AI systems can cite homepages, educational pages and reference sources, but the mix varies by platform and topic. A citation is a link to a source, not an endorsement of the brand or a guarantee that the cited passage supports every claim in an answer. For B2B SaaS teams, the useful question is not simply which page types appear in a sample; it is what the evidence says, what remains uncertain, and what to test on their own site.
The available evidence offers descriptive patterns, not a proven recipe for earning citations. One Ahrefs classification of ChatGPT citations and a separate study of Google AI Overviews use different platforms, samples and methods. Their results should not be combined into a single benchmark.
What is an AI citation, and what is a recommendation?
An AI citation is a source link attached to an answer; a recommendation is an evaluative selection of a brand or option. A mention names a brand without necessarily endorsing it, while source relevance concerns whether the linked material supports the associated claim. These are separate observations: a link does not by itself show that the source—or its brand—was recommended.
OpenAI's ChatGPT search guidance describes citations as sources readers can open. That makes a citation a traceable link, not proof of endorsement or complete factual support. Teams should distinguish four questions: was the brand mentioned, was it recommended, was a URL cited, and does the cited passage support the claim?
The related question of third-party brand mentions vs your website concerns source ownership and prioritization; it is distinct from whether a specific citation supports an answer.
Can a brand be cited without being recommended?
Yes. An answer could link to a brand-owned page for a factual detail while selecting a competitor—or naming no preferred vendor. The link shows that a source was attached to the answer; it does not establish a positive evaluation of the brand.
Illustrative example: a B2B software procurement answer links to Vendor A's documentation to describe an integration, then recommends Vendor B for the buyer's stated needs. This is a hypothetical example of how to label an answer, not an observed AI transcript. A reviewer would record the brand mention, recommendation, cited URL and whether the cited passage supports the integration claim separately.
Can a brand be recommended without its own URL being cited?
Yes. An answer can select a brand while citing third-party pages, or while providing no brand-owned link. Reviewers should record the recommendation and source attribution separately, then check whether the cited passage supports the recommendation. They should not infer how the system reached it.
Why do citation and recommendation rates diverge?
They measure different features of an answer, so a brand name, evaluative choice and source link need not appear together. Seer Interactive's ghost-citation analysis reported an association between brand mentions and citations in its behavioral tests. That association does not establish that a mention causes a citation or that a citation causes a recommendation. Nor does it show the sequence in which a system selects them.
A practical review can code response patterns like these without treating them as measured frequencies:
Response pattern | Brand mentioned | Explicitly recommended | Brand URL cited | Third-party URL cited | Support verified |
|---|---|---|---|---|---|
Background fact about a vendor | Yes | No | Yes | No | Check the cited passage |
Vendor selected; independent comparison cited | Yes | Yes | No | Yes | Check whether the comparison supports the selection |
Category explanation with no preferred vendor | No | No | No | Yes | Check whether the source supports the explanation |
The table is a qualitative illustration of a coding framework, not a report of observed response rates. Citation counts alone cannot show whether a source is relevant, whether a claim is supported, or whether a buyer would act on the answer.
What page types are cited by AI systems?
Observed categories include homepages or landing pages, educational pages and reference sources, but their proportions vary by platform, topic and sample. No page format is guaranteed to earn citations. ChatGPT citation URLs and Google AI Overview URLs come from different systems and should retain separate definitions and denominators.
For this article, a cited page type means a broad category assigned to a URL or its content. That label is descriptive: it does not establish why the system selected the page, how useful the page was to a buyer, or whether a similar page will be cited elsewhere. The broader AI citation fundamentals are best understood as a measurement problem, not a fixed template to copy.
What did the ChatGPT cited-URL sample show?
In Ahrefs' classification of ChatGPT's top 1,000 cited URLs from September 2025, Wikipedia accounted for 29.7%, homepages and landing pages for 23.8%, and educational pages for 19.4%. These figures describe that vendor's sample only; Claude-assisted content-type labels may be imperfect, and the results do not represent all AI systems or topics.
Category | Share of the sample | Sample and caveat |
|---|---|---|
Wikipedia | 29.7% | Ahrefs' top 1,000 ChatGPT cited URLs, September 2025; vendor classification |
Homepages and landing pages | 23.8% | Same sample; labels assigned with Claude and may be imperfect |
Educational pages | 19.4% | Same sample; directional category classification |
Source: Ahrefs' analysis of ChatGPT's most-cited pages. The listed categories are the figures reported in the analysis; this table does not infer the composition of the remaining URLs. Nor does the Wikipedia share imply that marketers can reproduce it or that a page type caused citation selection.
Must a page rank in Google's top ten to be cited?
No. In Ahrefs' March 2, 2026 analysis of 863,000 keyword search engine results pages (SERPs) and 4 million AI Overview URLs, 37.9% of cited URLs also appeared in the first ten result blocks, 31.2% appeared in result blocks 11–100, and 31.0% were beyond result block 100. This is descriptive overlap in one vendor dataset; it does not show that a lower ranking caused a citation, and it does not describe ChatGPT citations. Here, result blocks include organic listings, ads and search features, not organic rankings alone.
Google results-page block category for AI Overview URLs | Share reported | Dataset |
|---|---|---|
First ten result blocks | 37.9% | Ahrefs analysis of 863,000 keyword SERPs and 4 million AI Overview URLs, March 2, 2026 |
Result blocks 11–100 | 31.2% | Same dataset |
Beyond result block 100 | 31.0% | Same dataset |
Source: Ahrefs' analysis of AI Overview citations and top-ten results. The percentages are the reported categories and should not be recast as evidence that ranking position has no relationship to citation; they show that the cited URLs in this dataset were not confined to the first ten blocks. Organic visibility and AI citations should be tracked as distinct outcomes.
Does adding schema markup cause more AI citations?
The supplied matched difference-in-differences study does not support treating JSON-LD (a format for structured-data markup) alone as a reliable citation lever. Ahrefs compared 1,885 pages that added schema between August 2025 and March 2026 with 4,000 control pages. Relative to controls, AI Overview citations were reported at −4.6%. The reported changes for AI Mode (+2.4%) and ChatGPT (+2.2%) were statistically indistinguishable from zero.
Difference-in-differences compares changes over time in a treatment group with changes in a comparison group. The study's sample consisted of pages already heavily cited by AI, so it does not establish the effect on pages with no citations. It pooled schema types and measured results 30 days after the detected addition. It also could not fully separate schema from concurrent content or technical changes.
The study does not test every schema type or control for every content-quality distinction. Its result should not be described as a universal causal penalty from schema, and the nonsignificant positive estimates are not evidence of a citation lift. It cautions against assuming that markup alone changes citation outcomes, but does not settle every possible use of structured data.
Source: Ahrefs' schema and AI citation analysis. Google's guidance on AI features and your website addresses accessibility and discoverability; it is not evidence that a specific page type or markup guarantees citations.
Which B2B SaaS pages are worth testing first?
Start with pages that answer real buyer questions using specific product facts, original evidence, useful comparison context and clear provenance. This is an editorial hypothesis for prioritizing tests, not a proven ranking of page types that earn citations.
Candidate page purposes include product documentation, use-case explainers, comparison pages and original research. Available evidence does not show that these categories outperform other pages, so treat them as cohorts to examine rather than prescriptions.
For teams assessing original research AI citations, examine whether the research has a clear method, attributable evidence and repeatable citation measures. Author expertise is a separate question; neither authorship nor page type guarantees visibility.
Candidate page purpose | Buyer question to test | Evidence to inspect | Measurement question |
|---|---|---|---|
Product documentation | Can the product perform a specific task or integrate with a named system? | Current, accessible product facts and documentation | Is the page cited for a relevant product fact across repeated prompts? |
Use-case explainer | How might a team address a defined workflow or problem? | Specific process detail, limitations and applicable product facts | Does the page appear for the intended buyer question, and is its passage relevant? |
Comparison page | How do options differ for a stated need? | Clear criteria, dated evidence and transparent scope | Are claims supported, and are alternatives represented accurately? |
Original research | What does a defined dataset or method indicate? | Research methodology, sample boundaries and attributable findings | Is the research cited for the finding it actually supports? |
These are testable page purposes, not confirmed citation winners. Keep essential information accessible and internally linked, consistent with Google's guidance, and make product claims verifiable against product documentation. Where relevant, teams can also compare category options without implying that a comparison page itself earns a citation.
How should teams measure whether a page earns citations?
Use a defined prompt set and record cited URLs by engine over repeated runs, including collection date and source. Compare similar pages while noting changes in content, crawling and rankings; a single response is an observation, not a reliable estimate of a page-level treatment effect.
Define buyer prompts. Record the intended audience, question and prompt wording so the sample can be repeated.
Record baseline citations. Capture engine, date, response, cited URLs and whether the brand is mentioned or recommended.
Classify page purpose and evidence quality. Separate URL-pattern labels from what the page actually does and what its cited passage supports.
Compare matched cohorts. Note differences in page purpose, evidence, accessibility and changes during the observation period.
Repeat across engines and dates. Report prompt-sampling and platform limitations rather than treating one run as a stable rate.
This is a measurement proposal, not an established causal result. Keep mention, recommendation, citation presence, passage support and downstream action distinct. Prompt samples and platform behavior can vary, so report the collection conditions alongside any rate.
What is still unknown about page types and citations?
Available evidence does not establish that choosing a particular page type causes citations, especially for B2B SaaS. A stronger study would classify page purpose and evidence quality, match comparable pages, and measure citations and recommendations repeatedly across engines.
URL patterns are not the same as page purpose: two URLs with similar paths can serve different buyer needs and contain different evidence. Site-level tests should record what changed, maintain clear source provenance and avoid attributing a citation shift to one intervention when content, crawling or ranking also changed. Neither page structure nor schema ensures retrieval or citation.
A useful research question is whether a preregistered B2B prompt study that codes mentions, explicit recommendations, citations and passage support separately would produce stable results under blinded human review. Until such evidence exists, page-type patterns are useful for generating hypotheses, not declaring a winning format.
Conclusion
The evidence describes which kinds of URLs appeared in specific samples; it does not show that any format reliably causes AI citations. Keep ChatGPT and Google AI Overview results separate, assess the cited passage rather than count links alone, and treat candidate B2B page types as hypotheses to test. This makes the results more useful for prioritization without turning descriptive patterns into a promise.
Frequently Asked Questions
Which sites get AI citations?
The analyzed ChatGPT sample includes Wikipedia, homepages and landing pages, and educational pages among its classified cited URLs. That is a vendor-specific sample from September 2025, not a universal ranking of sites or a prediction of which site will be cited for a particular prompt.
What are AI citations?
An AI citation is a source link attached to an AI-generated answer. It does not automatically mean the linked source endorses the answer, that its brand is recommended, or that the cited passage supports every associated claim.
How Nortav Can Help
For teams deciding which pages to improve or test for AI citations, the useful starting point is to separate page-type observations from evidence about source relevance and claim support. Nortav can help assess the current visibility picture and identify where a more structured test may clarify priorities.
Our approach starts with understanding your current search visibility, identifying gaps against competitors, and developing a strategy focused on driving qualified traffic, generating leads, and delivering measurable business value.
Book a Discovery Call →
Sources
Ahrefs, ChatGPT's most-cited pages — September 2025 top-1,000 cited-URL classification; Claude-assisted categorization may be imperfect.
Ahrefs, AI Overview citations and top-ten results — March 2, 2026 observational analysis of 863,000 keyword SERPs and 4 million AI Overview URLs.
Ahrefs, schema and AI citations — supplied matched difference-in-differences analysis of pages adding schema and controls.
Google Search Central, AI features and your website — platform guidance, not proof that a specific page type will be cited.
OpenAI Help Center, Searching the web with ChatGPT — source links in ChatGPT search.
Seer Interactive, LLM Ghost Citations: Why Your Content Is Working and Your Brand Isn’t — vendor analysis of mentions and citations; association does not establish causality.
