An FAQ built to be cited by AI

Context

Admirals operates under three licences and publishes its FAQ in 15 language versions. Search had stopped being the only way in. A growing share of the audience now asks the question inside ChatGPT or Perplexity, or reads Google’s AI Overview, and never opens a result page at all.

For a broker this costs more than it does for most businesses. The questions people ask before opening an account are trust questions: is this company regulated, will they accept someone from my country, what documents will they want from me. If a model answers those without you, the customer decides somewhere you have no presence.

Problem

The FAQ had been written for a search engine and for a customer who already had an account. It answered branded, procedural questions: how to open an account, what a spread is, how rollover works. Useful to someone already inside the product, invisible to the way people now ask.

The queries that decide a broker choice look nothing like that. They are unbranded, comparative, and phrased as whole questions: which brokers accept French residents, what KYC documents do brokers require, is this broker safe. Nobody on the site answered them, so nothing could be cited.

A second constraint sat on top of the first. In a regulated environment you cannot fix this with promotional copy. Every claim has to survive compliance review, and marketing language is exactly what a language model discounts when it picks a source.

Approach

1. Wrote to the question, not the keyword

I built new sections around the unbranded queries people actually type: Choosing a Broker, Geographic Questions, Legal & Compliance, Education & Tools. Each heading is a full question in natural phrasing, because that is the form the query arrives in.

2. Put the answer in the first sentence

Every entry opens with a statement a model can lift whole. “Yes, negative balance protection applies.” “ESG stands for Environmental, Social, and Governance.” The explanation comes after it. A model takes the first sentence and has a complete answer; a reader who wants the detail keeps going.

3. Hedged honestly, on purpose

For comparative questions the answer states that there is no single best broker, sets out the criteria a trader should weigh, and places Admirals inside that frame instead of at the top of it. This runs against the instinct of a marketing page, and it is the decision I would defend hardest. Language models favour balanced, criteria-based sources over self-promotion, and competitors publishing “we are the best broker” ruled themselves out of the answer.

4. Made the entity structure explicit

Three regulated entities, each named, each linked to its own policy documents. A vague claim about being “regulated” gives a model nothing to quote. A specific, verifiable, attributable one is what it reaches for on a compliance question.

5. Held the architecture across 15 languages

The same answer structure carries into every market, so the FAQ works in French and Spanish the way it works in English rather than degrading into a translated wall of text.

Results

I measured it myself, with a method anyone can repeat:

1. Twenty target questions taken from the published FAQ sections, phrased the way a user would type them, unbranded.
2. Each question run through ChatGPT, Perplexity, and Google AI Overviews.
3. Recorded whether the domain came back as a cited source in the answer.

15 of the 20 questions now return the domain as a cited source.

Published FAQ: admiralmarkets.com/faq/general-faq

My contribution

• Selecting the unbranded, comparative queries worth answering
• Designing the section architecture around questions instead of keywords
• Writing every answer to open with an extractable first sentence
• Making the case for criteria-based answers over promotional ones
• Naming and linking the three regulated entities for compliance questions
• Holding one answer architecture across 15 language versions
• Building and running the 20-question citation test across three AI surfaces

---

This case describes work I led at Admirals between 2025 and 2026 on FAQ content and AI-search visibility. Published FAQ copy is public; test questions and internal reasoning are paraphrased to respect employer confidentiality.