How financial companies are being read by AI, and what breaks

When somebody asks an assistant "what does a $400 loan actually cost in Ohio" or "which card has no foreign transaction fee", the model does not open your homepage and admire it. It pulls text. If your numbers live inside a carousel, a PDF, or a script-rendered pricing widget, they are not in the answer — a competitor's plain sentence is.

Finance is the worst-hit category for three reasons.

Numbers are buried in components. Rates, fees and limits are the single thing people ask about, and they are usually the single thing rendered by JavaScript. A crawler that reads HTML sees an empty container where the APR should be.

Disclaimers outweigh facts. Regulated copy is long, and models summarise what is dense and declarative. Three hundred words of legal text around one number teaches a model very little.

The real answer is state- or country-specific. "It depends where you live" is true and useless. Pages that state the variation explicitly — this state, this figure — get quoted; pages that hedge do not.

What a machine-readable listing fixes

A markdown listing is a flat page with no scripts: who you are, what you sell, the actual numbers, and a short set of question-and-answer pairs. That last part matters more than anything else on the page, because a question with a direct answer under it is the exact shape a model reuses when it answers a person. Write the question the way a customer would type it, then answer it in one or two sentences, with a real figure in it.

What is listed here

Every company below has a listing page and a markdown twin that crawlers can take whole, plus an open count of how many times each named AI crawler actually fetched it. Those counts are raw server logs. Some are small. We publish them anyway — a directory that inflates its own numbers is worth nothing to the people paying for it.