Get cited by AI search

The old game was ranking: climb to the top of a list and earn the click. The new game is being quoted: the model reads a handful of pages, lifts the clearest answer, and names a source. You do not need to be first anymore. You need to be the most quotable, and you need to be talked about in the places the models trust. Here is how.

1. Write for the question, not the keyword

Keyword SEO trained us to target phrases like "B2B lead generation software". Nobody asks an AI that. They ask, "what is the best way to get leads for a small B2B agency without paying for ads?" Models match the MEANING of a real question, so your page has to contain the real question and a direct answer to it.

Start by writing down the actual sentences your buyers say. Pull them from sales calls, your inbox, the "people also ask" box, and your own DMs. One page should own one question.

Weak: a page titled "Lead Generation Strategies"

Strong: a page titled "How do I get B2B leads without running ads?" that answers exactly that, and nothing else

2. Answer in the first two sentences

An AI engine quotes the passage that resolves the question fastest. If the answer is buried under a 300-word warm-up, the model skips you for a page that just says it. So put the answer FIRST, then explain. Journalists call it the inverted pyramid, and it is the single biggest lever on your own page.

The opening passage should make sense lifted out of the page on its own, with no setup. Write it so that if a model pasted those two sentences into an answer and credited you, it would be correct and complete.

Lines that get quoted: "The fastest way to get B2B leads without ads is to mine your existing network: export your contacts, filter them to your ideal customer, and message the warm ones first." Direct, specific, self-contained.

3. Make every heading a question

Models parse a page by its headings. When a heading is the exact question someone asked, you hand the engine a clean match between the query and your answer. So phrase your H2s as the questions, not as topics.

Turn "Pricing" into "How much does it cost?" Turn "Integrations" into "Does it work with HubSpot?" Each section then answers its own heading in the first line. This also makes you eligible for the "people also ask" expansions, which feed the same answer boxes.

4. Be specific enough to be worth quoting

This is the strongest lever there is. A model will not cite a sentence that could have come from anywhere. Specifics are what make a passage quotable: a number, a name, a date, a step, a real result. "Improves engagement" is wallpaper. "Carousels reached 2.6x more people than single images in my last 40 posts" is a citation waiting to happen.

There is hard evidence for this. In the Princeton and Georgia Tech study that coined the term "generative engine optimisation" (GEO, KDD 2024), pages that ADDED statistics, quotations, and cited sources were picked up by AI answers up to 40% more often. It is the most reliable thing you can do to your own page.

The same research found the opposite move backfires. Copy that reads like an advert, packed with persuasive adjectives and self-praise, got cited around a quarter LESS often. So put the concrete thing in the sentence, not three paragraphs later, and cut the marketing gloss. Specificity is also how a model decides you actually know the thing, which is the trust behind a citation.

5. Cite your sources and quote real people

The GEO study did not just reward numbers. Adding a real quotation from a named person, and citing a real source for your claims, both lifted citations on their own. A model treats a page that shows its own receipts as more trustworthy, and trustworthy is what gets quoted.

So name your sources in the text. Quote the customer, the expert, the study. "According to a 2025 Ahrefs analysis…" or "as our head of sales puts it, '…'" does two jobs at once: it makes the passage more credible, and it hands the model a ready-made, attributable line. This guide does it too, which is the whole point.

6. Get mentioned where the models look

Everything above is about your own page. This tactic is about everywhere else, and it may matter more. An Ahrefs study of over 75,000 brands (December 2025) found that how often a brand is MENTIONED across the web predicts its AI visibility better than backlinks do. Mentions on YouTube were the single strongest signal, and Reddit, Wikipedia, YouTube, and review sites were among the domains AI answers cite most.

The takeaway is uncomfortable but freeing: being talked about in the right places often beats polishing your own page one more time. So get into the sources the models already read. Answer questions on Reddit in your niche. Get onto YouTube, or get mentioned in someone else's video. Earn a line in a roundup or a review site. Make sure your brand has a clean, factual presence where these engines go looking.

The engines differ a little in taste. Perplexity leans on fresh, well-sourced pages, ChatGPT cites few sources and skews Wikipedia-heavy, Google's AI answers weight established authority and YouTube, and Claude favours sources it already knows well. You do not need to game each one. Being genuinely useful, well-sourced, and widely mentioned helps with all of them.

7. Give the page a shape a model can read

Beyond the words, the shape of the page decides whether an engine can extract you cleanly. None of this is exotic, and all of it compounds.

1. Short, self-contained passages. Two to four sentences each, each making one point that survives on its own.
2. A one-paragraph summary near the top. Models love a tight abstract they can quote whole.
3. A short FAQ at the end. Three to five real questions with direct answers, which doubles your question coverage.
4. Go deep on one topic, do not just add pages. How many pages you publish barely moves the needle. One page that genuinely answers a question beats ten thin ones. Depth is what gets quoted.
5. Keep it fresh for real. Models favour current information, but changing the date at the top does nothing. Update the actual substance: new numbers, a new example, a changed recommendation.
6. Plain HTML and real text. Headings, lists, and paragraphs the crawler can read, not text baked into images.
7. Let the AI crawlers in. Do not block GPTBot, ClaudeBot, or PerplexityBot in robots.txt, or you cannot be cited at all.

One thing to skip: you may hear you need an llms.txt file. Almost no AI crawler actually fetches it today, so it is not where your effort should go. Spend that time on the seven points above instead.

8. Draft it with AI, then check it

Hand the question and your raw facts to a model and ask it to write the page to this shape: answer first, question headings, specifics and real sources kept, sales tone cut, an FAQ at the end. Then run the real test. Copy any single passage out of the page, read it cold, and ask: does this answer the question completely, on its own, with a specific in it? If yes, it is quotable. If you needed the rest of the page to make sense of it, rewrite it.

The exact prompt I use to turn a question and a pile of facts into a citable page is in the unlock below, alongside the Get Cited skill, which runs that whole audit and rewrite for you inside the Claude app.

Stop writing for the crawler and start writing to be quoted. Own one real question per page, answer it in the first two sentences, make the answer specific and sourced, and get your name mentioned in the places the models already trust.