# How to Count Citations by Page Domain and Organization By AI Visibility editorial team AI-assisted article. A citation report can say an AI answer linked to six pages, five hostnames, four domains and two known organizations. All four numbers can be accurate for the same answer. The practical question is what each count lets you decide. Page counts identify which documents earn citations. Domain counts reveal how widely links are distributed across sites. Organization counts tell you whether those sites represent several publishers or a single operator with several web properties. For a monthly AI visibility report, keep the raw links, calculate the different totals from the same capture, and label both the counting unit and the reporting window. You can then explain a change in the numbers without guessing whether the change came from new documents, more domains or a different mix of publishers. Keep one row per observed citation link The capture log records what the interface showed before you apply any grouping rules. Give each answer a run ID and save the question, interface, date and time, market, and any visible link label or source panel placement. On each citation row, record the visible URL and its position. If a panel groups several links under one citation marker, retain the marker ID so a future reviewer can reconstruct the grouping. A single marker may reveal several clickable links. Decide whether your raw count describes markers or links and label it accordingly. Retain the displayed URL even when you follow it to another address. Add separate fields for the resolved URL, resolution status and normalized page key. If a link produces an error or the destination cannot be established, leave a reviewable unresolved row. You still saw the citation. Its destination may be unavailable for page-level comparison, and the raw count should preserve that distinction. A useful capture log also separates citations attached to the answer from incidental navigation. For example, the links in a sources panel may form the citation set, while account controls, sharing links and interface help sit outside it. Write down what you included before calculating totals. Keep the same capture boundary in successive reporting periods. If the interface changes and the earlier boundary cannot be reconstructed, date the change and explain which trend lines it affects. Count pages after applying a written URL rule Use a URL parser to split a destination into its components. Then apply a short normalization policy. Start with the parser’s normalized hostname and scheme representation. For ordinary HTML pages, you can drop the fragment when your unit is the document, while retaining it as a field for readers who need the cited section. Keep fragments when the site uses them to identify distinct content and you can verify that behavior. Store the page key alongside the source URL so another reviewer can see how it was derived. WHATWG URL Standard (https://url.spec.whatwg.org/) Query strings require care. Removing a known campaign parameter can prevent a tracked copy of a link from looking like a second page. Removing every query parameter can collapse separate documents, language variants or listings that the site selects through the query. Preserve unknown parameters by default. Keep paths with different case or different trailing slash behavior separate until you establish how that site serves them. If two addresses redirect to the same verified page, a destination-based metric can group them while retaining both raw placements. Count distinct page keys once per answer for a run-level measure. Suppose one answer shows the same guide twice and a second guide once. That gives you three raw links and two page placements for the run. If the first guide appears again in a later answer, it earns another run-level page placement. Across the full reporting window, it is still one unique page. Those two metrics answer different questions. One measures recurrence and the other measures breadth. Keep hostnames and registrable domains separate A hostname describes the address the link leads to. A registrable domain is a grouping boundary derived from the applicable public suffix. For example, the two hosts guides.harbor.example and harbor.example belong to the same illustrative domain, harbor.example. Two sites beneath a suffix such as co.uk can represent separate registrable domains even though the last two labels match. Use a maintained Public Suffix List implementation rather than splitting hostnames at the final two dots. Record the library or list version and how it handles the private section of the list. Public Suffix List (https://publicsuffix.org/) Hosted publishing platforms make that last decision visible. Tenant addresses can have distinct editorial owners even when they share infrastructure. A domain report built with private suffix rules can group them differently from one built with ICANN rules alone. State your choice in the method note and retain host-level counts so someone can reproduce the published total. If a URL points to an IP address or a host the implementation cannot classify, give it an explicit unresolved or special status for the domain metric. Avoid reading domain breadth as independent corroboration. Two hosts may serve the same editorial operation. Conversely, contributors to one domain may speak for different organizations. Start from the domain table when you want to see which web properties receive links. Move to the organization register when the decision depends on who operates or publishes the cited material. Map domains to organizations with ownership evidence Choose a consistent definition of organization. For a source-diversity report, the operator responsible for publishing the page is a workable choice. A directory is the publisher of its listing page, while the business named in that listing is the subject. A parent-company metric would answer a different question, so keep parent relationships in a separate field if you want that view as well. Give verified publishers stable IDs. Record the mapping basis and the date checked. An operator’s official about page, legal notice or other direct ownership disclosure may support grouping more than one domain. Structured identity links can help you locate supporting evidence. For example, Schema.org defines sameAs as a reference URL that unambiguously indicates the item’s identity. Treat the presence of a markup field as a lead and read the linked page before applying the mapping. Schema.org sameAs (https://schema.org/sameAs) Use page-level exceptions when a domain hosts more than one publisher under your definition. For example, a directory listing written and published by a directory belongs to the directory in a publisher count. A platform hosting independently operated tenant publications can need a separate mapping by tenant or page. Write that scope into the register so a rule intended for one publication does not attach every URL on the platform to it. An official Instagram account offers a concrete version of the same distinction. An organization can publish through its own website and its account on a shared platform. If your method treats the account owner as the publisher and you verify the identity, the two destinations can contribute to one organization count. The destination still belongs to Instagram for domain reporting. The Home Box UAE post below illustrates the publication channel. It sits outside the worked example and the historical study counts. View the selected official Instagram post (https://www.instagram.com/homebox_ae/p/DejgCPQNlMe/) Official Home Box UAE post, previously verified against its regional account. Work through a two-answer example This constructed example uses reserved illustrative domains and invented publishers. In Run A, an answer links twice to the same Harbor guide. One occurrence has a known tracking parameter. The answer also links to a second page on the same hostname, another Harbor page on a subdomain, a trade page on a separately mapped Harbor domain, a directory listing about Harbor and a page whose publisher is unresolved. In Run B, a second answer links to the original guide and the same directory listing. For this fixture, count pages after stripping the known tracking parameter, calculate domains using the illustrated suffix boundary and assign organization IDs only where the mapping is verified within the example. Unit Run A Run B Sum across runs Distinct across window Raw link appearances 7 2 9 — Normalized pages 6 2 8 6 Hostnames 5 2 7 5 Registrable domains 4 2 6 4 Verified publishers 2 2 4 2 The three Harbor hostnames in the example resolve to two registrable domains and one verified publisher. The directory is a separate publisher, even though Harbor is the subject of its page. The unresolved publisher contributes a page, hostname and domain in Run A. It contributes no verified publisher. Report that gap beside the verified count. A single shared unknown ID would collapse unrelated unresolved sites into an apparent organization. In a published comparison, explain the denominator for the gap as well. Here you could say that one of six distinct pages had an unresolved publisher, or that one of nine raw link appearances had an unresolved publisher. Both are reproducible. They differ because the second treats each link occurrence as part of the population. A reader comparing two months needs to know which proportion you use. Read historical metrics at their original level A published Lunasol study of 12 Google AI Mode runs recorded 199 outbound links to non-Google domains and 100 host placements when a host was counted once per run. The second number has a run-level scope. A host appearing in several runs can add to it several times. The first counts link occurrences, so it cannot serve as the distinct page total. Neither aggregate identifies how many independent organizations published the destinations. That requires a fresh grouping of the link records under documented page and ownership rules. Lunasol source study (/article/which-sources-does-google-ai-mode-cite-for-best-in-dubai-questions-a-lunasol-study/) When importing another report, record the metric label, capture boundary and counting method beside the value. Even a metric labeled “sources” could refer to panel cards, linked pages, hosts, domains or operators. If the study offers only an aggregate, preserve that aggregate at its original level. If you have the raw data, build any new totals as derived fields and keep a path back to the original rows. This separation also helps when your report sits alongside a business-mention analysis. A business can appear in the answer text without its own website appearing in citations. Its site can also appear as a source without the answer naming the business in prose. Report publisher attribution and entity mentions in separate columns so a stakeholder can ask whether a business was named, whether its own pages were used, or whether other publishers supplied the information. Put the counting rules in the monthly report Present raw links as capture volume. Show distinct normalized pages and registrable domains for breadth. Show host and publisher placements per run when repetition matters. Add the verified publisher count, the unresolved coverage figure and the reporting window next to each other. If you have a long tail of unresolved mappings, give readers enough context to decide whether organization-level comparisons are ready to use. For month-over-month comparisons, decide how to handle revisions to the ownership register. If a domain is newly mapped because you obtained better evidence, the verified publisher count can move with no change in citation behavior. Preserve the version used in each published report. If you recalculate the earlier month under the revised mapping, label the comparable series and keep the originally published totals accessible. An actual ownership transfer needs dates in the register so a citation can be attributed using the applicable period. Use page counts when deciding what to update or investigate. For example, several pages from one domain may each support a different part of an answer. Use domain counts when checking whether the answer draws links from a broader range of sites. Use publisher counts when assessing how much apparent domain diversity reflects multiple operations. Inspect the underlying rows before treating a publisher concentration as a finding about trust, accuracy or editorial independence. Citation counting record For each report, record the run population, capture date range, citation interface, raw link total and inclusion boundary. Add the page normalization rules, suffix list version, publisher mapping definition and version, and unresolved counts. Label columns that sum unique units within each answer and columns that deduplicate over the whole window. Save the citation rows and publisher evidence so another reviewer can reproduce the totals from the same input. AI Visibility monthly reporting guide (/article/what-should-a-monthly-ai-visibility-report-contain/) AI Visibility is published by Lunasol, which offers GEO and AEO services. The worked example is constructed. The historical figures come from a published Lunasol study and describe its recorded sample. About AI Visibility (https://aivisibility.ae/about/)