26 AI citation statistics

An AI engine reads many pages and quotes a few. Which few it quotes decides whether buyers hear your name, and the rules differ sharply from one engine to the next. The statistics below show which sources get cited most, how little the engines agree, how loosely citations track Google rankings, and how quickly the cited set churns. For the mechanics behind the numbers, see How AI Engines Choose What to Cite.

The numbers in one minute

Community sources dominate, engines barely overlap, ranking is a weak predictor, and the cited set churns within days. These seven figures frame the rest of the page.

• Reddit is the single most-cited domain across LLMs, appearing in about 40% of citations, per Semrush.
• Perplexity cites 21.87 sources per answer; ChatGPT just 7.92 — almost three times fewer — per Qwairy.
• Only ~11% of domains overlap between ChatGPT and Perplexity, across 680 million citations, per Profound.
• Just 38% of AI Overview citations also rank in Google's top 10, down from 76% a year earlier, per Ahrefs.
• An AI Overview has a ~70% chance of changing between checks, lasting about 2.15 days, per Ahrefs.
• AI-cited URLs are 25.7% fresher than organic results (1,064 vs 1,432 days), per Ahrefs.
• Brand mentions predict AI visibility 3x better than backlinks (0.664 vs 0.218), per Ahrefs.

Which sources do AI engines cite most?

Across the major engines, the citation set leans heavily on community, reference, and video platforms rather than traditional media. A handful of domains do an outsized share of the work.

1. Reddit is the most-cited domain across LLMs (40.1%)

An analysis of roughly 150,000 LLM citations found Reddit was the single most-cited domain, appearing in 40.1% of them across ChatGPT, Perplexity, Gemini, and Claude. Engines reach for Reddit because its threads carry candid, first-person opinions and product comparisons that polished marketing pages rarely match. That preference rewards real discussion over corporate copy, so a brand named naturally inside a relevant thread can earn citations a homepage never would. This 40.1% is Reddit's pooled share across all four engines, far above the 1.8% it holds of ChatGPT's own citations later in this article. A brand should treat genuine community presence as a first-order channel, not an afterthought to its owned site. Source: Semrush, 2025.

2. Wikipedia is the second most-cited domain (26.3%)

Wikipedia ranked second among most-cited domains at 26.3%, reflecting how heavily engines lean on a single neutral reference source. Models favor it because its entries are structured, broadly verified, and rarely read as promotional, which lowers the perceived risk of citing them. That trust shows up unevenly across engines: Wikipedia tops ChatGPT specifically at 7.8% of its citations, the per-engine figure listed later here. The practical lesson is indirect, since brands cannot buy a citation on a neutral encyclopedia. What they can do is earn the kind of verifiable, widely reported presence that makes a Wikipedia entry possible in the first place. Source: Semrush, 2025.

3. YouTube is the third most-cited domain (23.5%)

YouTube placed third at 23.5%, showing that video is now a first-class citation source, not an afterthought. Engines cite it because transcripts give them searchable text while the underlying video supplies demonstrations and reviews that written pages lack. With Reddit at 40.1% and Wikipedia at 26.3% ahead of it, the top three cited domains are all community, reference, or video platforms rather than traditional media. YouTube also surfaces in the per-engine breakdowns later, ranking second on both Perplexity and Google AI Overviews. A brand that publishes clear, well-captioned video on its category gives every engine another readable source to quote. Source: Semrush, 2025.

4. .com domains account for 80% of all AI citations

Across 680 million citations, .com domains made up 80.41% of AI citations and .org 11.29% — commercial sites still carry most of the cited web. The skew exists because the public web is overwhelmingly .com, so engines draw on commercial pages simply because that is where most usable content lives. This matters for brands worried that AI only quotes nonprofits or encyclopedias: in raw volume, ordinary business sites dominate the cited set. It pairs with a later finding that brand mentions, at a 0.664 correlation, predict visibility far better than backlinks at 0.218. Owning a clear, well-written .com is table stakes, but being widely talked about is what moves a commercial page into the answer. Source: Profound, 2025.

5. One study tracked 100 million citations across 13 weeks

Semrush's most-cited-domains study tracked more than 100 million AI citations across 230,000+ prompts over 13 weeks, publishing weekly top-25 snapshots — the cited set is now measured at scale. The weekly cadence matters because citations move fast, and a single snapshot would freeze a moment that has already passed. Tracking over 13 weeks is what let the same study catch Reddit's ChatGPT share sliding from roughly 60% to about 10% rather than reporting one misleading average. For a brand, the takeaway is methodological: visibility findings built on a single pull can mislead, so trends beat one-off readings. The same logic underpins the volatility numbers later, where an AI Overview has about a 70% chance of changing between checks. Source: Semrush, 2025.

6. Reddit's ChatGPT share collapsed from ~60% to ~10% in weeks

On ChatGPT, Reddit's citation share fell from roughly 60% of prompt responses in early August 2025 to about 10% by mid-September after a platform change — a single tweak reshaped the citation map. Such swings happen because each engine controls its own retrieval, and a quiet change to how it weighs a source can erase or inflate that source overnight. The drop is a sharp reminder that Reddit's 40.1% pooled share across all four engines hides wide, engine-specific volatility underneath. For a brand, it means no single platform is a safe place to concentrate every effort, however dominant it looks this month. The defensible response is to spread presence across the community, reference, and video sources that engines draw on, so one tweak cannot wipe out visibility. Source: Semrush, 2025.

How different are the engines?

The engines disagree about almost everything: how many sources to cite, which ones, and how prominently. Winning one tells you little about the others — the central reason visibility has to be tracked per engine.

7. Perplexity cites the most sources per answer (21.87)

Across 118,101 answers, Perplexity averaged 21.87 citations per answer — the most of any provider studied. Perplexity is built as an answer engine first, so it surfaces many sources by design and treats citation as part of the product rather than a footnote. That generosity widens the field: with nearly 22 slots per answer, a brand has more openings to be one of the cited sources than on any other engine. The contrast is stark against ChatGPT's 7.92 and Copilot's 2.47, both detailed below, where each slot is far scarcer. For a brand, Perplexity is the engine where breadth of relevant content pays off, because more thorough coverage gives it more chances to fill one of those many slots. Source: Qwairy, 2025.

8. Google AI Overviews and Gemini cite ~17 sources each

Google AI Overview averaged 17.93 citations per answer and Gemini 17.11 — Google's surfaces cite generously. Both lean on Google's index and its long habit of linking out, so they fold many supporting pages into a single answer. At roughly 17 sources each, these surfaces sit below Perplexity's 21.87 but well above ChatGPT's 7.92, giving a brand a moderately wide field to compete in. They also draw partly on organic rankings, a tie covered in the ranking section, so search work still feeds AI visibility here more than on the chat-first engines. A brand that ranks and is widely cited elsewhere has two routes into these answers rather than one. Source: Qwairy, 2025.

9. ChatGPT cites 7.92 sources; Copilot just 2.47

ChatGPT averaged 7.92 citations per answer and Microsoft Copilot only 2.47 — the narrowest citation surfaces, where a single slot is hard-won. These engines answer from the model first and cite sparingly, so they quote only the few sources they weigh most heavily. Scarcity raises the stakes: at 2.47 citations Copilot offers a brand barely a tenth of Perplexity's 21.87 openings, so being one of the chosen sources is far harder. It also means the per-engine leaders matter more here, which is why Wikipedia's 7.8% lead on ChatGPT, shown below, is so telling. For a brand, winning these surfaces is less about volume of content and more about being the clearly authoritative source on a specific question. Source: Qwairy, 2025.

10. ChatGPT and Perplexity share only ~11% of cited domains

Across 680 million citations, only about 11% of the domains ChatGPT cited were also cited by Perplexity — a brand visible on one can be invisible on the other. The gap exists because each engine retrieves from its own sources and weighs them by its own rules, so they end up reading largely different slices of the web. This is the clearest single argument against treating "AI visibility" as one number: an 11% overlap means a win on one engine barely predicts the other. It also explains why the per-engine leaders below diverge so sharply, with Wikipedia heading ChatGPT and Reddit heading Perplexity. A brand should measure and optimize each engine as its own channel, not pool them into a single score that hides where it is actually absent. Source: Profound, 2026.

11. On ChatGPT, Wikipedia leads (7.8% of citations)

ChatGPT's most-cited sources were Wikipedia (7.8% of its citations), Reddit (1.8%), and Forbes (1.1%) — a reference-led profile. The mix suggests ChatGPT favors sources it reads as neutral and broadly verified, leaning on an encyclopedia far more than on raw community chatter. Note the scope carefully: Reddit's 1.8% share of ChatGPT's citations is a different measure from its 40.1% pooled share across all four engines, and the two should not be conflated. With only 7.92 citation slots per ChatGPT answer, that reference-led skew means a brand benefits most from verifiable, widely reported coverage rather than forum posts alone. The contrast with Perplexity's community-led profile, shown next, is exactly why these engines share so few sources. Source: Profound, 2025.

12. On Perplexity, Reddit leads (6.6% of citations)

Perplexity's top sources were Reddit (6.6%), YouTube (2.0%), and Gartner (1.0%) — a community- and video-led profile distinct from ChatGPT's. Perplexity appears to weight live, opinion-rich sources heavily, which is why forum threads and video outrank the encyclopedia that leads ChatGPT. Reddit at 6.6% of Perplexity's citations sits well above its 1.8% on ChatGPT, a per-engine contrast that maps directly onto the roughly 11% domain overlap between the two. Because Perplexity also cites nearly 22 sources per answer, there is room for both community presence and analyst coverage like Gartner to surface. A brand chasing Perplexity should invest in genuine discussion and well-captioned video, not just the reference-style authority that wins ChatGPT. Source: Profound, 2025.

13. On Google AI Overviews, Reddit and YouTube lead

Google AI Overviews most cited Reddit (2.2%), YouTube (1.9%), and Quora (1.5%) — community Q&A and video again outrank traditional media. Google leans on these because its users phrase questions conversationally, and Q&A threads and video answer those questions in the same plain language. The pattern echoes Perplexity more than ChatGPT, since both put community and video ahead of a single reference source. It also fits the broader picture: Reddit, YouTube, and Quora are community or video platforms, the same categories that take the top three spots overall. With Overviews citing about 18 sources per answer, a brand that maintains real presence on these Q&A and video platforms has multiple realistic paths into the box. Source: Profound, 2025.

Win ChatGPT and you may still be invisible on Perplexity — the two engines share barely a tenth of the sources they cite.

How do AI citations relate to Google rankings?

Ranking and citation overlap, but loosely — and the link is weakening. A strong organic position helps, yet a large share of citations now come from pages that don't rank well at all.

14. Only 38% of AI Overview citations rank in Google's top 10

Across 863,000 keywords, Ahrefs found just 38% of pages cited in Google AI Overviews also ranked in the top 10 — down from 76% the prior year. The fall from 76% to 38% suggests Google's AI is increasingly pulling sources by relevance to the question rather than by organic position alone. That decoupling reframes the old assumption that a top-10 ranking all but guarantees inclusion, since most cited pages now sit outside it. The figure runs lower than Originality.ai's 52.5% top-10 overlap reported next, but both point the same way: page one is no longer the gate it once was. A brand should keep ranking, yet also build the topical, widely cited content that earns a quote even when it sits below the fold. Source: Ahrefs via Search Engine Journal, 2026.

15. 52.5% of AI Overview citations sit in the top 10

A separate Originality.ai study put top-10 overlap at 52.5%, with 20.7% in the top 3 and 77.9% in the top 20 — ranking still helps, but isn't required. The gradient from 20.7% in the top 3 up to 77.9% in the top 20 shows ranking and citation are correlated, since higher positions are cited more often. But that same gradient confirms position is not decisive: nearly half of citations fall outside the top 10. The 52.5% here sits above Ahrefs' 38% top-10 figure, a gap that likely reflects different keyword sets and methods rather than a contradiction. For a brand, the honest read is that strong rankings tilt the odds without settling them, so ranking work and citation-building have to run together. Source: Originality.ai, 2025.

16. Half of AI Overview citations come from outside the top 100

Originality.ai found only 48% of AI Overview citations overlapped with the top-100 organic results — the other 52% came from outside the first ten pages of Google entirely. That over half of citations come from beyond the top 100 means the engine is reaching for pages that barely rank, or do not rank for that query at all. The likely mechanism is that AI Overviews retrieve passages that answer the question directly, even when the host page never earned a strong organic position. Set against the same study's 52.5% top-10 overlap, the two figures together say ranking helps at the top yet leaves a wide back door open. A brand can therefore win citations with sharply relevant content on a page that would never crack page one. Source: Originality.ai, 2025.

17. Organic-ranking overlap rose to 54.5% over 16 months

BrightEdge found 54.5% of AI Overview citations now rank organically, up from 32.3% in May 2024 — a 22-point rise as Google tightened the link between the two. The climb suggests Google has been re-anchoring its AI to its own ranking signals after an early phase of looser, more experimental sourcing. That trend cuts against the Ahrefs finding that top-10 overlap fell from 76% to 38%, a reminder that "ranking" measured as any organic position differs from ranking in the top 10. Read together, the studies imply citations are drifting back toward organic results broadly while loosening their tie to page one specifically. For a brand, the safe conclusion is that organic SEO still feeds AI visibility, so abandoning it would be premature. Source: BrightEdge, 2025.

18. 94% of AI Overviews touch the top 20 — but 44% of citations don't

Across 362,000 keywords, seoClarity found 94% of AI Overviews overlapped the top-20 results at least once, yet 44% of individual citations came from beyond the top 20. The two numbers describe different things: 94% is how often an Overview touches the top 20 at all, while 44% is the share of individual citations drawn from below it. That distinction matters because an Overview can pull one citation from a top-ranked page and several more from pages that rank poorly. It mirrors the Originality.ai pattern, where a majority of citations came from outside the top 100, confirming that deep-ranking pages still get quoted often. A brand should read these as complementary: ranking high gets you into most Overviews, but it does not corner the citations inside them. Source: seoClarity, 2025.

19. Position 1 is cited 43% of the time; position 20 just 7%

seoClarity found URLs ranking #1 were cited by AI Overviews 43% of the time, falling steadily to 7% at position 20 — higher rankings help, but guarantee nothing. The smooth slide from 43% to 7% shows position acts as a strong probability dial rather than an on/off switch for citation. Even the top spot is cited fewer than half the time, so ranking first wins the best odds but still misses most of the time. The pattern fits the section's wider message that ranking and citation are correlated yet loosely coupled, the same loose link BrightEdge tracked rising to 54.5%. For a brand already ranking well, the implied move is to make the page itself more directly quotable, since position alone leaves more than half of its potential citations on the table. Source: seoClarity, 2025.

How volatile are AI citations?

The cited set is in constant motion. The same prompt run twice often pulls a different roster of sources, which is why a one-time visibility check goes stale almost immediately.

20. An AI Overview has a ~70% chance of changing between checks

Ahrefs tracked 43,000+ keywords and found a Google AI Overview had about a 70% chance of changing between observations, with content persisting only about 2.15 days. The churn happens because the engine re-retrieves and regenerates each answer, so even a stable query can pull a fresh roster of sources hours later. A 2.15-day lifespan means a citation a brand earns today may be gone before the week is out, with no edit to its own page. That instability is the strongest case against the one-time audit: a single check captures a state that has likely already changed. It is also why a 13-week study, like the Semrush one cited earlier, reveals patterns that any single snapshot would miss. A brand should monitor its citations continuously, treating visibility as a live signal rather than a fixed result. Source: Ahrefs, 2025.

21. 45.5% of cited sources are new between responses

On average, 45.5% of the sources cited were entirely new between consecutive AI Overview responses — only 54.5% of URLs carried over. That nearly half the cited URLs turn over each time tells you the engine reassembles its evidence rather than reusing a fixed list. It puts hard numbers behind the ~70% change rate above: when an Overview shifts, a large slice of its sources is replaced outright, not merely reordered. For a brand, the swing cuts both ways, since a source absent from one response may appear in the next without any change on its part. The defensible response is durable, repeatable relevance, so a brand keeps reappearing across many regenerations rather than betting on a single lucky pull. Source: Ahrefs, 2025.

22. The meaning holds even as the sources churn (0.95)

Despite that churn, AI Overviews scored 0.95 out of 1.0 for semantic consistency — the citations change, but the answer's substance rarely does. The engine holds a stable view of the correct answer and swaps in whichever sources happen to support it on a given pass. That gap between a near-perfect 0.95 meaning score and the 45.5% of sources that turn over is the key insight: the message is fixed, the messengers are not. For a brand, it means the goal is not to author the single canonical page but to be one of many credible sources that consistently back the established answer. Aligning content with the settled consensus on a topic, rather than fighting it, is what keeps a brand in rotation across regenerations. Source: Ahrefs, 2025.

23. AI-cited URLs are 25.7% fresher than organic results

Across 17 million citations on seven platforms, AI-cited URLs averaged 1,064 days old versus 1,432 for organic results — 25.7% fresher. Engines lean newer because recent pages are more likely to reflect current facts, prices, and product states, which lowers the risk of citing something stale. A 25.7% freshness edge means updated content has a structural advantage that has nothing to do with rankings or links. It dovetails with the churn findings above: a cited set that turns over every few days naturally favors sources that keep changing. For a brand, the implication is concrete: revisiting and updating key pages is not housekeeping but a direct lever on whether AI keeps quoting them. Source: Ahrefs, 2025.

24. ChatGPT shows the strongest freshness bias

ChatGPT cited URLs 393–458 days newer than Google's organic results, the strongest freshness bias measured; Google AI Overviews were the exception, citing slightly older content. ChatGPT likely leans hardest on recency because it answers fewer questions from a wider open web and prizes the most current source it can find. The split also fits the earlier ranking data: AI Overviews stay closer to Google's organic results, which skew older, while ChatGPT roams further for fresh material. So freshness is not a uniform rule but another axis on which the engines diverge, reinforcing the case for per-engine strategy. A brand chasing ChatGPT should prioritize visibly recent content, while one chasing Google AI can lean more on enduring, well-ranked pages. Source: Ahrefs, 2025.

What kind of source gets cited?

When the off-page signals are compared head to head, how often a brand is mentioned beats how many links it has — a meaningful shift from the classic SEO playbook.

25. Brand mentions predict AI visibility 3x better than backlinks

Across 75,000 brands, branded web mentions correlated with AI visibility at 0.664 versus 0.218 for raw backlinks — being talked about beats being linked to. The reason is that language models read text, so a brand named across many pages registers as a recognized entity even where no link points to it. At 0.664 versus 0.218, mentions correlate about three times as strongly as backlinks, a sharp break from the link-centric SEO playbook. It also explains why the community sources at the top of this article matter so much: places like Reddit and Quora generate exactly the unlinked mentions engines weigh. For a brand, the priority shifts from acquiring links to earning conversation, since being discussed is what these models actually pick up. Source: Ahrefs (reported), 2025.

26. Domain Rating is a weak predictor of citation (0.326)

Domain Rating correlated with AI visibility at just 0.326 — about half the strength of brand mentions — so raw authority scores explain less than marketers assume. A high Domain Rating signals link authority, but engines reading for relevant, current text do not treat that score as a strong reason to cite. At 0.326 it sits between backlinks at 0.218 and brand mentions at 0.664, confirming the same off-page reordering: conversation outweighs link-based authority. The practical risk is that a brand with a strong domain assumes it will be cited by default, then finds smaller, frequently mentioned competitors quoted instead. The durable play is to pair whatever authority a brand has with broad, genuine mentions and fresh, quotable content, the signals these engines actually reward. Source: Ahrefs (reported), 2025.

Source: Profound, share of each engine's citations, 2025.

What this means for brands

The citation data carries one clear instruction: treat each engine as its own channel, and earn the kinds of mentions the engines actually read. A single visibility number, checked once, tells you almost nothing.

• Community and reference sources — Reddit, Wikipedia, YouTube, Quora — do an outsized share of the citing, so genuine presence there matters more than a press release.
• Engines barely overlap, so measure and optimize for ChatGPT, Perplexity, and Google AI separately rather than as one "AI" bucket.
• Ranking helps but doesn't guarantee citation; a large share of citations come from outside the top 10.
• Citations churn within days, so visibility has to be monitored continuously, not audited once.

That continuous, per-engine view is exactly what Mentionova provides — tracking where you're cited across six engines and flagging when the answer shifts. See the deeper teardown in How AI Engines Choose What to Cite.