Research
What AI engines actually say about crypto categories, measured across ChatGPT, Claude, Perplexity and Google AI Overviews. Every finding ships with the data underneath it. Both editions are written up together in The Crypto AI Visibility Playbook, a free PDF with a 30-day plan attached.
Published
The same ten questions asked from India, Brazil, Nigeria, Indonesia, the Philippines, Turkiye, Korea, Vietnam and the United States. Nothing changed but the country. Thirty-four different cards got named, two established ones never did, and one engine answers as though geography does not exist.
Read the report →Or go straight to the evidence: all 1,020 answers, with every product named and every domain cited, free and ungated.
Ten ordinary money problems that stablecoin rails genuinely solve, put to four engines in plain language, then asked again with crypto named. The engines raised crypto 35% of the time. They named a product a person could actually use six times out of 360.
Read the report →Or go straight to the evidence: the full dataset, 480 responses and 497 classified warnings, free and ungated.
Edition one, findings in detail
In progress — protocol visibility reports
The index looks at a whole category. The next series looks at one protocol: every question its buyers actually ask, run across all four engines, scored on who gets named and where the answer came from. Same method, same published data, one name at a time.
If you run a protocol and want yours measured, say so. The measurement is the same whether or not anything comes after it.
How these are done
Every prediction is written down before a single call is made, and every one gets published afterwards whether it held or not. Edition one registered six hypotheses and two of them failed outright, including the one the whole thing was built around. Both are printed at full size.
The raw responses go out with the report, so anyone can rescore it and reach a different answer. Where the numbers argue against the case Kunzum makes commercially, they still go in — edition one found that AI warnings about crypto are more accurate than warnings about traditional finance, which is not the finding a crypto marketing studio would have picked.
That is the whole reason to publish research rather than case studies. A case study is something you have to take on trust. This you can check.
All of it is run by one person who spent six years inside crypto before starting to measure it from the outside, using the method these findings feed back into. The scoring scripts and every raw response ship with each edition, so the checking is not hypothetical.
Each edition also has a category page built on it, for readers who want the commercial reading rather than the study: stablecoin rails from edition one, and crypto cards and neobanks from edition two.