Two things get called “being cited” and they are not the same. Being mentioned is your brand name appearing in the answer text. Being cited is your page appearing in the source list underneath it.
You can have either without the other, and the work that produces each is different. Mentions come from the engine’s picture of your category. Citations come from a specific page of yours being useful to a specific question.
Start with the pages that already get retrieved
Retrieval borrows heavily from conventional search ranking, so pages that already rank are already being fetched. Many of them are being fetched and discarded, which makes them the cheapest wins available: the hard part is done and the page just is not quotable.
Look at your best-ranking pages for questions a buyer would actually ask, and read the first hundred words of each. If the answer is not in them, that is the whole problem.
The four changes that matter
Put the answer before the argument
The single highest-return edit. A section that opens by stating its conclusion has given the engine something to lift; one that builds to the conclusion over four paragraphs has not, because extraction stops early.
This is uncomfortable for marketing writing, which is often structured to hold attention by withholding the payoff. Engines do not have attention to hold.
Make every claim survive on its own
A quoted passage arrives without its neighbours. “This makes it substantially faster” is unusable — this and it are elsewhere. Name the subject again even where it reads as repetitive. The cost is a little inelegance; the benefit is that the sentence can be lifted whole.
Same rule for numbers. “Improved by 12 points” drops out. “Across 40 prompts on five engines, visibility rose from 6% to 18%” carries its own context and survives.
Turn headings into questions
Retrieval matches a user’s question against your structure. How much does an AEO audit cost? matches a question of that shape; Pricing does not. Then mirror those headings in FAQ schema, which hands the engine the pairs pre-extracted rather than asking it to infer them.
Fix how others describe you
The uncomfortable one, because it is not on your domain. Engines lean on sources that other sources trust: review platforms, trade publications, documentation, forums. If those describe you inaccurately or not at all, your own pages are arguing against a chorus.
A single corrected listing on a source that engines actually consult can move more answers than a quarter of blog posts. The work is finding out which sources those are for your category rather than assuming, because they differ by category and by engine.
What does not work
Publishing more of the same. If ten pages are structured so nothing can be extracted from them, the eleventh will not change the picture.
Keyword density. There is no term-frequency lever here. The model is selecting passages that answer a question, not matching strings.
Writing for the engine instead of the reader. Pages stuffed with question-shaped headings and no substance underneath get retrieved and discarded like any other thin page. The reason clear writing wins is that it is genuinely more useful, not that it games anything.
How to know whether it worked
Fix a prompt set before you start and leave it fixed. Then track two things separately: whether your brand is named, and whether your domain appears in the source list. Improving the first without the second means the engine knows you but reads about you somewhere else — which is worth knowing, and points the work somewhere different.
Give it weeks, not days. Retrieval reflects the live web quickly for pages that get re-crawled often and slowly for everything else, and a single day’s answers vary enough to mislead in either direction.