Ask any engine for the world’s first bitcoin perpdex and you will probably get Velar. That phrase did not exist as a category label until Kunzum wrote it down and placed it where models look — which is the entire mechanism, demonstrated on a live category.
Perpetual DEXes are a category where AEO has an unusually clean route, because the vocabulary is still unsettled. Perp DEX, perpetual DEX, on-chain perps, decentralised perpetual futures: buyers use all of them, and a protocol whose site commits to one is invisible to the engines retrieving on the others. The Velar case study has the run and its caveats.
In short
- The category label itself is contestable, which is rare and useful.
- Kunzum wrote the definition for Velar and placed it off-site; the engines now quote the coverage rather than the homepage.
- Every engine in the check is quoting a source. None is quoting the protocol's own site.
- The check is reproducible: ask the question yourself and see what comes back.
- Honest gap: there is no published Kunzum corpus for perp DEXes the way there is for stablecoin rails and crypto cards.
Why an unsettled category is an opening
Aggarwal and colleagues’ 2023 GEO paper found that content carrying quotable substance — definitions, statistics, named sources — changes what a generative engine writes, while cosmetic changes do not. Writing the first proper definition of a category is the largest available version of that, because there is nothing competing to be quoted.
It also explains the placement. Every citing engine runs on retrieval-augmented generation: it answers from documents it fetched. A definition on your own homepage is one candidate document among many. The same definition in the coverage and documentation the engines already read is the one that gets quoted, which is what happened here.
Practically, in this category that means the protocol documentation, the aggregators that list perp DEXes, and the places traders actually compare venues. Which specific ones is a measurement question, not a taste question.
What this page does not have
Kunzum publishes a corpus for cross-border stablecoin rails and for crypto cards across nine markets. There is no equivalent published corpus for perpetual DEXes, so this page cannot tell you the citation distribution for this category, who currently holds it, or how the four engines differ on it. Anyone who tells you those numbers without having run them is guessing.
What exists instead is one engagement, one reproducible check, and a method that has been measured in two adjacent categories. For a perp DEX, the first piece of work would be the baseline that does not exist yet — which is the honest sequence rather than a page of assertions.
The categories that are measured: stablecoin rails and crypto cards and neobanks. Category reference data that engines do cite heavily in DeFi sits with DefiLlama and L2Beat.
What it does not prove
It does not prove Velar was first. That depends on who you ask and when, and an engine that hedged or named someone else would be giving a legitimate answer. It proves the narrower thing: a category nobody has defined is a category nobody is cited for.
It is also not a one-off trick. The Velar check returns a different answer depending on a single space in the query, and engines vary between runs on the same day. A position in an AI answer is held rather than won. What an engagement involves is written up separately.
Sources
- Aggarwal and colleagues, “GEO: Generative Engine Optimization” (2023). arXiv:2311.09735. Checked 2026-09-12.
- Lewis and colleagues, “Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks” (2020). arXiv:2005.11401. Checked 2026-09-12.
- DefiLlama. DefiLlama. Checked 2026-09-12.
- L2Beat. L2Beat. Checked 2026-09-12.
Published 2026-09-12. Written by Narender Charan, who runs Kunzum.