One study, three lenses: what AI tells people about money in the Netherlands

Why one study needs three lenses

In October 2026 we asked ChatGPT, Gemini and Claude eight questions that people in the Netherlands ask about money: mortgages, box 3, pensions, investing and the 30% ruling. Four were in Dutch and four in English. Each ran ten times on each engine, with a simulated buyer who added their details and then asked who to go to. That made 240 conversations and 720 answers.

The study reports what happened. This piece asks what it means.

Take the first finding: a firm that publishes its price is far more likely to be named when a buyer asks what advice costs. That one fact is three things at once.

• People & Behaviour. "What will it cost?" is the question a buyer asks just before choosing.
• Technology & AI. Two of the three engines search for a page with figures on it. A fee inside a PDF or behind a form is invisible to them.
• Brand & Narrative. When your fee is missing, the engine quotes a market range instead, and your value is judged against the average.

Sales knows the first. The web team can fix the second. Marketing decides the third. Fix it from one seat and the other two undo it, which is why we read every finding through all three.

Three gates, in plain words

While it answers a buyer, an AI engine makes three decisions about a business. We call them gates, and each has a small set of outcomes, which we call states.

Examples from this study:

• Credited: Viisi on Gemini's expat mortgage answers, cited and put forward.
• Unsourced: Gemini naming firms and prices without linking anything.
• Ghost-cited: KRC van Elderen's pages fed 22 Gemini conversations, and the firm was never named.
• Overlooked: a Dutch-only adviser in an English conversation, or Gemini telling a buyer that paid help is not worth it. We call the second case framed out.

The study did not check firms' own facts, so for Gate 2 we describe what each state would look like rather than give examples.

A firm can pass one gate and fail the next for the same buyer, and buyers do not move through the gates in order. Someone who asks about you by name skips straight to Gate 2. So we read each gate on its own and never add them into one score.

Ten findings, read three ways

Does publishing your fees help a financial adviser get recommended by AI?

Yes, on two of the three engines. When the buyer asked what advice would cost, ChatGPT reached 27 named firms through a price or tariff page and Claude 25. Gemini reached 11, and searched in only 15 of the 56 answers where it listed firms.

• People: the cost question comes right before the choice. A firm with no visible fee drops off the list at that moment.
• Technology: ChatGPT and Claude look for figures in plain text. Gemini answers with a range from memory.
• Brand: without your fee, the buyer hears the market average, not your offer.
• Gate · state: Understood · Undersold, which usually ends in dropped later. You were in the conversation until the buyer asked the one thing your site does not say.

What to do: advisers working on products under the commission ban already give clients a vergelijkingskaart, the fixed-format card that compares services and costs. Put the same figures, and what they include, on a normal web page.

Do English-speaking expats get recommended different firms than Dutch speakers?

Yes. Depending on the engine, only 11% to 19% of brands appeared in both Dutch and English conversations. ChatGPT named no firm in any of its 40 Dutch first answers, against 9 of 40 in English.

• People: an expat on the 30% ruling and a Dutch homeowner are different customers with different questions.
• Technology: each language draws on different sources, and engines behave differently in each.
• Brand: you have one reputation per language. A firm that only publishes in Dutch barely exists for an English-speaking buyer.
• Gate · state: Found · Overlooked, per market.

What to do: give each service an English page that answers the expat's version of the question, not a translation of the Dutch page.

Can you trust the mortgage figures AI gives?

Not as a single answer. For the same expat couple on €110,000, Claude's maximum mortgage ranged from €400,000 to €550,000 across ten runs. The €475,000 flat they wanted was out of reach in one conversation and within reach in another. ChatGPT declined to give a figure on most personal questions, with this mortgage as the one exception.

• People: buyers anchor on the first number they see, and it decides whether they call an adviser at all.
• Technology: an estimate that is not grounded in a cited source moves from run to run.
• Brand: the engine sets an expectation your adviser has to correct.
• Gate · state: Actionable · arrives misinformed. The buyer reaches you with the wrong picture.

What to do: publish the rules and worked examples you stand behind, so engines have a source to quote. Brief advisers that new clients may arrive with an AI's number, and ask where it came from.

Does ChatGPT recommend financial advisers if you do not ask for one?

Rarely. When the buyer only asked "what's my first step?", ChatGPT put a firm forward in 1 of 24 conversations, Claude in 6 and Gemini in 16, with 4.5 firms on average. Where the buyer did ask for names, Gemini had already named a firm in 66% of conversations, Claude in 27% and ChatGPT in 7%.

• People: most buyers ask how to start, not who to call.
• Technology: ChatGPT sends buyers to the Belastingdienst and the AFM first and keeps names back. Gemini names firms in its first answer.
• Brand: the route in differs per engine. On ChatGPT you win as the explainer it cites; on Gemini, as a name in its early list.
• Gate · state: Found, and for buyers who never ask, often not considered.

What to do: test with the question your buyers ask, "what's my first step?", on each engine, and write the page that answers it well.

Why does every AI give a different list of firms?

Because the lists are not settled. Across the three engines 252 commercial brands were named, and only 27 appeared on all three. About 40% appeared in a single conversation and never again. The one exception was the pension question, where Brand New Day and Meesman were put forward in every run on every engine.

• People: couples and families compare answers from different apps.
• Technology: each engine builds its own list, and the list changes between runs.
• Brand: your position is a share across many answers, not a single screenshot.
• Gate · state: across all gates, as how stable your position is.

What to do: ask your key question five times on each engine and count how often you appear. Do not react to one answer.

Which AI tells buyers to check the AFM register?

ChatGPT and Claude, in 13 and 23 of the 56 answers where the buyer asked for names. Gemini never did. ChatGPT linked a source in 96% of its answers, Claude in 81% and Gemini in 63%. Gemini named firms and prices from memory and never once flagged them as unverified. Claude said it had not checked the firms it named in 43 of 56 conversations.

• People: on two engines out of three, buyers are told to check whether an adviser is registered.
• Technology: each engine handles sources and trust in its own way.
• Brand: being verifiable is now part of your reputation. On Gemini, what it "knows" about you goes unchecked.
• Gate · state: Found · Unsourced when you are named from memory, and Understood · Misrepresented when that memory is out of date.

What to do: make sure your AFM registration, trading name and website use the same name. Keep directory listings and profiles current, because that is what an engine remembers.

Does AI talk buyers out of paying for financial advice?

Sometimes. Gemini told buyers that paid help was not worth it in 15 of 80 second answers, mostly on box 3 and the first tax return. On investing, all three engines steered towards a global index fund through a low-cost broker or a ready-made plan.

• People: buyers take the textbook answer, especially when it saves money.
• Technology: engines settle on the most common advice in their sources.
• Brand: a tax firm or active adviser can be ruled out before any firm is compared.
• Gate · state: Found · Overlooked, framed out. Your kind of service never comes up.

What to do: publish an honest page on when someone needs help and when they do not, with the situations where it pays. It is the page engines cannot find today.

Do big banks or specialists get recommended more?

It depends on the engine and the question. ING, Rabobank and ABN AMRO appeared in every Gemini run of both investing questions. On Claude, banks showed up mostly as cited explainer pages, not as recommendations. On expat tax, small specialists won: TaxSavers, Blue Umbrella, JC Suurmond and Broadstreet. The large accounting firms appeared only on Claude, and only for the 2027 question.

• People: the more specific the buyer's situation, the more they want a specialist.
• Technology: Gemini leans on well-known brands; search matches how specific the question is.
• Brand: a big name does not win niche questions, and a bank's explainer page can inform the buyer without the bank being chosen.
• Gate · state: Found for who is named, and Understood · Out-positioned for who looks like the better choice.

What to do: list the specific situations your best clients were in and give each its own page. Banks: put a product and a next step on your explainer pages.

Which websites does AI use to answer money questions?

Different ones per engine. ChatGPT leans on public bodies: belastingdienst.nl, afm.nl and rijksoverheid.nl. After the Belastingdienst, Gemini's most-cited sites are commercial: taxsavers.nl, krcvanelderen.nl, abnamro.nl and rabobank.nl. Claude leans on intermediaries such as iamexpat.nl, the Consumentenbond and comparison sites. Comparison platforms such as Advieskeuze, Independer and Trustoo were also put forward as the place to go.

• People: the proof a buyer is shown depends on the app they opened.
• Technology: each engine has its own source mix.
• Brand: intermediaries tell your story, and some of them also compete for the buyer.
• Gate · state: Found · Unsourced when others tell your story, and Actionable · via aggregator when the buyer acts through a comparison site.

What to do: check how you appear on the intermediaries in your field and correct them. Give buyers a clear reason and a clear way to come to you directly.

Can AI use your content without naming your firm?

Yes. KRC van Elderen's pages were cited in 22 Gemini conversations, mostly on expat tax and investing, without the firm being named once. Claude did the same with explainer pages from HomeFinance, Knab, Van Lanschot Kempen and Financer.

• People: the buyer learns from your expertise without learning your name.
• Technology: engines lift the facts and drop the credit.
• Brand: your best content is working for nobody.
• Gate · state: Found · Ghost-cited, which ends in not considered.

What to do: put your firm's name, the service and a next step on every explainer page, close to the facts engines quote.

The matrix on one page

Switch lenses to see one reading at a time.

Five things a Dutch financial firm can do this month

1. Put your fees on a web page. The figures from your vergelijkingskaart, in plain text, with what they include.
2. Publish in the language your buyer asks in. One English page per service, written for the expat's question.
3. Ask the engines what your buyer would ask. "What's my first step?" in Dutch and English, five times on each engine, and count how often you appear.
4. Match your names. AFM register, trading name, website and directory listings, all the same.
5. Sign your explainers. Your firm's name, the service and a next step on every page that explains a rule.

Not in finance?

The same three readings apply to any business whose buyers ask AI before they call. We wrote one piece per lens: how buyers decide inside an AI conversation, how AI engines decide which companies to name, and the version of your company AI tells.

Frequently asked questions

Do people in the Netherlands use AI for financial advice? Estimates differ. A bunq survey of 1,000 Dutch adults in March 2026 put it at 55%. The AFM's Consumentenmonitor 2026 put it at about one in seven over the past year.

Which AI engine recommends the most financial firms? In our study Claude named the most brands in total (153), then Gemini (142) and ChatGPT (56). Gemini put forward the most firms per conversation (8.0 by the end) and named them earliest. Claude named many brands but repeated few of them.

Is a financial adviser allowed to publish fees online? Advisers on products under the commission ban already have to give clients a vergelijkingskaart that compares services and costs. Publishing the same figures on a web page makes them readable for buyers and for AI engines. Check your own obligations with your compliance adviser.

How can I see what AI says about my firm? Ask the question your buyer would ask, without your firm's name, on ChatGPT, Gemini and Claude, in each language you serve. Repeat it several times, because answers vary. Then ask about your firm by name and compare the answer with the facts on your own site.

What are the three gates? Found (does AI name you), Understood (does it describe you correctly) and Actionable (where does it send a buyer who is ready to act). Each is read on its own, because a firm can pass one and fail another.

Where does your firm stand?

If you would like to know what ChatGPT, Gemini and Claude tell your buyers, we are happy to talk it through. In a 20-minute call we look at how a potential client would describe the problem you solve, what the engines say when they do, and whether optimising for the AI buyer journey is worth your attention right now. Sometimes the answer is that it can wait, and we will say so.

The call is free and you leave with a first read on your position either way.

Book a 20-minute call

Sources

1. CheckMarque AI, "Three AI engines, eight Dutch money questions," 10 October 2026. 240 conversations and 720 answers across ChatGPT, Gemini and Claude.
2. AFM, "Vergelijkingskaart." https://www.afm.nl/nl-nl/sector/adviseurs-bemiddelaars-en-gevolmachtigde-agenten/themas/vergelijkingskaart
3. bunq survey of 1,000 Dutch adults, March 2026, via Accountant.nl. https://www.accountant.nl/nieuws/2026/3/helft-nederlanders-vraagt-ai-om-financieel-advies/
4. AFM Consumentenmonitor 2026, via Risk & Business. https://riskenbusiness.nl/nieuws/insurance/afm-financiele-zaken-lang-niet-altijd-op-de-voorgrond-bij-consumenten/