A quarter of people in Germany have asked AI about money. What should banks make citable?
Bitkom's May survey shows a quarter of people in Germany have asked AI about money. The fee, product and disclosure pages banks should make easy to quote.
The short answer
Banks should make citable the facts an engine would otherwise guess: current fees, product terms with their dates, deposit protection, adviser status and how advice is paid for. Publish them as plain text in German as well as English, one fact per heading. Then test German-language prompts as their own sample, because answers differ by language and country.
Key takeaways
- In Bitkom's survey of 1,004 people aged 16 and over in Germany (March to April 2026), 25% had already asked an AI tool such as ChatGPT about their finances.
- In the same survey, 62% worried that AI increases the risk of fraud, so trust signals belong on the pages engines cite.
- In Semrush's analysis of US prompts (January to April 2026), the top three brands held 41.4% of AI visibility in finance, so answers concentrate on a few names.
- Fee tables, dated product terms and regulatory disclosures are the facts an engine quotes or gets wrong, and all three sit on pages a bank already owns.
- German-language prompts are a separate sample from English ones and need their own baseline.
In this article
What did Bitkom find?
Bitkom found that asking AI about money is already common in Germany. In its telephone survey of 1,004 people aged 16 and over, run by Bitkom Research in weeks 11 to 14 of 2026 (March to early April), 25% said they had already consulted an AI tool such as ChatGPT about their finances.1 Bitkom published the results on 21 May.
Willingness runs ahead of trust. In the same survey, 27% said they could imagine letting AI make most of their financial decisions, while 62% worried that AI increases the risk of fraud.1 Bitkom's president framed it as a matter of trust: AI applications in finance must be transparent, secure and understandable.
For a bank, a savings provider or an independent adviser, the practical question is narrower. When someone in Germany asks an assistant which current account has no monthly fee, or whether a fee-only adviser is worth it, what does the answer quote? Increasingly it quotes whichever page states the fact most plainly. This post covers which facts to publish that way, and how to check the result.
Germany, March to April 2026
Asking AI about money is already common
Why finance answers concentrate on a few names
Finance answers name few brands, so being one of them matters more than in most categories. In Semrush's 2026 AI Visibility Index, built on 126 million US prompts from January to April 2026, the top three brands held 41.4% of AI visibility in finance.2 In news and media the same measure reached 82.9%. These are US figures, because the index covers US prompts only.
The engines also differ in how many sources they draw on. In the same data, ChatGPT cited about 15 sources per response against about three for Gemini.2 A narrow answer leaves little room for a bank that is not already among its sources.
Our view: in a regulated category, engines lean on pages that state a fact plainly and carry an institution's name. Comparison portals, consumer-advice bodies and the banks' own price lists all fit that description. A bank whose fees exist only inside a PDF is relying on someone else to restate them correctly.
Which facts should a bank make citable?
Make citable the facts a customer would otherwise get from a guess: what it costs, what the product is, and who stands behind it. Each one should sit on a crawlable HTML page, in plain sentences, with the date it took effect.
Fees and prices
Most German banks already publish a price and service list, the Preis- und Leistungsverzeichnis, usually as a long PDF. Keep the PDF, and add an HTML summary of the ten fees customers ask about most: account fee, card fees, foreign transactions, overdraft rate, and so on. Give each its own row and the date it applies from.
Product terms with their dates
Interest rates, minimum terms and conditions change, and an engine quoting last year's rate hurts the customer and the bank. State the current rate with "valid from" next to it, and keep an archive of earlier rates on a separate page so old figures are clearly labelled as old.
Protection and regulatory status
Customers ask whether their money is safe and who regulates the provider. Say which deposit protection scheme covers which products, which supervisor the firm answers to, and where the key information documents for investment products live. Link the official register entry rather than paraphrasing it.
Advice and how it is paid for
For advisers, the most asked question is how they earn their money. State plainly whether advice is fee-based (Honorarberatung) or commission-based, what an initial meeting costs, and what qualifications the advisers hold. An engine cannot recommend a fee-only adviser it cannot identify as one.
| Fact | Where it usually lives | What to add |
|---|---|---|
| Account and card fees | Price list PDF | HTML summary, one fee per row, dated |
| Savings and loan rates | Product page, image or widget | Rate as text with "valid from" date |
| Deposit protection | Footer link, legal page | Plain statement per product line |
| Supervisor and register | Imprint | Named supervisor, link to register entry |
| Adviser pay model | Sales conversation | Fee or commission, stated on the team page |
Our checklist for a first pass. Illustrative; your own audit decides the order.
Where Google's AI Overviews come in
Google's AI Overviews add a second surface for the same questions, and money questions are among the kind that trigger them. Ahrefs found AI Overviews on 20.5% of 146 million desktop results pages in September 2025, data older than the other figures here.3 The rate was 57.9% for question queries and 44.1% for medical queries in the "your money or your life" group.3 Ahrefs did not break out finance in that figure, and the study did not state a geography.
The pattern still matters for banks. Customers phrase money worries as questions, and question queries are where Overviews appear most. The fix is the same as for chat assistants: plain, dated facts on pages Google already ranks.
Why German prompts need their own sample
German-language prompts give different answers from English ones, so measure them separately. An assistant asked "Welches Girokonto ist kostenlos?" may draw on German comparison portals and consumer-advice pages. The same question in English may lean on international sources and miss a regional bank altogether.
Treat each language and country as its own sample with its own baseline, as our measurement spec recommends. Mixing them averages away the difference you most need to see. A useful starting set is 20 to 30 prompts in German that customers actually ask, written by the people who answer the phone.
When an engine states a wrong fee or rate, trace and correct the source following our guide to fixing wrong AI answers. In finance, a wrong number can reach a customer as advice, so log each correction and the date you checked it.
How to check progress
Check progress on a fixed prompt set, run often enough to separate change from noise. At Sigzen AI, our AI answer monitor runs each prompt on six engines (ChatGPT, Gemini, Perplexity, Claude, Copilot and Google AI), through official APIs where they are offered and a licensed AI-answer data provider otherwise. Audits run each prompt on three separate days, because one run of a prompt is an anecdote.
Our view: for a bank or adviser, the first metric to track is not share of answer. It is whether the facts in answers that mention you are correct and current. Visibility that repeats an old fee does more harm than absence.
