Ranking first for a query does not mean being cited in the AI answer to it. Those are two different competitions, and for brands operating across several markets the second one is decided separately in every language.
The thing most AEO advice gets wrong
Most writing on this topic assumes that being cited is a reward for ranking well. It is not. Engines synthesise an answer from sources they treat as authoritative for the question, and that set is often noticeably different from the top ten organic results.
Run the exercise yourself and the pattern is obvious within an hour: for a lot of commercial categories the cited sources skew heavily toward review sites, trade publications, forums, community answers and reference sites, and away from the vendors those sources are discussing. Your competitor is not outranking you in the answer. A publication you have never pitched is.
How engines appear to choose
Nobody outside these companies knows the weighting, and anyone claiming otherwise is guessing. But the observable behaviour is consistent enough to act on:
- They cite passages, not pages. A single well-formed paragraph that directly answers the question gets pulled; the rest of the page is irrelevant to that citation.
- They prefer sources that are not selling. Between a vendor page and a comparison article covering the same ground, the comparison article usually wins.
- They reward specificity. A claim with a number, a date or a named constraint is more quotable than the same claim stated generally.
- They need to be able to read you. Content that only exists after client-side rendering is content some engines never see.
- They reuse the same sources across related prompts. Once a source is established as the authority for a topic cluster, it appears repeatedly, which makes early presence in those sources compound.
The multi-market part nobody measures
Here is the finding that matters most if you sell in several countries, and the one I see missed almost universally: engines answer the same question differently by language.
Ask about a category in English and in German and you frequently get different cited sources — not translations of the same sources, genuinely different ones. That follows directly from how these systems are built: different language corpora, different authorities, different volumes of available material.
The practical consequences:
- Measuring in English tells you about one market. If you are running a prompt set only in your home language, you have an anecdote about the others.
- Your target source list is per market. The publication that dominates answers in Spanish is not the one that dominates in Dutch.
- Smaller language markets are easier. Less material exists, so fewer sources compete to be the authority. If you are going to win this anywhere first, it will be in your smallest market, not your largest.
That last point is the actionable one. Most brands attack this in English, where the competition is hardest, and ignore the markets where a single well-placed piece could make them the cited source for a whole cluster.
What actually moves it
Be present in the sources engines already trust
Establish which third-party sources are cited for your category, per market, and work to be represented there. This is digital PR with a different target list — and increasingly the same list, which is a genuinely new development in this work.
Make individual passages quotable
A direct answer in the first hundred words. One claim per paragraph. Specifics rather than generalities. A named author and a review date adjacent to the claim. You are writing for extraction, not for a scroll.
Fix the entity layer
Consistent name, role and description everywhere you appear, with Organization and Person schema and
sameAslinks tying the profiles together. Engines that cannot resolve who you are will not assert things about you.Answer the questions buyers actually ask
Including the uncomfortable ones — price, limitations, who you are not for. Those questions get asked of engines constantly, and pages that answer them honestly get cited because so few exist.
Measure per market, on a schedule
A fixed prompt set, run monthly, per language. Without a baseline you cannot tell improvement from variance, and these systems vary a great deal between runs.
Stuffing pages with question-shaped headings that answer nothing. Publishing volume in the hope that something gets picked up. Schema markup describing claims the page does not actually make. And any tactic premised on a specific engine's current behaviour — these systems change without notice, and the only durable strategy is being genuinely the best source on a question.
Where to start if you are behind
Run a prompt set in your smallest market first. It costs the same as running one in your largest and tells you more, because the competitive set is thinner and any movement is visible sooner. Establish who gets cited, take the top five recurring third-party sources, and work out what it would take to be present in them.
That is the first module of an AI visibility audit, and in most categories it produces a shorter, more actionable list than the ranking report it sits next to.