Definition
Citation diversity is the spread of distinct, independent sources that an AI answer engine cites. It can be read per answer (how many unique domains one response draws on) or across a prompt set (how concentrated all citations are among a few domains). High diversity means many sites share the answer. Low diversity means a handful of domains shape most of what the model says.
It is a different question from citations per response. An answer with eight citations is not diverse if six of them point at the same review site, or at copies of the same syndicated press release (the wire echo problem). Diversity counts independent sources, not links.
There is no single standard formula, but most teams combine a few signals:
- Unique domains per response: cited URLs grouped by domain, so several pages from one site count once.
- Concentration across a prompt set: the share of all citations held by the top 5 or 10 domains, or a concentration index such as Herfindahl-Hirschman.
- Source-type mix: brand sites, publishers, review platforms, forums, video, documentation and government sources.
- Independence: whether cited pages share an owner or republish the same text.
Promptwatch's study of unique domains per ChatGPT response (responses with at least one citation, September 28 to October 4, 2026) found that 47.6% of answers cited 10 or more different domains, and only 1% cited a single domain. Prompt type changed the picture: 52.5% of organic prompts reached 10 or more domains, against 21.3% of brand-specific prompts. Across the whole citation pool, concentration at the top is low. In the June 2026 ChatGPT citation share data, the most cited domain, reddit.com, held 3.71% of all citations and Wikipedia came second at 0.56%.
For GEO, diversity tells you what kind of work earns a place in the answer. In diverse answers, typical for category and comparison prompts, no single page carries the response, and third-party mentions on listicles, reviews and forums are often the way in. In concentrated answers, typical for prompts about one brand, the few cited pages decide how the model describes you, so owning or correcting those pages matters more than adding new ones.
Low diversity also creates fragility. If one engine leans on a single outdated comparison page, that page can distort hundreds of answers until it is updated or displaced. High diversity reduces that risk but raises competition, because each source carries less weight in the multi-source synthesis.
To act on it, measure diversity per AI engine and per prompt type, list the domains that dominate your prompts, and classify them by citation content type. Then decide per cluster whether to get onto the pages that already win or to publish a source that is missing. Read it next to citation share, which shows how much of that pool belongs to you.
Examples of Citation Diversity
- A CRM vendor finds that answers to 'best CRM for startups' cite 11 domains on average, while answers to '[brand] pricing' cite 4, so it targets listicles for the first group and fixes its own pricing page for the second.
- An analyst flags a cluster of AI Overviews answers as low diversity after spotting five cited URLs that all republish the same wire story.
- A healthcare publisher checks that answers about a medication cite clinical guidelines and patient resources, not only forum threads, before recommending content investments.
- After a model update, a team sees the top 10 domains' share of citations in its category rise from 30% to 45% and reprioritizes outreach to those 10 sites.
