Getting cited by ChatGPT and AI Overviews: what actually changes

A meaningful share of research now ends inside an AI answer that either cites you or does not. The work that earns a citation overlaps with classic SEO without being the same thing.

Ranking asks: is this page the best result for this query? Extraction asks: is this paragraph a usable, quotable answer on its own, without the rest of the page around it?

That single difference drives everything below.

A model retrieves chunks, not documents. If your paragraph begins "As mentioned above, this means..." it is unusable the moment it leaves the page.

Make every passage self-contained. Name the subject in the sentence rather than relying on a pronoun pointing three paragraphs back. Repeat the entity name more than feels elegant.

Answer in the first sentence, then elaborate. The inverted pyramid, unfashionable in web copy for a decade, is exactly right again.

One claim per paragraph. A paragraph making four points cannot be quoted for any of them.

If people ask "how much does a Google Maps listing service cost", the heading should be close to that string and the paragraph beneath it should answer it in the first line — with a number.

This is not keyword stuffing. It is matching the question form because retrieval matches question form.

Models cite passages carrying specifics: numbers, dates, prices, named tools, measured outcomes, version numbers. "We help businesses grow their online presence" is unquotable. "Foundation fixes typically produce visible Maps movement in 30 to 60 days; competitive urban categories take three to six months" is quotable, and checkable.

The corollary is that padding actively hurts. A 2,000-word page saying what 600 words could say has diluted its own citability.

Publish an llms.txt. A markdown map at your root pointing to your important surfaces, with one-line descriptions. Cheap to write, and it exists precisely for this.

Verify AI crawlers can reach you. Check your robots.txt and any WAF or bot-protection rules against GPTBot, ClaudeBot, PerplexityBot, Google-Extended and Bingbot. Blocking them is a legitimate choice — but make it a choice, not an accident of a security plugin's defaults.

Serve clean text where rendering is heavy. If the content only exists after client-side hydration, assume some retrievers never see it. Server-render anything you want quoted.

Schema does not make an AI cite you. It makes the entity unambiguous — that this Organization, this LocalBusiness, this Service, this price, this rating all belong together. Ambiguous entities get summarised vaguely or attributed to a competitor.

Interlink your schema by @id so the graph is connected rather than a pile of disconnected blocks.

Retrieval leans on corroboration. A claim that appears only on your own domain is weaker than one echoed on a directory, an industry site or a forum thread. This is the least glamorous item on the list and frequently the one that moves citation rate most.

Not impressions. Track whether your brand appears in AI answers for your target questions, on which platforms, and with what framing. Run the questions manually, monthly, and keep the log. The tooling is immature; the discipline of actually checking is not.