Claude Has Never Cited Reddit. Perplexity Cites It More Than Anything Else.

Most advice about AI citations assumes there is a list. Get on the list — the handful of domains the engines trust — and you get cited everywhere. We had a month of measurement sitting in a database, so we counted.

There is no list.

Across 55,693 citations from 10,500 runs on 40 sites and six answer surfaces, the ten most-cited domains account for 10.4% of all citations. 39.4% of the 6,771 domains cited were cited exactly once.

And the domains that do lead a market almost never lead a second one.

Disclosure, up front. Lead Media operates sites in several of the markets measured here, verdikt.bet among them. The measurement engine, GetCited, is our own. We sell nothing. We are not a neutral party. Every scan's own target domain is excluded from every ranking below — a scan trivially cites its own subject, and leaving that in would have flattered us.

Concentration is local, not global

Corpus-wide, the top ten domains take 10.4% of citations. Inside a single vertical, the top ten take between 15% and 28%:

| vertical | distinct domains | citations | top-10 share | top-20 share | cited exactly once | |---|---:|---:|---:|---:|---:| | local services | 2,159 | 21,684 | 21.2% | 32.5% | 44.4% | | trading / finance | 1,658 | 16,418 | 18.6% | 26.6% | 34.6% | | gambling | 1,595 | 12,001 | 15.0% | 22.0% | 28.9% | | software / consumer tech | 922 | 4,822 | 27.7% | 37.8% | 47.0% | | other | 697 | 2,456 | 26.5% | 33.6% | 49.9% |

Every market has a head. The heads are just different heads. Pool them and the concentration dissolves, because there is no shared centre for it to concentrate on.

Forty of forty-three lead exactly one vertical

Take the top ten domains of each of our five verticals. Fifty slots. They are filled by 43 distinct domains — and 40 of those 43 lead exactly one vertical.

| vertical | top five (excluding each scan's own target) | |---|---| | gambling | casino.org (311), coljuegos.gov.co (200), deadspin.com (194), reddit.com (192), trustpilot.com (160) | | local services | yelp.com (616), standardpest.com (504), reddit.com (440), advancedexterminating.com (426), nyc.gov (392) | | trading / finance | youtube.com (503), forexbrokers.com (446), sashares.co.za (376), brokerchooser.com (372), scribehow.com (323) | | software / tech | youtube.com (354), techradar.com (242), reddit.com (174), security.org (116), tomsguide.com (86) | | other | absolute.digital (78), linkedin.com (76), awisee.com (37), youtube.com (37), seo.casino (31) |

Vertical authority does not transfer. A domain that owns pest control in New York is invisible in forex, and the reverse. There is no reservoir of general trust to draw down.

Only two domains cross at all: reddit.com, in the top ten of all five verticals, and youtube.com, in four.

So — Reddit and YouTube are the universal AI citation channel. That was the headline we had. It survived about twenty minutes.

The crossers are not consensus

Split those same citations by engine. This is the raw table — no exclusions at all, so nothing in our filtering can manufacture what follows.

| surface | all citations | reddit.com | youtube.com | x.com / twitter | quora | wikipedia | trustpilot | |---|---:|---:|---:|---:|---:|---:|---:| | claude | 20,724 | 0 | 0 | 0 | 6 | 28 | 33 | | claude-opus | 870 | 0 | 0 | 0 | 0 | 1 | 7 | | aio | 18,022 | 721 | 975 | 2 | 94 | 29 | 9 | | chatgpt | 16,549 | 75 | 302 | 11 | 0 | 172 | 61 | | perplexity | 6,599 | 476 | 250 | 13 | 0 | 11 | 90 | | gemini | 3,052 | 25 | 23 | 0 | 7 | 5 | 8 |

Across 21,594 Claude citations, reddit.com appears zero times. YouTube, zero. X, zero.

Perplexity, meanwhile, makes reddit.com its single most-cited domain — 7.15% of everything it cites. On AI Overviews, Reddit ranks second and YouTube first. On ChatGPT, Reddit ranks 30th.

Reddit is not an AI citation channel. Reddit is a Perplexity and AI Overviews citation channel, and an irrelevance on Claude. The aggregate number said "universal". The aggregate number was hiding two opposite behaviours that happened to sum to something.

Is that our bug?

It is the first thing we checked, because a zero across 21,594 rows is either a real fact or a broken pipe.

The runner has no domain blocklist, allowlist, or filter of any kind — we went and looked. Claude's citations come straight out of the Anthropic Messages API web_search tool's own citation objects; we record what it hands us. And Claude is plainly not refusing user-generated content as a class: it cites Quora (6) and Trustpilot (40) in the same corpus. It is these specific platforms, absent.

Why? We don't know. Reddit, YouTube and X each have distinctive crawler and licensing postures, and different engines have struck different content deals. We can see the absence. We cannot see the cause, and we are not going to invent one.

Two things we found by getting them wrong first

google.com looked like a top cross-vertical domain. It isn't a publisher at all. Of its 1,274 citations, 655 are Google Maps and Business-profile search URLs — many carrying ?utm_source=openai — and 618 are Google Shopping URLs. That is an engine citing Google's own surfaces, not a source earning a citation. We pulled it out of the head analysis and reported it here instead. Left in, it would have produced a confident paragraph about "Google's authority" that meant nothing.

Our forty sites are not forty independent markets. Roughly ten are New York pest-control sites; roughly eight are forex sites. Compute head-overlap at site level and pest-control competitors light up as "cross-market" domains — when in truth we measured one market ten times. That is why every overlap number above is computed at vertical level. The site-level version of this post would have been wrong, and it would have looked more impressive.

And a correction to our own thinking

Wikipedia is cited on 0.1% to 1.0% of citations, depending on surface. ChatGPT is the most Wikipedia-friendly engine in our data at 1.0%; Claude sits at 0.1%.

We have written elsewhere — in our own internal notes — that Wikipedia is "heavily weighted in LLM retrieval". This data does not support that, and it does not refute it either: Wikipedia's influence plausibly runs through training and grounding rather than through a visible citation. But the two things are not the same, and we had been treating them as if they were. Getting cited by Wikipedia and getting cited by an answer engine are different objectives, measured on different surfaces. Both are worth pursuing. Neither is evidence for the other.

What we are not saying

This is our prompt corpus, not a random sample of the web. The prompts exist to measure sites we and our clients operate, and the vertical taxonomy is ours — "local services" spans pest control, mould, flood damage, plumbing and a clinic, and a finer taxonomy could move the overlap numbers.

The surfaces are unevenly represented: Claude contributes 20,724 citations, Gemini 3,052. Within-surface shares are safe; cross-surface volume comparisons are not, and we have not made any. Google AI Overviews contributes almost nothing to gambling, because it does not render for commercial gambling queries at all — gambling is effectively a three-engine vertical here.

And the claude surface is the Anthropic API's web_search tool, not the claude.ai browsing interface. They may not behave the same.

A citation is not traffic. A citation is not an endorsement.

What we would do with this

(Interpretation. Reasoned, not measured — held to a lower standard than everything above, and marked as such.)

There is no general-purpose AI authority to buy. Forty of forty-three leading domains lead one vertical. A high-authority domain in an unrelated niche is not a citation asset in yours, and no amount of domain rating changes that.

Pick the engine before you pick the tactic. Reddit work is worth real money on Perplexity and AI Overviews and worth exactly nothing on Claude. Anyone selling "AI citation optimisation" as a single channel is selling a category error. Ask which engine, and ask for the number.

The long tail is the opening. 39.4% of cited domains were cited exactly once. The engines reach past the head constantly, looking for something specific. Being the best available source for one narrow, verifiable claim is a real route in — which is, conveniently for us, the entire thesis we already published. Treat that convenience with suspicion, and check the number yourself.

One thing did land on both sides of the ledger: casino.org tops the gambling vertical here, and it is also the most Wikipedia-cited gambling domain we found. Two independent measurements, one answer — publish reportorial fact, not toplists.

Corrections

If any number here is wrong we will correct it in place, dated, with the error left visible. This post already contains three corrections against ourselves: the Reddit headline we abandoned, the google.com artefact we removed, and the Wikipedia claim we walked back.

exp-015 will re-run monthly. The interesting question is whether Claude's zero stays a zero.