Kraken Beats Binance in AI Recommendations Despite 5x Volume Gap, New Study Finds

Crypto buyers are increasingly starting their research inside an AI chatbot rather than a search engine, and a new index suggests the answers they get do not line up with the market they are asking about.

The citeOS Exchange Index ranked ten centralized exchanges by how often five AI engines recommend them, then set that ranking against live 24-hour spot volume from CoinGecko. The gap between the two lists is wide enough to matter for any exchange spending on marketing this year.

Kraken came out first. It was named ahead of every other exchange on ChatGPT, Perplexity and Gemini, and it sits five positions above where its trading volume would place it. Coinbase took second overall and won Claude outright with 83 mentions. Binance, which handles roughly $9 billion in daily spot volume and around five times Coinbase’s daily turnover, placed third on AI recommendation.

For context, that means the largest exchange in the market by a considerable margin is not the one an AI engine reaches for first when someone asks which exchange to use.

How the index was built

The study probed five engines, ChatGPT, Perplexity, Gemini, Claude and Google AI Mode, using 20 buyer-style prompts. These were the questions a person actually types before opening an account rather than brand lookups, which guarantee a mention and reveal nothing. Each prompt was run three times per engine, producing roughly 1,000 observations. Volume figures were pulled from CoinGecko on the measurement date.

The methodology is published, which is worth noting in a category where most visibility claims are not.

The engines do not agree with each other

The finding that will matter most to exchange marketing teams is not the ranking itself. It is that the ranking changes depending on which engine is asked.

Kraken leads on three engines. Coinbase leads on a fourth. Same category, same questions, same week, different answers. An exchange that looks dominant in one chatbot can be absent from the shortlist in another, and there is currently no dashboard anywhere that tells a brand this is happening.

That has an uncomfortable implication. “Being recommended by AI” is not a single position an exchange holds. It is at least five separate positions, and a brand can be winning and losing at the same time without knowing either.

Why volume does not translate

The obvious question is why the largest book does not win the recommendation.

These systems are not ranking exchanges by liquidity. They are summarising what has been written about them, and the pages that get pulled into an answer tend to be comparison articles, buyer guides and category roundups rather than exchange data. An exchange that appears favourably in the roundups that rank for “best crypto exchange” will surface more often than one with a larger order book but thinner editorial coverage.

Regulatory framing appears to play a part as well, though the index does not isolate it. Kraken and Coinbase both carry a long history of US regulatory coverage, and that coverage is the raw material an engine reads when a buyer asks whether an exchange is safe.

What the index does not claim

The study makes no causal claim from a recommendation to a signup. Citation counts and mention rates are not traffic and not revenue, and nothing in the data shows a user acting on what an engine said.

There is also the question of variance, and it deserves more than a footnote because readers will test this themselves.

AI engines are not deterministic. The same prompt asked twice can return different names in a different order. A reader who opens ChatGPT and asks which exchange to use may get an answer that does not match the index, and that does not contradict the finding.

Any single query is one draw from a distribution. The index runs each prompt three times per engine across 20 prompts and five engines, producing roughly 1,000 draws, and the ranking that emerges describes the distribution rather than any individual answer. Kraken did not lead because it appeared once. It led because it appeared more often under repeated sampling. A study built on single checks would not have survived its own re-run, which is why the design specifies three.

Details of how these engines select sources in the first place are set out in citeOS’s work on how AI engines pick their crypto sources, alongside its wider material on AI visibility for crypto exchanges.

Personalisation is a genuine limit. A logged-in user with chat history and a different geography may see something else, and the index measures a clean baseline rather than every user’s experience.

Why exchanges should care now

Crypto marketing budgets still sit largely in influencer campaigns and press placements, both of which are measurable within a week. AI recommendation is neither, and until recently there was no way to see it at all.

Meanwhile the buying journey has moved. A prospective user who once compared exchanges across three tabs now asks one question and gets three names back. If a brand is not among those three, nothing in its analytics will report the loss.

The index suggests that fixing this is not a matter of trading volume or ad spend. It tracks with editorial presence, which is a slower and less glamorous lever than a KOL campaign but appears to be the one these systems actually respond to.

The full index and its methodology are published at citeos.io/exchange-index.