AI SEO: how to get cited inside AI search results

Ranking and being quoted are two different jobs

Google's AI Overviews, ChatGPT search and Perplexity all do the same thing: read a handful of pages, then write one answer and credit a few sources. Your page is no longer competing only for a position. It is competing to be the passage the model finds easiest to lift.

That changes what to optimise. A page can be technically perfect and still be skipped because its answer is spread across six paragraphs and three subheadings. A worse page that answers the question in one clean block gets the citation instead.

Write answers that can be lifted whole

When we audited our own site in August 2026 we measured the FAQ answers on our course pages. They averaged 28.6 words. The FAQ hub averaged 55.3. Not one answer on the entire site sat in the 130-170 word band that tends to get extracted.

That is a fixable problem and it is the highest-leverage one on this list. Take your most common customer question, answer it in a self-contained block of roughly 130 to 160 words, and make sure that block does not depend on the paragraph above it for context. A model reading the page in isolation should be able to quote it and be correct.

• Answer first — the claim in the first sentence, the reasoning after it. Never the other way round.
• Self-contained — no "as mentioned above", no pronouns pointing at earlier paragraphs.
• Named specifics — numbers, tool names, places. "Fees start from ₹4,999 per month" is quotable; "affordable fees" is not.
• One question per block — two questions in one answer means neither gets cited cleanly.

Publish the number you keep hiding

Price is the most-asked question in almost every category, and the most commonly withheld. "Contact us for pricing" cannot be cited, because there is nothing to cite. Any competitor who publishes a figure or a range gets quoted on the question you refused to answer.

We had this exact fault. Our FAQ said fees were quoted on a call rather than published, and the highest-intent question in our category had no extractable answer anywhere on the site. Fixing it was a copy change, not a technical one.

Check that the assistants can actually read you

Two things block AI crawlers, and both are quiet. The first is robots.txt: GPTBot, ClaudeBot, PerplexityBot, OAI-SearchBot and Google-Extended are separate user agents, and a wildcard rule that blocks aggressive scrapers can take them out too. Google-Extended is the one people miss — it governs AI Overview grounding without touching your Search ranking, so blocking it costs you AI visibility and nothing tells you.

The second is client-side rendering. If your content only appears after JavaScript runs, some crawlers will see an empty shell. Test it the blunt way: fetch your page with curl and read what comes back. If your answer is not in that HTML, an assistant may never see it.

Structured data helps entity resolution, not rankings

Be careful with the claims here, because a lot of advice overstates it. Schema markup does not make an assistant cite you, and Google retired FAQ rich results for every site in May 2026, so FAQPage markup no longer earns a SERP feature.

What schema still does is remove ambiguity about who you are. One Organization node with a stable @id, referenced by your other markup rather than re-declared, tells a machine that the course, the reviews and the business are one entity. We found three unlinked nodes for one organisation on our own site — a common fault, and worth fixing for correctness even though nobody should promise you a ranking for it.

Say what you actually know, including the limits

Assistants are drawing on pages that read like they were written by someone who has done the thing. Concrete numbers, named tools, stated trade-offs, and an honest limitation do more than another round of adjectives.

The honest limitation on this topic: nobody can currently prove which of these changes moves citation rates, because the platforms publish no citation analytics. What is verifiable today is whether your answers are extractable, whether crawlers can reach them, and whether your facts are consistent. Those are worth doing on their own merits. Anything beyond that is inference, and anyone selling it as certainty is guessing.