Gemini repeats a cited domain in two of every three responses. Perplexity repeats the exact same link instead.
An earlier post on this blog counted citations per response as a raw array length, Gemini averaging 13.8 per response against Claude's 6.0. That count treats every citation as a distinct source. It often is not. We went back through the same 8,664 real responses and checked whether each engine's citation list repeats a domain it already cited, and what kind of repeat it is.
Raw citation count overstates distinct sources, by different amounts per engine
For every real, non-error response across our 270 live audit panels, we took the citations array (excluding Gemini's opaque vertexaisearch redirect wrappers, the same exclusion used in the citation-count-per-response post) and counted the distinct domains alongside the raw citation count. A response citing g2.com twice and capterra.com once has 3 raw citations and 2 distinct domains.
| Engine | Responses with 1+ citation | Avg raw citations | Avg distinct domains | Gap |
|---|---|---|---|---|
| Gemini | 2,286 | 14.01 | 11.91 | 15.0% |
| ChatGPT | 1,179 | 7.14 | 6.32 | 11.5% |
| Claude | 2,547 | 6.34 | 5.95 | 6.1% |
| Perplexity | 2,358 | 8.80 | 8.36 | 5.0% |
Gemini's raw count overstates its distinct source count by 15%, the widest gap of the four. Perplexity's gap is the narrowest, 5%, even though Perplexity's raw citation count (8.8) sits above both Claude and ChatGPT in the earlier post's ranking. A high citation count and a diverse source list are not the same claim, and the size of the gap between them is itself an engine trait.
How often does a response repeat a domain at all
Looking at whether each response repeats any domain at least once, not just the average gap size, the engines split into two clear tiers.
| Engine | No domain repeated | At least one domain repeated |
|---|---|---|
| Gemini | 34.2% | 65.8% |
| ChatGPT | 55.6% | 44.4% |
| Perplexity | 69.5% | 30.5% |
| Claude | 68.4% | 31.6% |
Gemini repeats a domain in nearly two-thirds of its responses, almost double ChatGPT's rate and more than double Claude and Perplexity's. That tracks with Gemini's much higher raw citation count from the earlier post: the more sources an engine lists, the more chances a domain shows up twice.
Not all repeats are the same kind of repeat
A repeated domain can mean two different things. It can be two different pages on the same site, for example g2.com's HR-software category page and a separate g2.com product review, which is a real second citation with a real second URL. Or it can be the identical URL cited twice in one response's citation list, which adds nothing a source-diversity count should credit. We split every repeat into these two categories.
| Engine | Repeat is a distinct page, same domain | Repeat is the identical URL twice |
|---|---|---|
| Gemini | 65.4% | 0.4% |
| ChatGPT | 44.4% | 0.1% |
| Claude | 31.2% | 0.4% |
| Perplexity | 9.4% | 21.1% |
Why the split matters, not just the total
If you only looked at the "any repeat" numbers, Gemini and Perplexity would look like they sit at opposite ends of a similar spectrum, one repeating a lot, one repeating a little. The breakdown shows that is not the right comparison. Gemini's repeats are a byproduct of citing broadly, real second and third pages from sites it returns to across a long answer. Perplexity's repeats are closer to a formatting quirk in how its citation list gets assembled, the same link appearing twice without being deduplicated. Neither pattern is wrong, but they would call for a different fix if you were building a tool on top of these citation lists, one needs domain-level dedup and the other needs URL-level dedup.
Zero-citation rate confirms this is the same dataset
Restricting to responses with at least one citation, the counts above land on 2,286 (Gemini), 1,179 (ChatGPT), 2,547 (Claude), and 2,358 (Perplexity) out of the 8,664 non-error baseline. The zero-citation share this implies, Gemini 1.7%, ChatGPT 9.3%, Claude 5.0%, Perplexity 0%, matches the citation-count-per-response post's published figures on the same dataset exactly, confirming this is a fresh read of the same 270 live panels, not a different sample.
What this means for a visibility strategy
If an audit shows your brand cited on a page, the citation count next to that finding is not automatically the number of independent sources vouching for you. On Gemini in particular, a long citation list is more likely than any other engine's to include the same site twice under a different URL, so getting cited once on a competitor's domain-heavy site counts for less diversity than the raw number implies. The practical read is the same one from prior posts on this data: which specific domains an engine returns to, not how many total citation slots it fills, is what a content strategy should target.
See which domains each engine actually cites for your category, and whether your brand's pages are among the ones it returns to more than once.
Get an AI Visibility Audit, $490Methodology
Data drawn from the same 270 live, search-grounded audit panels used in the citation-count-per-response post (project management, customer support, and HR software brands), each run across four AI engines: ChatGPT with web search, Gemini with grounding, Perplexity Sonar, and Claude with web search. We pulled the citations array from every successful (non-error) response, both branded and organic prompts, 8,664 responses total, matching that earlier post's baseline exactly. Domain is the hostname of each citation URL after excluding Gemini's vertexaisearch redirect wrappers, the same normalization the engine's own scoring code applies. A repeat is "same URL" if the full URL (minus query string) matches an earlier citation in the same response's list, and "distinct page, same domain" if the hostname matches but the path does not. No development-rail or fixture data is included; all responses came from live engine calls. Data collected June-July 2026.