I run Kunzum, a one-person studio doing AI search visibility for crypto companies. Before that I spent six years writing crypto content and running marketing for protocols.
This post is the argument for why the studio exists, plus the two research findings behind it. Both studies are published with the raw data, so you can disagree with me using my own numbers.
They open ChatGPT, Perplexity, Gemini, or they get an AI Overview before they see a single blue link. They type a question, they get four names, and they pick from those four. Nothing about your product quality enters that decision. What enters it is whether a retrievable document exists that answers their question and mentions you.
Crypto is unusually bad at producing those documents, for reasons that are mostly technical:
Client-side rendering. A huge share of crypto frontends are React apps that ship an empty shell and paint the content after hydration. Googlebot handles that reasonably well now. Most AI crawlers do not execute JavaScript at all. If your value proposition only exists after hydration, it does not exist.
Documentation treated as a support cost. Docs are usually the most factual, most structured, most chunkable content a crypto team owns, and they are maintained by whoever had time.
Unsettled category names. Is your product a crypto neobank, a stablecoin card, a payments app, or a wallet? You do not get to decide. The engines decide from published language, and if your category has four competing names your share of each one is thin.
Marketing spend that produces nothing retrievable. Threads, spaces, KOL posts, Telegram. All of it is real distribution and none of it is a document a model can retrieve six months later.
The net effect is that in crypto, being good is close to uncorrelated with being recommended. That is a testable claim, so I tested it.
Most AI visibility research asks branded or category questions. "Best perp DEX." "Top stablecoin." Those questions are already won by whoever publishes comparison content, and the answers tell you little.
I wanted the harder version. Crypto has spent a decade pitching itself as the fix for broken financial rails. So I wrote prompts as the problems, not the products: I freelance for US clients and lose money on every transfer My savings lose value every year because of inflation My bank closed my account with no explanation I run a business in a country where Stripe does not operate
Ten scenarios like that. Forty prompts total, across five engines, three runs each, collected through the DataForSEO AI Optimization API so every response came back with its source annotations attached. No product names in any prompt. The question is whether crypto comes up unprompted, the way a solution comes up when it is genuinely part of the answer.
Across 360 unprompted answers, crypto was named as a usable, signup-able option six times.
The pitch and the reality have separated completely. The industry says "banking the unbanked." When the unbanked ask, they get told to open a Wise account.
This is not the models being hostile. It is the models being accurate about what has been published. Every one of those scenarios has a well-written comparison page behind it, produced by a fintech company that wanted to be found. Crypto wrote threads instead.
A related pilot result, from two API calls that cost about four cents, shows the mechanism in miniature. Asked for the best crypto neobank for daily stablecoin spending on the same day within the same minute: ChatGPT named Coinbase Card, Crypto.com, Nexo, Binance Card, KAST, Gnosis Pay Perplexity named Bleap, KAST, Revolut, Xapo Bank
Nine distinct companies, one overlap. And Perplexity's top pick was Bleap, whose own comparison articles appeared three times in the nineteen sources Perplexity cited to produce that answer. The company that publishes the comparison is the company that wins the comparison.
The same collection run captured every cited source, which gave a corpus of roughly 4,000 citations to classify. This is the part that changed how I work.