Most cited domains are cited by one engine. On 70,879 prompts where ChatGPT, Gemini, Perplexity, and Google AI Overviews all returned citations, Writesonic found 72 to 73 percent of domains on exactly one engine and 3.8% on all four. Profound's older two-engine pool, ChatGPT and Perplexity, put the shared slice at 11%. A domain that shows up on every engine somewhere in a quarter is a different, and much more common, event.
- 3.8% of domains appeared on all four engines in Writesonic's same-prompt set. 72 to 73 percent appeared on exactly one. Matching was domain-level, which the authors call the lenient version. Exact URLs would score lower.
- No pair cleared a Jaccard of 0.25. The highest was Perplexity and AI Overviews at 0.237. ChatGPT against Gemini was 0.119.
- Profound's 11% is two engines, pooled. ChatGPT-only 37.4%, Perplexity-only 51.6%, both 11.0%, from 100,000 prompts. A search for a "23 of 253" Profound chart did not find one.
- Same-day agreement is harsher than a months-long window. Authority Tech, on 2,704 question-days, found 1.0% of cited domains on every engine that cited that day.
- Window consensus, same-prompt consensus, domain matching, and URL matching are four statistics. Treating 11%, 3.8%, and 1% as one "overlap" number is how this literature gets a single answer it does not have.
Same prompt, four engines
Writesonic ran 161,286 prompts across ChatGPT, Gemini, Perplexity, and Google AI Overviews, published 22 July 2026. The overlap math uses the 70,879 prompts that returned citations from all four at once. Domains were lowercased and www. was stripped. Partial The vendor sells in the category. The method is described.
| Pair | Jaccard | Prompts in the pair |
|---|---|---|
| Perplexity and AI Overviews | 0.237 | 126,299 |
| Gemini and AI Overviews | 0.216 | 87,558 |
| Gemini and Perplexity | 0.174 | 82,221 |
| ChatGPT and Perplexity | 0.130 | 119,513 |
| ChatGPT and AI Overviews | 0.126 | 128,623 |
| ChatGPT and Gemini | 0.119 | 88,969 |
Jaccard is an overlap score, not a percentage of citations. A 0.24 means the shared domains are a small fraction of the union. The authors say a directional "of A's sources, what share also appear on B" cut was still in progress. The two-engine and three-engine shares were described as a small band and not printed as numbers. The chart shows the two ends they did print.
The 11% figure, and what it is
Profound's 1 July 2025 overlap post is the source of the figure this site already carries on the hub. 100,000 distinct prompts, ChatGPT and Perplexity. The combined pool splits as ChatGPT-only 37.4%, Perplexity-only 51.6%, both 11.0%. A broader pairwise range in the same post ran from 6% (AI Overviews versus Copilot) to 16.4% (Perplexity versus AI Overviews).
Average domains per response in that post: AI Overviews about 7.7, Perplexity about 7.3, ChatGPT about 5.0, Copilot about 2.5. An engine that shows more links contributes more rows. That is one reason a pooled "most cited on AI" list leans toward whichever product is chattiest. The hub's argument against pooling is the cross-engine page. This page is the measurement under it.
A circulating "23 of 253 domains, about 9%" was not in any Profound page opened for this pass. 23 divided by 253 is 9.1%, near the 11% and not the same statistic. It is not used.
A month of overlap is not one answer
Authority Tech, published 22 September 2026, on 3,217 domains, separates the two readings. Across the window, 71.8% of domains are ever cited by only one engine. Among domains with ten or more citations, single-engine falls to 23.0%, and the mean is 2.74 engines. That is "this host showed up on more than one product at some point."
The answer-level reading is the one that matches a user's afternoon. Across 2,704 question-days with two or more citing engines, 13.5% of cited domains showed up on more than one engine that day, and 1.0% showed up on every engine that returned a citation. URL-level multi-engine rate that day was 10.1%. Same-day exclusivity: ChatGPT 83.2%, Gemini 78.0%, Perplexity 71.3%, AI Mode 70.7%, Claude 68.0%.
Fifty-seven domains in that panel have been cited by all five engines at some point. Those 57 appear across 7,133 question-days. All five engines cited one of them for the same question on the same day nine times. That is 0.13%. Being a universal source in the abstract is almost unrelated to being the shared footnote on a given question.
Wellows, on 596,723 prompts answered by two or more engines, reports 10.2% of cited URLs on more than one engine and 17.9% at domain level, and says 97.15% of URLs ChatGPT cites appear on no other engine. Wellows sells in the category and discloses that its own domain ranks fifth in the resulting table. The direction matches. The level is another vendor's.
The few names that did show up everywhere
Foglift's Q2 2026 benchmark is small and specific: 75 buyer-intent prompts, 375 responses, 1,119 domains, 2,583 domain-citations. Twelve domains were cited at least once by each of ChatGPT, Claude, Gemini, AI Overviews, and Perplexity. That is 1.1% of the domains in the set.
| Cited at least once by all five, in this 75-prompt set |
|---|
| healthline.com |
| outdoorgearlab.com |
| sleepfoundation.org |
| salesforce.com |
| cleverhiker.com |
| monday.com |
| klaviyo.com |
| expedia.com |
| posthog.com |
| postmarkapp.com |
| solereview.com |
| theruntesters.com |
The union of the five engines' top-25 lists is 81 domains. Fifty of them, 61.7%, are in exactly one engine's top 25. Healthline is the only domain in all five top-25s. A gear-lab and a CRM vendor can share a tiny buyer panel without being the web's consensus sources. The list is here so it is not mythologised, and so it is not dropped. The sample size is the caption.
Engines also fail to move together over time. Profound's half-life study, 883,000 pages, found cross-engine correlations of 0.03 to 0.09, except Google's own surfaces at 0.12 to 0.17. Of pages in that set, 679,077 were cited by one engine. The decay schedule is on the half-life study, which is still a protocol on this site, and the vendor figure is attributed to Profound rather than claimed as ours.
What this does not prove
Open question No study opened here uses the same prompt set, the same engines, and the same matching rules. The spread from about 1% to about 11% is several measurements, not a contradiction of one. Writesonic did not print the middle of the distribution. A publisher cannot yet say "if I win ChatGPT I have a 3.8% chance of the others." That 3.8% is a property of domains in a prompt set, not a conversion rate for a URL.
Verification status
Each overlap figure keeps its unit: same prompt or any time in a window, domain or URL, two engines or five. The 23-of-253 claim was searched and not found. Foglift's twelve names are reproduced with the prompt count in the same section. This site has not run the concordance panel pre-registered at cross-platform citation concordance.
When four engines all cited something, 3.8% of domains were on all four. About three quarters were on only one.
Share on XNamdev, R. (2026). Domains cited by every AI engine, and domains cited by only one (v1). Retrieved from https://ritiknamdev.com/blog/domains-cited-by-every-ai-engine Published under CC BY 4.0 — reuse freely with attribution.
The per-engine lists should stay separate because of this gap. The hub is most-cited domains in AI search.