How do AI answer engines decide what to cite?
They do three things: retrieve a set of candidate sources for the question, rank them on quality, and generate an answer that cites the ones it leaned on. Google's AI Overviews and AI Mode run on Google's existing ranking and quality systems; tools like ChatGPT and Perplexity add their own retrieval over search and the open web.
So “how do I get cited?” decomposes into two familiar questions: is my page retrievable for this query, and is it ranked as one of the best answers? Both are problems SEO has always worked on.
Is AEO a different game from SEO?
No. AI visibility is not a separate discipline with its own tricks — it runs on the same ranking and quality systems, plus retrieval-augmented generation and query fan-out. The practical identity is: SEO quality is AI-citation quality. Optimize for genuine helpfulness and the citations follow; there is no shortcut that beats being the best answer.
This matters because a whole market is selling “AEO” and “GEO” as new tactics. Loudly positioning your own site around those acronyms can itself read as a spam signal. The operator's move is to do the work, not wear the badge.
What about llms.txt and schema — don't those help?
Honestly: there is no controlled evidence that either lifts AI citations. llms.txt is an emerging convention the major engines haven't confirmed using as a ranking input. Structured data genuinely helps eligibility for rich results and entity understanding — a real, proven job — but it is not a demonstrated citation lever either.
Treat both as neutral hygiene: cheap, sensible, worth doing for what they actually do — and not the thing that earns the citation. Anyone selling them as the lever is selling the bumper sticker as the engine.
What actually earns citations?
The durable moat is the same content quality that has always won, and that a content factory cannot fake: first-hand experience, original data and research, and transparent methodology. An answer that demonstrably knows the thing — because it was built, measured, or operated by the author — is what gets pulled into AI answers across tools, because it's the answer the model can't generate on its own.
That's why this estate publishes its own measured numbers, its own specifications, and its own methods rather than rephrasing the consensus. The citation goes to the source of something, not the summarizer of everything.
Query fan-out: why specificity wins
Answer engines rarely run your exact question once. They fan out — decompose it into sub-questions, retrieve for each, and assemble the answer from several sources. The implication is sharp: the way to be in the answer is to completely own one specific sub-question, not to half-cover a broad one.
That rewards a page that leads with the answer to a precise question and supports it thoroughly (see answer-first content) over a sprawling page that mentions everything and resolves nothing.
The operator's checklist (no snakeoil)
- Be the best answer to a specific question — first-hand, original, more useful than the alternatives.
- Lead with the answer and structure for a precise sub-query (see answer-first).
- Be retrievable — in the HTML, crawlable, fast (see build-time rendering).
- Use accurate schema for what it does; ship llms.txt as hygiene if you like — expect neither to be the lever.
- Don't buy mentions, fragment into thin chunks, or wear the AEO badge.
That discipline is our Search & AI visibility practice.