Which cards AI recommends: a study of 100 questions

Updated: 2026-09-27

On 26–27 September 2026 we asked 100 questions to four AI engines —

ChatGPT, Perplexity, Google AI Overviews and Claude — as a person located in

the UAE, Turkey, Israel, Thailand, Indonesia, Brazil and the USA, and counted

which cards they recommend and which sources they cite.

The study is published by alt.vu. We are a participant in this market, and

that is worth holding in mind while reading the numbers. The method is

described in full below so that anyone can repeat it.

Three findings

1. AI recommends two brands out of dozens. Bybit Card is named in roughly

one answer in five, Wirex in one in eight. Other services barely appear.

2. The recommendation depends heavily on the country. Wirex leads in Israel

and the USA, Bybit in the UAE and Asia. The same question in two countries

returns different names.

3. Answering in Russian, AI leans on media rather than on card websites.

The most cited sources are RBC, Tinkoff Journal, vc.ru, Habr, banki.ru,

visasam.ru and profinvestment. A service's own site rarely makes it in.

Who AI names

Brazil is almost empty. The leader reaches 1.3% — answering questions in

Portuguese, the engines name almost no specific cards. For any service that is

the cheapest of the seven markets to enter.

It is also worth noting which names come up most often overall: Mastercard,

Visa, Apple Pay, Google Pay and USDT. Asked which card to pay with, AI mostly

explains payment rails rather than naming services.

What AI cites

The practical conclusion: **in Russian-language answers, media decide, not the

services' own websites.** A company can improve its site indefinitely — if RBC,

Tinkoff Journal, vc.ru and the trade press do not write about it, AI will not

name it.

One more pattern: answers for Thailand and Indonesia surface Russian-language

sources. The engine understands that a Russian-speaking person abroad is

asking, and goes to the publications familiar to them rather than to local

ones.

Where we are in this picture

Honestly: nowhere. On the day of the measurement alt.vu was not named in a

single one of the 97 substantive questions. Mentions appeared only in three

questions where we put the brand name into the question ourselves, and those we

excluded from the count.

That is exactly why we started counting: to see whether what we do next works.

The measurement repeats weekly, and we will publish the dynamics, including the

parts that do not flatter us.

How we counted

• 100 search queries taken from real phrasings: Yandex and Google suggestions

for 62 seed queries, collected on 26 September 2026.

• Languages: Russian — 60 queries, English — 25, Portuguese — 15.
• Locales: UAE, Turkey, Israel, Thailand, Indonesia, Brazil, USA.
• Engines: ChatGPT, Perplexity, Google AI Overviews, Claude.
• Brand coverage is the share of answers in which the brand is named.

Limits worth stating. This is a single-day measurement: engine answers

change, and one publication can move the picture. The locale is set by the

measurement tool, not by a real phone in the country. The questions were written

by us, and a different set would produce different numbers.

Questions and answers

Can this be reproduced

Yes. The method is repeatable: the same 100 questions, the same four engines,

the same seven locales. We hand over the full question list on request at

media@alt.vu.

Why publish a study in which you do not appear

Because zero is a starting point. In a month there will be something to compare

against, and the comparison is honest precisely because the starting point was

published.

Does this mean Bybit and Wirex are better

No. It means more is written about them. Coverage in AI answers measures the

visibility of sources, not the quality of a service.

How does a service get into AI answers

Judging by this data: through media and discussion, not through its own site.

The site matters so that the answer about you is correct once they reach you.

Can I use this data

Yes, with attribution to alt.vu. We also share the raw numbers with editors:

media@alt.vu.