Getting cited by ChatGPT: what AI search optimisation actually involves

A growing share of buying research now starts with a question typed into an assistant, and the answer arrives as a paragraph with two or three sources attached. That changes the goal. Ranking first is a position on a list; being cited is being the thing the answer is made of. The work overlaps with SEO but is not the same work.

How should a page be structured to get cited by AI?

One thing that used to be part of this answer no longer is. FAQ rich results stopped appearing in Google Search on 7 May 2026, so an FAQ block no longer buys a visible search feature and the markup no longer earns its keep on that basis alone. The question-and-answer shape still works, because a self-contained answer under a real question is exactly what an engine can lift, but the reason changed: see what to do with FAQPage markup now.

In self-contained passages, each one written to be lifted out on its own. A model quoting your page does not quote the page; it takes a passage from it. If understanding that passage requires the three paragraphs above it, it is a poor candidate however good the article is. The shape that gets cited is consistent: the question restated as a heading, the answer in the first sentence, and the qualification after it rather than before. That ordering feels wrong to write, because good prose usually builds towards its point. Here the point goes first and the reasoning follows it, because the reasoning may never be read by the thing doing the quoting.

This is close to the old featured snippet advice, with one difference that matters: an assistant is synthesising rather than excerpting, so hedging language and long preambles get dropped or, worse, summarised into something you did not mean.

By matching what your site says about you against what everywhere else says, and trusting the parts that agree. Before a model can recommend you it has to be confident about what you are. Schema does most of that work: Organization markup stating the name, the location, the services, and the links to the profiles that corroborate them. Where businesses lose is inconsistency rather than absence. A company described three different ways across its own site, its profiles and its directory listings is three weak signals instead of one strong one, and a model resolving that ambiguity will usually reach for a competitor it can describe without hedging.

Off-site sources do a lot of the deciding

Assistants lean heavily on sources they treat as consensus: established directories, review platforms, community discussion, trade press. A brand mentioned across those is far more likely to surface than one whose only advocate is its own homepage. This is the part that cannot be done purely on your own website, and it is usually the part that gets skipped.

• Let the crawlers in. Check that the AI user agents are not blocked in robots.txt by an old default nobody revisited.
• Serve content in the HTML. Anything that only exists after a client-side fetch may never be seen.
• Date and attribute your pages. Freshness and a named author both feed how much a model trusts a source.
• Track it. Record the prompts you care about and re-run them, or you are guessing about whether any of this worked.

Ranking first puts you on the list. Being cited means the answer is partly made of you.

The reassuring part is that almost none of this conflicts with ordinary good SEO. Clear structure, honest specifics, a site that renders its own content and a brand that exists somewhere other than its own domain. The difference is mostly that AI search punishes vagueness harder, because a model cannot quote a sentence that does not commit to anything.