Banking and Insurance: What ChatGPT Recommends to the French

"Best online bank," "cheap home insurance": when someone asks ChatGPT these questions, the answer almost always names a comparison site or broker before it names a specific bank or insurer. Retail banking and insurance are also among the most tightly regulated sectors for AI: French regulators AMF and ACPR impose specific rules on tools that interact with customers, and no generated answer can stand in for personalized advice.

These topics fall under what Google calls YMYL, Your Money Your Life: an inaccurate answer can have real financial consequences for the person reading it. But unlike a law firm or a doctor's practice, the entity being cited is almost never a named individual — it's a brand, most often relayed through a comparison site. Here is what AI engines recommend today on money questions, what AMF and ACPR regulate, and how a bank or insurer can exist in these answers without crossing a regulatory line.

Why banking and insurance are a different kind of terrain for AI

YMYL topics require a higher standard of proof from search engines and generative AI, which favor institutional or verifiable sources over anonymous content. We cover this mechanism, originally built around regulated professions, in our article on AI visibility for lawyers, doctors and experts.

Banking and insurance share this requirement, with one structural difference: the trust signals a law firm builds around a named professional (degree, bar registration) are built here around a regulated brand instead. The signals AI engines can verify are different: ACPR authorization, ORIAS registration, dated and comparable pricing data, verified customer reviews. A bank or insurer's website that doesn't surface these elements clearly misses out on the same kind of trust proof an individual YMYL professional relies on.

A second difference: the facts move fast. A loan rate, a coverage cap or a deductible can change within months, sometimes weeks. An AI that cites a stale figure makes a mistake the user can immediately verify, which pushes engines to favor sources that state their freshness explicitly over a page that never says when it was last updated.

What questions do people ask AI about their money?

Usage remains a minority behavior, but a real one. According to a study published in February 2026 by France's Autorité des marchés financiers (AMF), 11% of French savers say they use AI to inform themselves before making an investment, a share that rises to 19% among those under 35, compared with 42% who consult a bank or an advisor. Generative AI hasn't replaced the advisor; it has added itself as one information source among others, upstream of the decision.

The typical questions look like what people already type into a search engine, just phrased as natural language: "what's the best online bank for a young professional?", "cheap home insurance for a tenant", "should I cancel my car insurance to switch providers", "how does France's Hamon law work for switching insurers". In these answers, AI tends to compare categories of offers and criteria (account fees, coverage caps, waiting periods) rather than settle on one specific brand — consistent with the fact that a general-purpose model is neither a registered investment advisor (CIF in French law) nor a licensed insurance broker, and so has no business issuing personalized advice that carries legal liability.

Who dominates the answers: comparison sites, brokers and financial media

Comparison sites hold a structuring place in these answers, for a simple reason: their content is already built like an answer. Homogeneous criteria tables, dated figures, a published methodology, sometimes a monthly refresh. That is exactly the format an AI citation system is built to extract, the same principle we describe for product pages in our guide on how AI engines choose the products they recommend.

UFC-Que Choisir, an independent French consumer association, illustrates this format well: its banking and insurance comparisons rely on analyzing dozens or hundreds of contracts against homogeneous criteria and tested policyholder profiles, with a public methodology. An AI arbitrating between sources on "which home insurance to choose" has more reason to cite a structured, sourced comparison than a sales page that details neither its figures nor its method.

For a bank or insurer, the consequence is direct: if your own site doesn't present rates and coverage in a format as readable and dated as a comparison site's, you hand the citation to the intermediary that compares you, not necessarily to yourself. Customer reviews play the same role as verifiable proof; we cover how to structure them in our article on customer reviews and AI visibility.

What AMF and ACPR regulate about AI in banking and insurance

Financial services is one of the most tightly regulated sectors for AI use, and that regulation directly shapes what a brand can publish or automate.

No personalized advice from a general-purpose tool. In France, investment advice (conseil en investissement financier, or CIF) is a status regulated under the Monetary and Financial Code, requiring ORIAS registration and client documentation and protection duties. A general-purpose chatbot like ChatGPT, Gemini or Claude doesn't hold that status: its answers, even when useful to understand a topic, don't carry the liability of a licensed professional and should not be presented as individualized advice. The AMF study makes this distinction explicit, describing AI as a tool to help people inform themselves rather than a substitute for advice from an investment professional.

Mandatory transparency for chatbots since August 2, 2026. Article 50 of the EU AI Act, which took effect on that date, requires any system designed to interact directly with people to clearly inform them they are talking to an AI. This obligation applies in particular to the chatbots and conversational agents banks and insurers run on their own websites, even when those systems don't fall under the regulation's "high-risk" category, according to analysis from French law firm Haas Avocats.

Pricing systems are classified as high-risk. The ACPR (Autorité de contrôle prudentiel et de résolution) has been designated as the supervisory authority for high-risk AI systems used to assess risk and set pricing in life and health insurance, with supervision becoming effective from December 2027. On July 1, 2026, it also published a discussion paper on algorithmic fairness in financial services, open for public consultation until September 30, 2026, aimed at preventing the discrimination that pricing algorithms could introduce.

For content published on a website, the practical consequence is the same as for any YMYL content: stay factual, dated, sourced, and explicitly point toward a licensed advisor or a subscription journey as soon as a personal situation is involved, rather than letting text or an in-house chatbot imply individualized advice it isn't legally positioned to give.

→E-E-A-T and AI: Google and ChatGPT want the same proof

What this changes for a bank or insurer that wants to be cited

Getting cited by an AI on a "best bank" or "cheap insurance" question doesn't happen through a generic sales page. Five levers account for most of the impact.

• Structured, dated pricing and coverage pages. Amounts, caps, deductibles and conditions presented in a comparable format (table, criteria list), with a visible last-updated date. That's the format comparison sites already use, and the one AI engines cite most easily.
• Factual FAQs, never disguised as personalized advice. Answer real customer questions ("how do I cancel my car insurance", "what's the refund timeline") with short, sourced answers, FAQPage schema, and a clear pointer toward an advisor or a simulator for any individual situation. Our guide to GEO details the answer structure that favors citation.
• Authorization and registration made visible and structured. ORIAS number, ACPR authorization, complete legal notices: these are trust signals AI engines can cross-check, the same way a bar registration number works for a lawyer.
• Consistent presence on the comparison sites that already cite you. Up-to-date listing, accurate rates, correctly described coverage. A stale listing on a major comparison site can steer an AI answer against you, with consequences close to those described in our article on competitors cited by ChatGPT.
• Structured, recent customer reviews. On financial products, social proof weighs as much as price in the decision; see our article on customer reviews and AI visibility.

For a broader measurement that compares your brand to competitors across ChatGPT, Gemini, Claude and Perplexity, Beeleven's AI visibility audit, from the agency that built Detekia, asks the real questions a sector's customers ask (comparison, cancellation, claims) and measures who gets cited, in which position, and against which comparison sites or competitors.

What should you do in practice?

1. List your customers' and prospects' real questions. Comparison, cancellation, claims, eligibility: the questions your advisors already hear on the phone are the ones AI gets too.
2. Measure what ChatGPT answers about your brand today. Are you cited? Which comparison site is cited ahead of you? Methods are compared in our guide to measuring AI visibility.
3. Structure your pricing and coverage pages with dated, up-to-date tables and the right schema (Product, Offer, FAQPage depending on the content).
4. Make your authorization and registration visible on key pages, not only buried in legal notices.
5. Check the consistency of your listings on the comparison sites that already cite you: rates, coverage and cancellation terms up to date.
6. If you run a chatbot on your own site, check its compliance with AI Act Article 50 (clearly informing users they're talking to an AI) before any other optimization.

Frequently asked questions

Can a bank or insurer let its chatbot give personalized advice?

No, unless the tool and the entity operating it hold the required regulated status (registered investment advisor, insurance broker registered with ORIAS). A general-purpose chatbot or an unlicensed in-house assistant should stick to general information and point toward an advisor for any individual situation.

Why are comparison sites cited so often by AI on banking and insurance?

Because their content is already structured like an answer: homogeneous criteria, dated figures, a published methodology. An AI that has to synthesize a comparison finds a directly usable format on these pages, more easily than on a sales page that details neither its figures nor its method.

What's the risk for a bank or insurer if its chatbot doesn't comply with the AI Act?

Article 50 has required, since August 2, 2026, that users be clearly informed they're talking to an AI. Non-compliance exposes the operator to scrutiny from the relevant supervisory authorities; for pricing systems classified as high-risk in life and health insurance, the ACPR will be the supervisory authority starting in December 2027. This is a legal question: when in doubt, specialized legal advice remains necessary.

How can you tell if ChatGPT recommends your bank or insurer over a competitor?

By asking your customers' actual questions to ChatGPT, Gemini, Claude and Perplexity yourself, and noting who gets cited, in which position, and which sources appear alongside you. For a complete, repeated measurement across all four AI engines, Beeleven's AI visibility audit does this work and weights it by each AI engine's real usage in France.

Sources

• Autorité des marchés financiers, "The use of AI by French financial market participants," February 2026 (in French).
• Argus de l'Assurance, on the ACPR's new consultation on AI oversight (in French).
• Haas Avocats, on what applies to banks and insurers under the AI Act from August 2, 2026 (in French).
• UFC-Que Choisir, Money and Insurance section (in French, accessed October 2, 2026).

→E-commerce: Get Your Products Recommended by AI

→Lawyers, Doctors, Experts: AI Visibility and YMYL Content