In one Zumi run, 4,308 AI answers cited 12,625 different pages, and 7,024 of those pages were cited exactly once. The most-cited page appeared in 230 answers, which is 5.3% of the total.
That shape matters more than any single ranking of page types. Citations in this run were spread across a large number of pages, and a small group of sites took a disproportionate slice.
How was this measured?
Zumi ran 33 buyer questions about AI search measurement software across eight engines: ChatGPT, Gemini, Microsoft Copilot, Google AI Mode, Google AI Overviews, Perplexity, DeepSeek, and Grok. Runs happened between July 19 and August 7, 2026, from the United States and India, and produced 5,120 answers.
Of those, 4,308 answers returned at least one source. Every rate in this post uses those 4,308 answers as the denominator, so a figure such as 5.3% reads as "5.3% of answers that cited sources." Claude was not part of the run.
Who gets the citations?
Pages that are not owned by any vendor in the category took nearly all of them. The split below counts citations, and a page cited in five answers counts five times.
| Who owns the cited page | Citations | Share | Domains |
|---|---|---|---|
| Third parties | 46,118 | 91.6% | 3,333 |
| Software vendors in the same category | 3,420 | 6.8% | 9 |
| The tracked brand | 798 | 1.6% | 1 |
The 3,333 third-party domains include publishers, review sites, video platforms, forums, and research repositories. The tracked brand's 798 citations were mostly product pages (486) and the homepage (176).
Which page types get cited?
Lists and articles carry most of the volume. The table shows the six largest of the 31 page types the platform assigns.
| Page type | Citations | Share |
|---|---|---|
| Best-of list | 17,730 | 35.2% |
| Blog article | 13,552 | 26.9% |
| Product page | 6,080 | 12.1% |
| Homepage | 3,064 | 6.1% |
| How-to tutorial | 2,718 | 5.4% |
| Comparison page | 1,824 | 3.6% |
The other 25 page types and the unclassified pages together took 10.7%, led by reference articles at 1.7% and community threads at 1.5%. Best-of lists, comparison pages, and alternatives pages together took 39.0% of citations. That is likely inflated by the prompt set, since the questions ask which tools are best, and a set built around how-to questions would probably shift weight toward tutorials by an amount this run cannot say.
How concentrated are citations?
Citations skew toward a small group of domains, though the long tail beyond the top 100 still accounted for 43.4%.
| Group | Share of all citations |
|---|---|
| Top 10 domains | 19.8% |
| Top 50 domains | 44.1% |
| Top 100 domains | 56.6% |
The 3,343 domains in the run include 1,549 that were cited a single time, which is 46.3%. Earning a place among the top 10 domains is a narrow target for any single brand, while appearing at all draws on a very wide field of possible sources.
Why are most pages cited only once?
Each answer draws on many sources. The 4,308 answers averaged 11.7 citations each, spread over 12,625 pages.
Pages cited at least twice made up 44.4% of pages and took 86.0% of citations. Pages cited at least ten times were 8.9% of pages and took 54.0%. Pages cited exactly once were 55.6% of pages and 14.0% of citations.
A page cited once is common, and a page cited ten times is rare. Repeat citation is where the volume sits, and it is the harder result to produce.
What does this mean for a brand?
Three readings follow from the data, each limited to this category and window.
- Third-party coverage carries the volume. A brand's own pages accounted for 1.6% of citations here, so presence on pages the brand does not control matters more than the brand's own page count.
- List and comparison pages take a large share of citations. Whether a brand appears on the best-of lists that engines cite is a measurable question, and citation tracking is how to answer it.
- A citation count alone says little. Counting pages cited once against pages cited ten times separates broad presence from repeat presence, which is the difference mention rate and share of voice are built to capture.
How can a brand read its own category the same way?
The method carries over to any category, and it needs only the cited pages from a set of buyer questions.
- Collect the sources cited for the questions buyers in the category ask, across the engines that matter to them.
- Label each cited page by owner (the brand, a competitor, or a third party) and by page type.
- Count how many pages were cited once and how many were cited ten times or more.
The result is a map of where citations land for that category. It also shows which few pages repeat, and those are the ones worth a closer look.
What does this run not show?
One category, one 20-day window, and eight engines. Claude is absent, and the prompts are questions a buyer of measurement software would ask, so the best-of share is likely higher than it would be for how-to questions.
Answers to one prompt are repetitions, not independent observations. The 33 prompts were each run 3 to 21 times per engine and region, so the 4,308 answers carry less independent evidence than the count suggests. Page types are the platform's own classification, and 148 pages were left unclassified.
This is a measurement of what engines cited, which is the discovery layer only. It does not measure traffic, clicks, or revenue from any citation.
Zumi is an AI Search Intelligence Platform that tracks which pages engines cite for a brand's category, and which of those pages mention the brand. Book a demo to see the citation view for a specific category.