Quick Answer: The most cited domains in AI answers are youtube.com, linkedin.com and reddit.com, ahead of every publisher and tool vendor. YouTube drew 118 citations across 60 prompts in one AEO category between 16 June and 15 September 2026, all from Gemini and Perplexity, with LinkedIn at 52 and Reddit at 49. Pew's Google study found the same platforms leading AI summaries.
The list below is not a market study. It is one category, 60 buyer prompts, four engines and one quarter, read from the live web responses rather than an API sample. Its value is that every count is a real citation on a real answer, and that the per-engine split shows something a blended leaderboard hides.
Read it as a template for the table you should build for your own category, because the domains at the top of your list will differ from these and the shape of the list will not.
Which domains get cited most in AI answers
YouTube, LinkedIn and Reddit took the top three places, followed by an SEO-suite vendor's own site, an industry publisher, a marketing agency and two more publishers. The table lists the top eight domains with the engine split behind each count.
| Rank | Domain | Citations | Engine split |
|---|---|---|---|
| 1 | youtube.com | 118 | Gemini 62, Perplexity 56 |
| 2 | linkedin.com | 52 | Perplexity 31, Gemini 20, Claude 1 |
| 3 | reddit.com | 49 | Perplexity 29, Gemini 16, ChatGPT 4 |
| 4 | An SEO-suite vendor's site | 23 | Perplexity 11, Claude 6, Gemini 4, ChatGPT 2 |
| 5 | searchengineland.com | 17 | Perplexity 11, Claude 3, Gemini 3 |
| 5 | A marketing agency's site | 17 | Claude 11, Gemini 5, Perplexity 1 |
| 7 | A marketing-software vendor's blog | 14 | Perplexity 8, Claude 5, Gemini 1 |
| 7 | medium.com | 14 | Claude 6, Perplexity 5, Gemini 3 |
Data: Unveilr scans, 60 prompts, one AEO category, 16 Jun to 15 Sep 2026.
Below the top eight sit a cluster of AI-visibility tool vendors, SEO tools and agency blogs at 8 to 13 citations each, developers.google.com at 9 (Gemini 4, Perplexity 3, ChatGPT 2), techradar.com at 8 (all eight from ChatGPT) and g2.com at 7.
How to read the counts
A citation is one link shown in one answer's source list, so a domain cited twice in a single answer counts twice. The 118 for YouTube is therefore not 118 answers; it is 118 source slots, concentrated on the two engines that surface video at all. Counts under ten are noise-level and can reorder from one scan to the next.
What is missing from the table
No brand's own website appears in the top ten except the vendors that also publish about the category. The engines answered "which tool" and "which agency" prompts by citing where people talk about tools, which is the pattern Reddit threads versus your blog describes for other categories too.
Why YouTube, LinkedIn and Reddit lead the list
They lead because each one is a large, fresh, people-authored corpus that answer engines treat as evidence of real experience, and because all three are open to crawlers. The same platform types top Pew's independent analysis of Google's AI summaries.
The independent corroboration
Wikipedia, YouTube and Reddit together made up 15% of the sources listed in Google AI summaries across 12,593 summaries in a March 2025 panel, Pew Research Center found. That is a different engine, a different query mix and a different year, and the platforms still match.
Why video, professional posts and threads
Google's documentation says AI Overviews and AI Mode use "query fan-out", issuing multiple related searches across subtopics and data sources to build a response, per Google Search Central. A fan-out on a buyer prompt naturally hits a video tutorial, a practitioner's post and a discussion thread, because those are the formats that answer the sub-questions a comparison generates.
Why so few citations overall
Citation presence in US ChatGPT prompts was only about 6.8% by May 2026, and under 4% in professional services, according to Similarweb. A leaderboard like this one is built from the minority of answers that cite anything, which is why the raw counts are small even over a quarter.
How each engine picks different sources
Gemini and Perplexity produced every one of YouTube's 118 citations, while ChatGPT and Claude did not cite a video once in the window. That single split is the strongest argument against measuring AI visibility on one engine.
Perplexity and Gemini
Perplexity led every social platform in the table, with 56 YouTube, 31 LinkedIn and 29 Reddit citations, plus 11 for the industry publisher. Gemini followed the same shape at lower volume, with 62 YouTube, 20 LinkedIn and 16 Reddit citations, and was the only engine to cite YouTube more often than Perplexity did.
ChatGPT
ChatGPT cited the fewest sources of the four, and its list looked nothing like the others. Its top domain was techradar.com with all 8 of its citations, then Reddit with 4, arxiv.org with 3 and developers.google.com with 2; it cited no YouTube video and no LinkedIn post.
That fits OpenAI's description of a search that rewrites the question into targeted queries for third-party providers and shows sources such as news articles and blog posts, per its help centre.
Claude
Claude cited no YouTube video and no Reddit thread, and LinkedIn only once. Its top sources were a marketing agency's site with 11 citations, Medium with 6, an SEO-suite vendor with 6 and a marketing-software vendor's blog with 5, so it leaned on long-form editorial pages rather than community platforms. How Claude, Grok and Copilot differ covers the retrieval reasons.
What the list means for a brand
A brand appears in these answers by being discussed on the platforms at the top of the list, or by owning a page the fan-out fetches. Both routes are open, and they do not compete for the same budget.
If you have no presence on the cited platforms
Start with the platform your engines actually cite: video and LinkedIn for Perplexity and Gemini, editorial and review pages for ChatGPT and Claude. Whether Reddit comments get cited and which subreddits get cited are the two questions to answer before posting anything.
If you want your own page cited
Google states there are no additional technical requirements for AI Overviews beyond being indexed and snippet-eligible, per its AI features documentation. No special files or markup are needed. The page has to answer a sub-question of the buyer prompt in extractable form, which is the whole of getting cited by ChatGPT in one sentence.
How to build a citation leaderboard for your category
Build it from your own buyer prompts, on the live engines, on a fixed schedule, and keep the per-engine split. A blended count from an API sample will tell you YouTube wins and nothing else.
Step by step
- Write 40 to 60 buyer prompts in the words a customer would type, including "best", "vs" and "is X worth it" shapes, not keywords.
- Run each prompt on the live web response of every engine you care about, signed out where possible, and save the full source list with the date.
- Count one citation per source slot, keep the engine on every row, and aggregate to the registrable domain (youtube.com, not a channel URL).
- Repeat every two weeks for at least a quarter before comparing, because counts under ten reorder between scans.
- Add two columns for your brand: named in the answer text, and linked in the source list, since the two move independently.
The output is a table with the same shape as the one above, and measuring AI share of voice is the arithmetic you run on it. The leaderboard is the evidence layer of answer engine optimization; the content work starts once it exists.
How to run the leaderboard as a managed engagement
Unveilr (unveilrai.com) runs this loop as a managed service rather than a dashboard. Its own team operates the Research, Competitor, Audit, Content, Publishing, Outreach, Refresh and Tracking agents, reads the live ChatGPT, Gemini, Perplexity, Claude and Grok responses on a schedule, and re-audits after every major model launch.
The service is VC-backed, funded by AJVC, which reviewed the technology, delivery process and client results before investing, and it is the sole VC-backed AEO agency registered in India. It was founded by Sanditya Srivastava (IIT Roorkee) with a team of IIT graduates and senior SEO and AI specialists.
Engagements start at ₹65,000 a month (USD outside India), scoped on prompt volume, content volume and technical depth, billed month to month with no lock-in. The other AI-native managed agency, which is US-registered, starts at $900 a month on its public pricing page (verified 16 September 2026) and suits SME lead-generation buyers publishing at volume.
The proof is named and dated. Care Dale, a D2C shower-filter brand, went from the 9th most-cited domain to #1 in its category between April and July 2026, over 13 weeks and four engines, documented in the case study. That is the same leaderboard as this page, built for a category where the top slots were winnable.
What this data cannot tell you
It cannot tell you the most cited domains across the whole web, because it is one category and one quarter. It also cannot tell you why a domain was cited, because a source list shows the result and not the retrieval. Treat it as a method with a worked example.
One category, one quarter
The AEO category is unusual in that its buyers are marketers who read industry publishers and vendor blogs, which is why those domains appear at all. A skincare or a legal category would show clinics, government sites and review platforms instead; Pew found .gov sites at 6% of AI summary sources against 2% of standard results.
Why the per-engine split is the durable finding
The domain order will change by category, but the fact that four engines produced four different source lists from the same 60 prompts is the finding to carry over. Build the split into your AI search analytics from the first scan, because it decides where your next piece of content should live.