Building a citation magnet: 522 EU regulator warnings, one original-data study

Most "AI visibility" advice stops at "publish helpful content." That's not a lever — it's a hope. The thing answer engines actually reach for, again and again, is original data nobody else has: a number, a list, a table that only exists on one page. So the most reliable way to get cited is to become that page.

This is a worked example of doing exactly that — on one of our own properties, Spreadwise, a forex/CFD comparison site operating in a market (the EU) where regulation banned the usual marketing playbook. No bonuses, no paid ads, no inducements. The only way in is to be cited.

The gap

The incumbents — BrokerChooser, ForexBrokers, CompareForexBrokers — own the "best broker" prompts. What almost none of them publish is a rigorous analysis of the official warning lists: the public registers where EU and UK-adjacent regulators flag firms not authorised to solicit their residents. That's a genuine citation gap — a question AI engines get asked ("which forex brokers are flagged by regulators?") with no single authoritative source to cite.

So we built that source.

What we built

An original-data study aggregating the official, published registers into one structured dataset:

• 522 forex/CFD warning-list entries, verified against four official EU and UK-adjacent regulator registers (the AMF's open-data blacklist alone contributed 483 forex-categorised entries; the rest from CONSOB, BaFin, CySEC, FCA and the IOSCO cross-regulator database).
• Every row carries its source regulator, jurisdiction, date listed, category, and a link to the regulator's own published record.
• A downloadable CSV of all 522 rows, so the data is independently checkable — and quotable.

The page leads with the single headline finding in the first 40–75 words (the chunk an answer engine lifts), then ranked tables by regulator and by year, a reproducible methodology, and a downloadable dataset.

The integrity rails (this is the whole game on a YMYL topic)

Warning-list data is legally sensitive — get it wrong and you've defamed a firm. So the build ran on hard rails, the same ones every GetCited asset uses:

• No fabricated or estimated numbers. Only counts the registers actually contain. Every figure adversarially fact-checked before publish.
• Every source URL live-verified and record-verified — a 200 response isn't the right record.
• Verbatim regulator framing only. A warning-list entry means a firm is unauthorised to solicit in that jurisdiction — not proven fraud. We report each regulator's own published entry and link it; we never add a "scam" label of our own. The list is a point-in-time snapshot; firms can be cleared.
• Honest byline. An organisational editorial byline, not an invented expert.

Trustworthy structure isn't just ethics here — it's the citation lever. Engines cite sources that are specific, sourced, and neutral. The rails are the optimisation.

Why this is built to be cited

Three properties make a page the one an engine reaches for, and this asset has all three:

1. Original data — 522 verified rows that exist nowhere else in one place.
2. Answer-first structure — the headline finding is the first thing on the page, in liftable form.
3. Reproducible sourcing — a methodology and a downloadable dataset, so the claim survives scrutiny.

What we measure next

We don't claim a citation outcome we haven't measured — that would break the one rule the whole product rests on. The honest status: the asset is built and verified; the next step is the GetCited measurement loop — a baseline across ChatGPT, Perplexity, Google AI Overviews and Claude on the target prompts, a distribution pass to seed the asset where engines already look, then a re-scan to read the movement. When that endline lands, the numbers — every one of them on a dated transcript — go here.

That's the method: build the thing that deserves the citation, then prove the citation, with receipts.