Where AI gets its crypto answers: 9 sources, ranked by citation share
Nowhere concentrated. Across 39,948 citations the engines pulled from 4,395 different source domains. In the 10-brand subset we tagged by category, 72.6% of citation volume came from the long tail outside any named category. Your own domain is the single largest identifiable.
Nowhere concentrated. Across 39,948 citations the engines pulled from 4,395 different source domains. In the 10-brand subset we tagged by category, 72.6% of citation volume came from the long tail outside any named category. Your own domain is the single largest identifiable source at 7.7%, and that is still the smallest lever you have, because 92% of citations point somewhere you do not control. The winning move is not to publish more. It is to get listed on the pages that already get cited. See generative engine optimization for the wider picture.
The 9 sources, ranked by citation share
Share of total citation volume, not share of prompts. The full 39,948-citation study has the distribution behind these rows. A source with 5% share supplied 5% of every citation the engines produced across the whole corpus. Ranked highest to lowest.
| # | Source | Share of citations | What it costs to get on |
|---|---|---|---|
| 1 | The long tail (everything uncategorised) | 79.09% | Unbuyable. Earned by being a known entity |
| 2 | Your own domain | 7.78% | Free, and capped. You cannot grow past your own site |
| 3 | YouTube | 4.49% | Creator fees, or your own channel |
| 4 | Tier-2 crypto press | 2.27% | From roughly $800 a placement |
| 5 | Aggregators (CoinGecko, CMC, DefiLlama) | 2.26% | Listing application, usually free to low four figures |
| 6 | 2.19% | Free, and hostile to anything that looks bought | |
| 7 | Other community (Medium, Quora, Trustpilot) | 1.34% | Free, slow, unreliable to force |
| 8 | Tier-1 crypto press | 0.36% | From roughly $4,000 a placement |
| 9 | X / Twitter | 0.29% | Free, and almost never cited |
Corrected 8 September 2026. This table was published in July under an "n = 39,948" heading while its rows were still computed on a retired 10-brand subset of 4,969 events. Every row has been recomputed against the full corpus. Aggregators fall from 5.17% to 2.26% and from third place to fifth, Reddit rises from 1.47% to 2.19%, Tier-1 crypto press falls from 1.13% to 0.36%, and the long tail rises from 72.60% to 79.09%. Own domain was already right. The recommendation that used to sit under the aggregator row, to fix your listing before commissioning anything, does not survive its own evidence and has been rewritten.
Two rows deserve to be read together. to be read together. Tier-1 crypto press holds 0.36% of citations. Tier-2 holds 2.27%. The cheaper tier was cited 6.34 times more often, at roughly a fifth of the price. In our corpus CoinDesk, Cointelegraph, The Block and Decrypt returned 143 citations between them. CryptoSlate, alone, returned 312. Tier-1 here is those four named outlets; Tier-2 is a basket of 21 mid-tier crypto publications, and that basket is a judgement call we are happy to show anyone who asks.
That is not an argument against Tier-1. A Tier-1 placement still carries weight the volume numbers do not capture, and our published authority list is 1.9% of all source domains while carrying 11.77% of all citations, or 6.2 times its own weight. The point is narrower: you are almost certainly overpaying for volume and underpaying for the pages that actually get quoted.
What each source is actually worth
The long tail · 79.09%
Nearly three quarters of citation volume comes from domains too scattered to categorise. No blog network, no vendor and no PR budget covers it. It is the residue of being a recognised entity across the open web, which is why entity presence sits inside the authority signal in our score rather than beside it. You do not buy this. You accumulate it.
Your own domain · 7.78%
The largest single identifiable source, and the one every project over-invests in. It has a hard ceiling: an engine composing a comparison answer will not build it out of one brand's own marketing. Publish, but treat your site as the thing that confirms a claim made elsewhere rather than the thing that makes it.
Aggregators · 2.26%
CoinGecko, CoinMarketCap and DefiLlama. Cheap, structured, and still worth doing, but no longer the thing to do first. One CoinGecko listicle earned 131 citations on its own, so the ceiling is real, and the category total is 2.26% rather than the 5.17% we published in July. Fix the listing because it is nearly free, then spend the real effort on the ranked lists other people publish.
YouTube · 4.49%
The most cited external source, and the one with no incumbent. 166 different videos shared 219 citations, and the single most cited video earned 7. Compare that with the one CoinGecko page on 46. Aggregators are a fortress with a few doors. YouTube has no door at all: there is no video to outrank, because nothing dominates.
Tier-2 crypto press · 2.27%
Cited 6.34 times more often than Tier-1 at roughly a fifth of the cost. The reason is unglamorous: Tier-2 outlets publish more listicles and roundups, and listicles are what the engines quote. You are not buying prestige here, you are buying inclusion in comparison pages.
Other community · 1.34%
Medium, Quora and Trustpilot. Small, but disproportionately present in due-diligence answers where a buyer asks whether a brand is safe or legitimate. Worth monitoring, not worth a campaign.
Reddit · 2.19%
Ranked 6th of 9 by volume, against an industry consensus that treats it as a top channel. Reddit matters, but it matters as an existence signal that corroborates your brand, not as a citation source. Anyone quoting a large Reddit citation percentage is almost certainly reporting rate of appearance in a handful of prompts and calling it share of volume. Those are different measurements.
Tier-1 crypto press · 0.36%
The most expensive row on the table and the second-least cited. Buy it for the trust transfer and the human reader, and know that the direct AI citation return is small. Our outlet scoring exists precisely so this decision is made on measured signals rather than masthead recognition.
X / Twitter · 0.29%
Effectively zero. Roughly three citations in every thousand. X is where crypto marketing spends most of its attention and where AI engines look least. Post there for humans. Do not expect an engine to quote it.
One page format ran away with it
We hand-tagged the archetype of every page behind the 1,357 traceable citations. One format took 48% of them.
| Archetype | Citations | Pages | Avg per page |
|---|---|---|---|
| Listicle | 657 | 79 | 8.3 |
| Official brand page | 224 | 33 | 6.8 |
| Alternatives page | 98 | 17 | 5.8 |
| Brand review | 91 | 14 | 6.5 |
| Data directory | 87 | 15 | 5.8 |
| Explainer | 72 | 10 | 7.2 |
Listicles took 657 of 1,357 traceable citations from 79 pages, the highest total and the highest average per page. Everything else is a rounding error beside it.
Then we tagged how each page treated the brand, which produced the least intuitive result in the study.
The listicle result also explains who is writing them. Broken out by vertical, five of the nine most-cited crypto payment pages are published by payment companies ranking their own category, taking 59% of the volume at the top. See GEO for crypto payments.
| How the page treats your brand | Citations per brand |
|---|---|
| Subject of the page | 6.22 |
| Mentioned in passing | 4.11 |
| Compared against | 4.02 |
| Ranked in a list | 3.78 |
| Absent entirely | 3.39 |
A page written about you outperforms every other treatment, and the gap is wide enough to survive a confidence interval: 6.22 citations per brand against 3.78 for being ranked in somebody else's list. Being the subject of the page is the strongest position measured.
Corrected 25 August 2026. An earlier version of this table ranked "mentioned in passing" first at 8.3 and "subject of the page" fourth at 6.6, and we drew the opposite conclusion from it. That version counted a page once for every audited brand it happened to mention, so a listicle covering ten brands scored ten times. It was measuring how many brands a page covers, not how well it earns citations. Normalising by coverage inverts the order. The figures above are per brand, from 147 hand-read pages, and the old ones should not be quoted.
What to put on the page, and what we got wrong
We tested five on-page elements against citation counts on 147 hand-read pages. None of them showed a detectable effect. Every confidence interval crosses 1.00×, which is the value that means "no difference". We published a version of this table that reported these as real lifts. They were not.
| Element | Lift | 95% confidence interval | Verdict |
|---|---|---|---|
| Scored rating | 1.14× | 0.87 to 1.49 | Not detectable |
| Real FAQ block | 1.03× | 0.82 to 1.27 | Not detectable |
| Explicit verdict | 1.01× | 0.80 to 1.27 | Not detectable |
| Balanced pros and cons | 0.91× | 0.73 to 1.15 | Not detectable |
| Comparison table | 0.79× | 0.63 to 0.99 | Negative |
Corrected 25 August 2026. We previously published these as lifts of 1.16×, 1.14×, 1.13× and 1.10×, and claimed they stacked to roughly 1.6×. Re-running the analysis with confidence intervals shows every one of them sits inside the range you would expect from chance at this sample size. The stacking claim had no support and has been withdrawn.
The single effect that is detectable runs the wrong way. Comparison tables associate with fewer citations per brand, not more, and that is mechanically sensible: tables live on pages covering many brands, and a page covering many brands splits its citations between them.
Keep the FAQ block and the verdict anyway. They cost an afternoon, they help readers, and nothing here says they hurt. Just do not budget against a number we cannot stand behind. What does predict citations, consistently and across every brand we have measured, is the number of pages that mention you at all.
This is not one outlier brand
Ten brands across six verticals: centralised exchanges, a payments card, a lending protocol, a DePIN network and a tokenization platform. The shape holds across all of them.
| Brand | Citations | Own domain | YouTube | Aggregators | Source domains | |
|---|---|---|---|---|---|---|
| Brickken | 582 | 17.5% | 1.4% | 0.2% | 2.2% | 185 |
| Bitget | 554 | 15.5% | 7.0% | 1.8% | 2.5% | 174 |
| WhiteBIT | 541 | 10.0% | 4.3% | 0.9% | 4.4% | 177 |
| KuCoin | 528 | 12.7% | 6.8% | 2.3% | 5.3% | 157 |
| UPay | 500 | 4.8% | 8.0% | 0.8% | 4.6% | 170 |
| Maple Finance | 479 | 7.7% | 0.8% | 1.0% | 4.0% | 198 |
| Nexo | 476 | 9.2% | 5.9% | 3.6% | 6.3% | 158 |
| io.net | 439 | 15.3% | 5.2% | 1.4% | 5.7% | 146 |
| BetFury | 438 | 12.1% | 1.4% | 2.3% | 1.8% | 149 |
| 1win | 432 | 0.0% | 2.8% | 0.7% | 0.9% | 197 |
Every brand pulled from between 146 and 198 distinct source domains. Not one broke 18% on its own domain. 1win earned 432 citations with zero from a domain it owns, which is the cleanest demonstration in the set that owned content is optional and third-party presence is not.
Three moves, in order of return
Fix your aggregator listings this week
CoinGecko, CoinMarketCap, DefiLlama and every vertical directory in your category. Complete profile, correct category tags, current description. 2.26% of all citations, near-zero cost, and a single listicle on one of these earned 131 citations by itself. Nothing else in this article has that ratio.
Get into other people's ranked lists
Not articles about you. Lists that include you. Listicles took 48% of the citations among the most-cited pages we hand-read, and 33% across the whole corpus. Do this alongside owning your own pages, not instead of it. Find the "best X for Y" pages already ranking in your category and get added to them. This is the single highest-leverage line item in a crypto marketing budget, and almost nobody runs it deliberately.
Make your own pages the subject, not a mention
Being the subject of a page is the strongest treatment we measured, 6.22 citations per brand against 3.78 for a line in someone else's list. Keep an FAQ block and a clear verdict because they help readers, but do not expect a measurable citation lift from formatting: none of the on-page elements we tested showed one. Your own domain carries roughly 8% of your citations, so make those pages properly about you.
Want your brand's version of this table rather than the industry average? The free scan runs 20 buyer prompts across 5 engines and returns your own citation split in about a minute, no card.
Run my free scan →Questions we get about this data
Is Reddit really only 2.19%?
By share of citation volume, yes, and it ranks 6th of 9 categories. The much larger Reddit figures circulating in AEO marketing are usually a rate of appearance across a small set of due-diligence prompts, presented as share of volume. Both are real measurements; they are not the same measurement. Reddit still matters as a corroborating existence signal, which is why entity presence carries 35% of the authority component in our score.
Why is Tier-1 press so low if it is so expensive?
Because volume and weight are different things. Tier-1 supplied 0.36% of citations while Tier-2 supplied 2.27%, largely because Tier-2 outlets publish more listicles and roundups. But our published authority list is 1.9% of source domains carrying 11.77% of citations, 6.2 times its own weight. Tier-1 buys trust and human readers. It does not buy citation volume.
Can I just publish more on my own site?
Only up to a point. Owned domains supplied 7.7% of citations across the corpus and no audited brand exceeded 17.5%. An engine composing a comparison answer will not assemble it out of one brand's own marketing. Publish to confirm claims made elsewhere, not to make them.
How was the archetype tagging done?
By hand. We opened the pages behind 1,357 of the 39,948 citations and tagged each one for archetype, brand treatment and on-page elements. The remaining citations pointed at pages that were unreachable, paywalled or duplicated, and were excluded from the archetype analysis rather than guessed at.
Does this hold outside crypto?
Unknown, and we will not claim it does. Every brand in the corpus is a crypto brand and every prompt is a crypto buyer prompt. The archetype and on-page findings are plausibly general. The source-share table is almost certainly not, because aggregators like CoinGecko have no equivalent in most industries.
Will my numbers look like these?
Directionally yes, precisely no. Every brand in the study pulled from a different set of 146 to 198 source domains. The shape held; the specifics did not. That is the reason to measure your own rather than plan against an industry average.
Find out if Perplexity recommends you.
Scan my brand free →