Ranking and being cited are not the same thing
Classic SEO competes for a position in a list. An AI answer does not show a list. It gives a paragraph and, if you are fortunate, names its sources.
That changes what gets rewarded. A page that ranks by being comprehensive can still lose to a page that states one thing clearly, because the model needs a quotable, self-contained claim rather than a thorough document.
It also changes who competes. A small operator with genuinely specific information can be cited alongside a large site, because the model is looking for the sentence that answers the question, not the domain with the most links.
What actually helps, in order
Let the crawlers in. GPTBot, OAI-SearchBot, PerplexityBot, ClaudeBot and Google-Extended each need to be allowed in robots.txt. Many sites block them by default through a security plugin and never find out.
Answer in the first two sentences. Put the direct answer at the top of the section, then support it. Models extract from the top of a passage far more often than the middle.
Make claims specific and checkable. "We serve the Atlanta area" is unquotable. "Eight airport transfer operators across Georgia and Michigan" is a fact a model can carry into an answer.
Structure it. Real headings, real lists, FAQ blocks with the question as the heading. Structure is how a passage gets isolated from the page around it.
Add schema. Organization, Service, FAQPage and Breadcrumb markup give machines an unambiguous reading of what the page is.
The part most guides skip
You cannot optimise for a query you have never checked. Before changing anything, ask the actual questions a buyer would ask, in the actual assistants, and write down who gets cited.
You will usually find one of three things: a competitor is cited, a directory is cited, or nothing relevant is cited at all. The third is the opportunity, and it is more common in niche B2B than people expect.
Re-check monthly. These answers change far faster than search rankings do, and the only way to know whether a change worked is to have recorded what came before it.
An honest caveat
Nobody can guarantee a citation. The mechanisms are not public, they change often, and anyone selling certainty here is selling something else.
What is reliable is the input side: being crawlable, being specific, being structured, and being genuinely the best answer to a narrow question. That is a description of good publishing, which is why it survives whatever the retrieval layer does next.