the 260 markets in this study with complete eligible Top-100 portfolios on all six AI surfaces, the median market contained 310 distinct domains across the six platform lists. That raw universe is far too large to treat as a practical outreach list.

A useful shortlist emerges when the market is filtered by cross-platform recurrence and source depth:

• median 126 domains appeared in the Top 100 on at least two platforms;
• median 73 appeared on at least three;
• median 46 appeared on at least four;
• median 28 appeared on at least five;
• median only 14 appeared on all six.

Depth narrows the field further. The median market had 68 domains that reached a Top 25 somewhere, but only 22 that reached a Top 25 on at least three platforms and only 3 that reached a Top 25 on all six. At the Top-10 level, the median market had 28 distinct domains that reached a Top 10 somewhere, only 9 that reached a Top 10 on at least three platforms, 5 on at least four, and 1 on all six.

Time adds another filter. Among the 246 markets with all three months and all six platforms represented, the median market had 70 domains that remained in the monthly Top 100 across July, August, and September, but only 20 that remained in the Top 25 in all three months. Of those persistent Top-25 sources, a median 18 appeared in the Top 100 on at least four platforms, and 10 appeared on all six.

That is the practical discovery funnel:

market relevance → platform recurrence → source depth → temporal persistence → URL-level relevance → publisher and commercial verification

The final two steps are essential. The data ranks domains, not audited publishers. A domain can belong to a publisher, company, government organization, marketplace, community, software vendor, retailer, or unrelated source that surfaced because of prompt or taxonomy ambiguity.

The Stairlifts market illustrates the problem. It contained legitimate market-specific sources such as Lifeway Mobility, NCOA, 101 Mobility, Stannah, Harmar and Modernize, but the Top-100 portfolios also contained Ticketmaster, SeatGeek, BrunoMars.com and Concerts50. Those appearances are real observations in the stored source portfolios. They are not evidence that a stairlift marketer should buy media from a concert-ticket site.

A market-specific AI source audit therefore needs both quantitative ranking and human diligence.

Key Findings

These are medians across the 260 markets with six eligible platform portfolios. They are not global publisher counts and should not be added across markets.

Start With the Commercial Market, Not a Global Publisher List

Questions This Section Answers

1. How should a CMO find the websites AI systems actually cite in a specific commercial market?
2. Why can a global AI citation leaderboard miss important category-specific publishers and source domains?
3. How many potential source domains does a typical six-platform market audit surface before filtering?

The first mistake in publisher discovery is starting with the world's most-cited domains and asking which ones might matter to a category.

The evidence suggests the opposite workflow.

Start with the commercial market. Then identify the source portfolios actually observed within that market.

The reason is structural. The parent study found 308,807 distinct registrable domains across the 283-market benchmark. 192,753 of them, 62.42%, appeared in only one market. Even inside the stricter Top-100 source portfolios, 24,381 of 36,273 domains, 67.22%, appeared in only one market.

A global leaderboard therefore favors broadly recurring domains. It can reveal important generalist sources, but it compresses away the exact thing a category marketer often cares about: which domains have unusual authority inside this buying decision?

The companion study, Why Niche Publishers Matter More in AI Search Than Global Citation Rankings Suggest, found 7,719 domain-market pairs where a source reached a Top 10 on at least one platform. Nearly half of those local Top-10 authority relationships came from domains with global Top-100 breadth below 25 markets.

That means category-specific discovery should be treated as a local-ranking problem, not merely a global-reputation problem.

The raw local universe is still large. Across the 260 complete six-platform markets, the median market had 310 distinct domains across the six Top-100 lists. The middle half ranged from 298 to 332 domains, with observed markets ranging from 266 to 421.

A 310-domain list is a research universe, not an outreach plan.

The rest of the process is about turning that universe into a defensible shortlist.

Use Cross-Platform Recurrence as the First Shortlist Filter

Questions This Section Answers

1. How many AI citation sources typically recur across two, three, four, five, or six platforms inside the same market?
2. Should a marketer prioritize domains that appear on several AI platforms over sources that appear on only one?
3. How can a CMO reduce a 300-domain AI citation universe to a manageable source shortlist?

The cleanest first filter is platform recurrence within the market.

For each complete market, count how many of the six eligible Top-100 platform portfolios contain each domain.

The median market funnel looks like this:

This is already a useful operational reduction.

A marketing team does not have to manually investigate all 310 domains equally. It can create tiers:

• Tier 1: domains present on all six platforms;
• Tier 2: domains present on four or five;
• Tier 3: domains present on two or three;
• Tier 4: single-platform sources that may still matter locally.

That last tier should not be discarded automatically. Platform-specific authority is real. The Persistence vs Portability analysis shows why a source that matters to one AI system may fail to transfer to another.

The recurrence filter is therefore not a quality score. It is a way to organize investigative effort.

A six-platform source belongs in a different diligence lane from a one-platform specialist, but both can be commercially relevant.

Then Add Depth, Because Presence Is Not Prominence

Questions This Section Answers

1. How should a CMO distinguish an incidental AI citation source from a genuinely prominent market source?
2. How many domains typically reach the Top 10 across multiple AI platforms inside one market?
3. Why should a market-specific publisher audit track Top-10 and Top-25 depth separately from Top-100 presence?

A source can appear in a Top-100 portfolio and still be peripheral to the market.

Depth is the second major filter.

Across the 260 complete six-platform markets, the median market had:

• 68 domains reaching a Top 25 on at least one platform;
• 22 reaching a Top 25 on at least three platforms;
• 3 reaching a Top 25 on all six.

At the Top-10 level, the median market had:

• 28 distinct domains reaching a Top 10 somewhere;
• 14 reaching a Top 10 on at least two platforms;
• 9 on at least three;
• 5 on at least four;
• 3 on at least five;
• 1 on all six.

The difference between a 310-domain Top-100 union and a 28-domain Top-10 union is enormous.

For publisher discovery, this supports a simple rule:

Start broad enough to avoid missing specialists, then use depth to identify which sources actually occupy leading positions.

This is different from saying that a Top-10 source is always a better media partner. The ranking is based on response presence in the observed AI source portfolio. It does not measure editorial quality, audience, traffic, sponsorship policy, price, conversion value, or recommendation influence.

But it tells the marketer where to investigate first.

The companion analysis Citation Breadth vs Citation Depth: Which AI Authority Metric Should CMOs Care About? shows why the same global breadth can conceal radically different local depth.

Add Time Before Calling a Source Durable

Questions This Section Answers

1. How many AI citation sources typically remain in a market's Top 100 across three consecutive months?
2. How can marketers distinguish a durable source from a one-month AI citation spike?
3. Which persistent Top-25 sources also tend to appear across several AI platforms?

A source shortlist becomes much more useful when the time dimension is added.

Among the 246 markets with all three months and all six platform portfolios represented, the median market had:

• 70 domains present in the monthly Top 100 in July, August, and September;
• 20 domains present in the monthly Top 25 in all three months;
• 18 of those persistent Top-25 domains also appearing in the Top 100 on at least four AI platforms;
• 10 of those persistent Top-25 domains appearing in the Top 100 on all six platforms.

This is a powerful shortlist layer because it separates three different source types:

1. durable cross-platform sources, which recur over time and appear broadly across AI surfaces;
2. durable platform-specific sources, which recur over time but remain concentrated on selected systems;
3. episodic sources, which may appear strongly in one month without becoming part of the recurring source core.

The first category is attractive for broad diligence because it combines temporal recurrence with platform reach.

The second category is not inferior. It can be highly valuable when the company cares about a particular AI surface.

The third deserves caution before money or outreach effort is committed.

The detailed persistence evidence is documented in How Stable Are AI Citation Sources Month to Month? and Top 25 vs Top 100 AI Citation Sources: Where Durable Authority Actually Lives.

Hearing Aids: A Clean Example of a Market-Specific Source Map

Questions This Section Answers

1. Which websites were the strongest cross-platform AI citation sources in the Hearing Aids market?
2. How can a marketer identify specialist hearing-aid sources that compete with broad national publishers in AI citations?
3. What does a six-platform market-specific publisher shortlist look like in practice?

Hearing Aids provides a relatively clean example of the method.

Across its six eligible Top-100 platform portfolios, the market contained:

• 267 distinct domains;
• 133 appearing on at least two platforms;
• 83 on at least three;
• 59 on at least four;
• 37 on at least five;
• 21 on all six.

The Top-10 layer was much tighter:

• 25 domains reached a Top 10 somewhere;
• 14 reached a Top 10 on at least two platforms;
• 10 on at least three;
• 6 on at least four;
• 2 on all six.

Those two six-platform Top-10 sources were NCOA.org and HearingTracker.com.

HearingTracker's platform-specific evidence is particularly strong:

NCOA.org was also Top 10 on all six platforms, with ranks ranging from 1 to 7.

The next layer included Forbes, SeniorLiving.org, Consumer Reports, HearAdvisor and Soundly, but their platform profiles differed materially.

Forbes reached the Top 10 on five platforms but fell to rank 19 on Perplexity. SeniorLiving.org reached a Top 10 on four platforms. Consumer Reports appeared in five eligible platform Top-100 portfolios and reached a Top 10 on four. HearAdvisor appeared on all six platforms and reached a Top 10 on three. Soundly appeared on all six and reached a Top 10 on two.

This is what a useful market-specific shortlist looks like: not one winner, but a layered source ecosystem that includes broad generalists, category specialists and platform-specific strengths.

The data alone does not tell a marketer which of these organizations sells sponsored content, accepts contributed content, has relevant audiences, maintains editorial independence, or is commercially appropriate. Those are the next diligence questions.

Air Purifiers Shows Why the Platform Columns Must Stay Separate

Questions This Section Answers

1. Which sources had strong AI citation depth in the Air Purifiers market across different platforms?
2. Can a source rank first on Google AI or Perplexity but much lower on ChatGPT or Copilot?
3. Why should an AI publisher shortlist preserve separate ranks for each platform instead of using one blended score?

Air Purifiers shows why collapsing platform ranks into a single authority score can hide useful information.

The market contained 316 distinct Top-100 domains across six platforms, but only 11 appeared on all six. Twenty-five domains reached a Top 10 somewhere, and only one reached a Top 10 on all six.

That all-six Top-10 domain was Forbes.

But several category-specific sources were extremely strong while showing large rank differences across platforms.

HouseFresh.com is the clearest example:

A blended global or even blended market score could obscure that shape.

HouseFresh was a leading source for Google AI Overviews, Google AI Mode and Perplexity, strong on Gemini, but much weaker in the Copilot portfolio.

Other locally important sources included RTINGS, AirPurifierFirst, Consumer Reports, Good Housekeeping and TechGearLab.

AirPurifierFirst ranked 1 on Gemini, 3 on Perplexity, 10 on Google AI Mode, but 46 on Copilot. RTINGS ranked 1 on ChatGPT and 2 on Perplexity, but 13 on Gemini.

This is not noise to be averaged away. It is exactly the information a platform-specific publisher strategy needs.

Life Insurance Shows the Difference Between a Shared Core and Platform-Specific Opportunity

Questions This Section Answers

1. Which sources formed the strongest cross-platform AI citation core in the Life Insurance Companies market?
2. How can marketers distinguish a shared AI citation core from platform-specific life-insurance sources?
3. Why should life-insurance publisher research track both specialist and generalist sources?

The Life Insurance Companies market contained 351 distinct domains across its six Top-100 platform portfolios, but only 7 appeared on all six.

Twenty-seven domains reached a Top 10 somewhere. Only one domain reached a Top 10 on all six platforms: Policygenius.

Policygenius ranked:

• 6 on ChatGPT;
• 4 on Gemini;
• 1 on Google AI Mode;
• 1 on Google AI Overviews;
• 8 on Copilot;
• 5 on Perplexity.

Its response presence ranged from 34 of 500 retained Copilot responses to 480 of 800 Google AI Overview responses.

The next layer was more fragmented:

• Reddit reached a Top 10 on five platforms but ranked 100 on Copilot;
• Ethos reached a Top 10 on five but ranked 69 on Perplexity;
• NerdWallet reached a Top 10 on five platforms but did not appear in Gemini's Top 100;
• MoneyGeek reached a Top 10 on four, but ranked 45 on Google AI Mode and 37 on Google AI Overviews;
• Forbes reached a Top 10 on three;
• Insure.com reached a Top 10 on three;
• Money.com reached a Top 10 on three and appeared in five platform Top-100 portfolios, but not Copilot's.

That is the kind of source architecture a CMO needs to see before deciding where to invest attention.

A list that only says “Policygenius, NerdWallet, Forbes, Money.com and Reddit are important life-insurance sources” is directionally useful but operationally incomplete.

The platform columns tell the marketer where each source is strong.

The persistence columns tell the marketer whether that strength lasts.

The URL review tells the marketer what kind of content is actually being cited.

Only after those steps should commercial outreach begin.

Job Posting Sites Shows How a Market Can Have Both Universal and Platform-Specific Leaders

Questions This Section Answers

1. Which sources were cited most consistently across AI platforms for job-posting and job-search questions?
2. How do Indeed, ZipRecruiter, LinkedIn, Glassdoor, Money.com and FlexJobs differ in AI citation-platform coverage?
3. What can a job marketplace learn from platform-specific source rankings before planning publisher or PR outreach?

Job Posting Sites had 312 distinct domains across the six Top-100 lists. Ten appeared on all six.

At the Top-10 level, the market had 27 distinct sources, with 2 appearing in the Top 10 on all six platforms.

Those two were Indeed and ZipRecruiter.

Indeed ranked 1, 2, 3, 2, 2 and 5 across ChatGPT, Gemini, Google AI Mode, Google AI Overviews, Copilot and Perplexity respectively.

ZipRecruiter ranked 9, 7, 2, 3, 5 and 4.

The next layer illustrates why platform specificity matters:

• Money.com reached a Top 10 on five platforms but ranked 12 on Copilot;
• LinkedIn reached a Top 10 on five but ranked 49 on Gemini;
• Glassdoor reached a Top 10 on four platforms but did not appear in Gemini's Top 100;
• FlexJobs reached a Top 10 on three, but ranked 83 on Copilot;
• G2 reached rank 6 on Perplexity but 87 on Gemini;
• Forbes ranked 8 on ChatGPT but 84 on Gemini.

This is not a stable one-dimensional hierarchy.

It is a market-specific source map with a shared core and platform-specific edges.

Stairlifts Demonstrates Why Human Relevance Review Is Non-Negotiable

Questions This Section Answers

1. Why can an AI citation-source shortlist contain domains that are not commercially relevant to the intended market?
2. What should marketers manually verify before treating a highly ranked AI citation domain as a publisher opportunity?
3. How can prompt or taxonomy ambiguity affect a market-specific AI source audit?

The biggest risk in an automated source-discovery system is assuming that every high-ranking domain belongs in the final publisher shortlist.

The Stairlifts market demonstrates why that assumption is unsafe.

The market contained 320 distinct Top-100 domains, with 11 appearing on all six platforms. Thirty domains reached a Top 10 somewhere, but none reached a Top 10 on all six.

Several results were clearly aligned with the commercial category:

• Modernize appeared on all six platforms and reached a Top 10 on five;
• Lifeway Mobility appeared on all six and reached a Top 10 on four;
• NCOA appeared on all six and reached a Top 10 on four;
• 101 Mobility appeared on all six and reached a Top 10 on three;
• Stannah and Harmar each appeared on all six and reached a Top 10 on three.

But the same Top-100 universe also contained domains such as:

• Ticketmaster;
• SeatGeek;
• BrunoMars.com;
• Concerts50;
• Live Nation.

Ticketmaster, for example, ranked 3 on ChatGPT, 4 on Gemini and 7 on Perplexity in the stored Stairlifts platform portfolios.

SeatGeek ranked 6 on ChatGPT, 6 on Gemini and 3 on Perplexity.

Those are real observations from the stored ranking data. They are not evidence that these domains are relevant publishers for a stairlift marketing campaign.

The correct response is not to delete inconvenient results silently. It is to investigate them.

The analyst should open the cited URLs, inspect the prompts or answer context where available, determine whether the domain is semantically relevant, and document whether the source belongs in the final commercial shortlist.

This is also why the dataset should be described as a source-domain discovery system before it is described as a publisher database.

The data can identify where AI systems cited domains. Human review is still required to determine what those domains represent.

A Six-Step Market-Specific Publisher Discovery Workflow

Questions This Section Answers

1. What is a repeatable process for turning AI citation data into a market-specific publisher shortlist?
2. Which filters should a marketing team apply before contacting publishers found through AI citation research?
3. How should platform coverage, source depth and persistence be combined without creating an arbitrary authority score?

A practical workflow can be built without inventing a composite score.

Step 1: Define the market and buying intent

State exactly what commercial decision the source map is supposed to support.

“Insurance” is not the same market as “life insurance,” “final expense insurance,” “travel insurance,” or “long-term care insurance.”

“Home services” is not the same market as “roofing companies,” “stairlifts,” or “emergency plumbers.”

If the market definition is too broad, the source list will mix unrelated intents.

Step 2: Build separate platform portfolios

For each priority AI system, rank domains by distinct response presence rather than raw within-answer citation repetition.

Keep:

• platform;
• retained response denominator;
• response count containing the domain;
• rank;
• Top-10, Top-25 and Top-100 status.

Do not average away the platform differences at this stage.

Step 3: Build the market-level recurrence tiers

Count how many platforms include each source.

Start with the all-six and four-to-five-platform sources for broad diligence, but preserve platform-specific domains as a separate lane.

The median market provides a useful scale reference:

• 310 total Top-100 domains;
• 126 appearing on at least two platforms;
• 73 on at least three;
• 46 on at least four;
• 28 on at least five;
• 14 on all six.

Step 4: Add depth

Flag whether each source reaches the Top 25 or Top 10 on each platform.

A source present on all six but never reaching a Top 25 tells a different story from one that ranks in the Top 10 on four systems.

The median market has only 9 sources reaching a Top 10 on at least three platforms. That is a much more manageable manual review list.

Step 5: Add persistence

Where repeated collection exists, determine whether the source recurs over time.

A source that ranks well in one month but disappears in the next two should not be described the same way as one that remains part of the market's durable core.

Do not substitute global persistence for source-level persistence. The market-level persistence metric describes portfolio stability, while the source audit should directly record whether the specific domain recurs.

Step 6: Perform URL, ownership and commercial diligence

Before calling a domain a publisher opportunity, verify:

• the exact cited URLs;
• the subject of those pages;
• ownership and publisher identity;
• editorial versus company-controlled content;
• whether the cited content is current;
• whether the site accepts advertising or sponsored content;
• whether sponsorship is clearly separated from editorial rankings;
• audience relevance;
• pricing and commercial terms;
• conflicts of interest;
• whether the source is actually reachable for the proposed partnership.

That final step converts a source map into a commercial prospect list.

The Source Table Should Preserve the Evidence, Not Just the Rank

Questions This Section Answers

1. Which columns should be included in a market-specific AI publisher research table?
2. How can a CMO preserve enough evidence to audit why a publisher was added to an AI source shortlist?
3. Which fields prevent AI citation research from becoming an unreviewable black-box authority score?

A useful source table should be auditable.

The most important field may be source type.

The underlying study deliberately does not classify every domain as a publisher. Doing that automatically would overstate what the data proves.

The commercial workflow should make that classification explicitly and preserve who made it and why.

Do Not Use One Composite Score to Hide the Tradeoffs

Questions This Section Answers

1. Should AI publisher discovery combine rank, breadth, persistence and platform coverage into one authority score?
2. Why can an opaque composite AI authority score mislead publisher selection?
3. How should CMOs compare broad generalist sources with deep category specialists?

The temptation is to turn every metric into one number.

That makes dashboards simpler, but it can make decisions worse.

Consider the different source profiles already visible in this study:

• HearingTracker is extremely deep inside Hearing Aids and reaches the Top 10 on all six platforms, but it is not globally broad across dozens of unrelated markets.
• Forbes is globally broad and often locally deep, but its rank still varies materially by market and platform.
• HouseFresh is highly prominent on several Air Purifier platforms and weak on Copilot.
• Money.com is strong in several finance and employment-related markets but does not occupy the same platform position everywhere.
• Lifeway Mobility is highly prominent on Google AI in Stairlifts but much weaker on ChatGPT and Perplexity.

None of these profiles can be reduced to “better” or “worse” without defining the business decision.

For a national multi-category advertiser, breadth may matter more.

For a stairlift company, local category depth may matter more.

For a company focused on Google AI Overviews, platform-specific prominence may matter more than six-platform reach.

For a publisher sponsorship program, persistence may matter more than a one-time rank spike.

The metrics should remain visible rather than being buried inside arbitrary weights.

What a CMO Should Ask Before Buying Anything

Questions This Section Answers

1. What questions should a CMO ask before buying sponsored content from a publisher found through AI citation data?
2. How can a marketing team separate observed AI citation authority from claims about future recommendation or ranking impact?
3. What evidence is missing from citation data alone when evaluating a publisher partnership?

A source's AI citation footprint can justify further diligence. It cannot justify a media buy by itself.

Before spending money, ask:

1. Is the publisher strong in our actual market?
2. Which platforms cite it?
3. How deep is that citation presence?
4. Does the source recur across multiple collection periods?
5. Which URLs are being cited, and what topics do they cover?
6. Does the publisher reach our customers outside AI systems?
7. Can the publisher produce content relevant to the buyer questions where it already has authority?
8. Is paid content clearly separated from editorial rankings and research?
9. What conventional distribution, referral or audience value accompanies the placement?
10. How will we measure brand citation, recommendation and traffic separately after publication?

The study does not contain a sponsorship experiment. It does not establish that buying content on a cited publisher will cause an AI system to cite the new page, recommend the advertiser, increase rankings, drive traffic or create revenue.

The defensible use of the data is narrower and more valuable:

It identifies which source domains deserve investigation because AI systems already use them in the commercial market being studied.

The separate article How CMOs Should Evaluate Publishers Before Buying Sponsored Content, PR, or Distribution extends that diligence into a buying framework.

Methodology

The research 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 consumer AI surfaces are:

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

API copies and experimental surfaces are excluded from the strict benchmark.

Platform portfolios

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

Strict Top-100 eligibility requires:

• at least 100 retained responses;
• at least 100 resolved registrable domains;
• an actual ranked list of 100 domains.

Thin or missing cells are excluded rather than assigned zero.

The practical six-platform funnel in this article uses the 260 markets with all six eligible platform Top-100 portfolios.

For each market, domains were counted by the number of platform Top-100 portfolios in which they appeared. The same source sets were used to count Top-25 and Top-10 appearances by platform.

Time-persistence filter

The durability subsection uses the 246 markets with all three monthly portfolios and all six platform portfolios represented in the marginal source sets.

For each market, the July, August and September Top-100 domain sets were intersected. The same was done for the Top 25. Persistent Top-25 domains were then compared with the six platform Top-100 portfolios to count how many appeared on at least four platforms and on all six.

These counts describe source recurrence in the stored portfolios. They do not establish a controlled identical-prompt panel.

Response presence

The platform case studies report each domain's rank and the number of retained responses containing the domain, together with the platform-market response denominator.

That is different from raw citation occurrences. A domain cited multiple times within one response still contributes one response presence to the rank calculation.

Limitations

This article is designed as a discovery method, not a publisher-quality rating system.

Several limitations matter.

Domains are not publishers

A registrable domain can represent a publisher, company, marketplace, government entity, community, social platform, retailer, nonprofit, software vendor or other organization.

The study does not automatically classify ownership or editorial model.

Market labels can still contain semantic or taxonomy ambiguity

The Stairlifts example demonstrates why apparently irrelevant source domains must be investigated rather than silently removed.

Historical market taxonomy is retrospective, and the broader research package documents alias and mapping concerns. A high-ranked domain is an observed source, not automatic proof of category relevance.

The samples are observational

The historical collection does not certify identical prompts, prompt versions, model versions, geographies, retrieval settings or collection configurations across every month and platform.

The source portfolios describe observed samples rather than a controlled experiment.

Domain-level ranking hides URL-level variation

Two platforms can cite the same domain while citing different articles. A publisher can have one highly influential page without having broad authority across its entire site.

URL inspection is mandatory before commercial conclusions are drawn.

Citation authority is not recommendation authority

A source can be cited while a competitor is recommended. A company can be recommended without its own domain being cited.

Publisher-source authority and company recommendation behavior need separate measurement. The distinction is developed in AI Citation Authority vs AI Recommendation Authority.

Source presence does not predict campaign performance

The study contains no controlled media-buy experiment, no verified sponsorship lift model and no causal estimate connecting publisher placement to AI ranking, traffic, conversions or revenue.

A source map improves targeting and diligence. It does not guarantee results.

Conclusion

Finding the publishers AI systems actually cite is not a matter of downloading a global “Top AI Sources” list.

A real market-specific source audit begins with hundreds of domains and then narrows them systematically.

In a typical complete six-platform market in this benchmark:

• about 310 domains appeared somewhere in the six Top-100 portfolios;
• 126 appeared on at least two platforms;
• 73 on at least three;
• 46 on at least four;
• 28 on at least five;
• only 14 on all six.

Depth shrinks the list further. The median market had only 9 domains reaching a Top 10 on at least three platforms and 5 reaching a Top 10 on at least four.

Time adds another layer. In the balanced 246-market subset, the median market had 20 sources remaining in the monthly Top 25 across all three months, with 10 of those also appearing in the Top 100 on all six platforms.

That gives marketers a practical research funnel:

1. define the commercial market;
2. build platform-specific source portfolios;
3. measure cross-platform recurrence;
4. add Top-25 and Top-10 depth;
5. add month-to-month recurrence;
6. inspect cited URLs;
7. classify source type and ownership;
8. verify commercial relevance before outreach.

The Hearing Aids, Air Purifiers, Life Insurance and Job Posting examples show how this process reveals both broad generalists and category specialists.

The Stairlifts example shows why the process cannot be fully automated.

The goal is not to produce the longest possible list of cited domains.

It is to identify the smallest defensible set of sources that actually matter to the market, platform and buying decision being studied.

That is the point where AI citation data becomes commercially useful research rather than another leaderboard.

Related LLM Authority Index Research

• AI Search Has a Tiny Universal Core and a Massive Specialist Long Tail is the parent market-breadth study.
• Why Niche Publishers Matter More in AI Search Than Global Citation Rankings Suggest examines local authority hidden by global breadth rankings.
• Citation Breadth vs Citation Depth: Which AI Authority Metric Should CMOs Care About? separates coverage from prominence.
• How Publishers Can Build Category Authority Without Becoming a Generalist Media Site examines the publisher-side implications of category-specific authority.
• The Persistence-Portability Gap at Scale provides the methodological anchor for temporal stability and platform fragmentation.
• Persistence vs Portability: Why a Source That Wins on ChatGPT May Not Transfer to Gemini or Google AI focuses on platform transfer.
• Six AI Engines, Six Different Webs provides the full six-platform overlap comparison.
• How CMOs Should Evaluate Publishers Before Buying Sponsored Content, PR, or Distribution extends source discovery into commercial diligence.
• Which AI Citation Metrics Belong on a CMO Dashboard? translates the research dimensions into reporting fields.

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 a broad source footprint is not mistaken for deep market influence.