Ask an AI engine an ordinary question about spending money abroad and a crypto card is named in 8 of 510 answers. Put the word stablecoin in the same question and it is named in 319 of 510, from an open field of 34 distinct products.
That gap is the whole opportunity in this category, and Kunzum measured it across nine markets in September 2026 — 1,020 responses, prompt text byte-identical everywhere, only the API location changing. The Crypto Card Report publishes all of it.
In short
- Unprompted, a crypto card appears in 8 of 510 answers. Prompted, 319 of 510.
- 34 distinct cards were named across 1,020 responses in nine markets.
- Most-named: Crypto.com (117), Coinbase Card (110), Kast (109), RedotPay (95).
- Two established cards were named zero times in any country, at either stage.
- Localisation is wildly uneven between engines, and one engine did it zero times in 180 answers.
Who currently holds the answers
Scoring was open-vocabulary: products were read out of the responses rather than matched against a shortlist, so a card absent from this table is absent because no engine said it.
| Card | Answers naming it |
|---|---|
| Crypto.com | 117 |
| Coinbase Card | 110 |
| Kast | 109 |
| RedotPay | 95 |
| MetaMask Card | 88 |
| Wirex | 78 |
The tail matters more than the head here. 34 distinct products were named across the corpus, which means the category has no settled default the way cross-border payments has Wise. In a category without a default, being the name an engine reaches for is still available.
Geography is a live variable
The same ten questions were sent from nine markets with no country named in any prompt. ChatGPT named a product specific to the asker's country in 85 of 240 answers and Google AI Overviews in 60 of 240. Perplexity managed eight. Claude did so zero times in 180 answers and never cited a local domain either.
For a card issuer this is actionable in a way most AEO advice is not. If your market is one where an engine localises, local sources are worth building. If your buyers use an engine that does not localise at all, local press will not reach them no matter how much of it you buy.
The one card any engine volunteered unprompted was Cardtonic, a Nigerian service, named four times without being asked out of five appearances total. The unprompted slot in this category is very nearly empty.
What this does not tell you
This measures who gets named. It does not claim a named card is available to you, or that an unnamed one should have been. Claude refuses to answer as though it were in Nigeria or Vietnam, so its figures rest on seven markets rather than nine. Asked in English on one day; Vietnam was the weakest market for local answers and asking in Vietnamese might change that entirely.
Two analyses from the first scoring pass were discarded before publication as artefacts of Kunzum's own counting rules, and are recorded as discarded. The full corpus is downloadable without an email gate.
Adjacent categories with their own measurement: stablecoin rails and perpetual DEXes. The approach is the method.
Sources
- Google Search Central, “AI features and your website”. Google. Checked 2026-09-12.
- Perplexity’s crawler documentation. Perplexity. Checked 2026-09-12.
- Aggarwal and colleagues, “GEO: Generative Engine Optimization” (2023). arXiv:2311.09735. Checked 2026-09-12.
- CoinGecko. CoinGecko. Checked 2026-09-12.
Published 2026-09-12. Written by Narender Charan, who runs Kunzum.