Getting Cited by AI Search

A growing share of the people who used to type a question into a search box now ask an assistant instead, and get an answer with a handful of citations rather than ten blue links. If you are not one of those citations, you are not in the conversation at all.

The good news is that most of what makes a page citable is what made it good in the first place. The bad news is that the things people used to get away with — padding, vagueness, keyword repetition — now work against you actively.

How this differs from ordinary search

A search engine ranks pages. An assistant reads them, extracts an answer, and attributes it. That changes what wins.

Specificity beats breadth. Ranking rewarded a page that covered a topic broadly. Citation rewards a page that answers one question precisely, with a number, a date or a concrete statement that can be lifted out.

Structure matters more. An assistant needs to find the answer inside your page. Clear headings that pose the question, short paragraphs that answer it, and lists where a list is genuinely the right shape all make extraction easier.

Being wrong is more visible. If your page says something inaccurate and an assistant repeats it with your name attached, that is worse than not being cited.

Volume stops working entirely. Fifty thin pages gave you fifty chances to rank. They give you zero chances to be cited, because none of them says anything worth quoting.

What actually makes a page citable

Answer the question in the first paragraph. Not after four hundred words of preamble about how important the topic is. State the answer, then explain it. This is also better for human readers, which is a recurring theme.

Use headings that are the questions people ask. "How much does a webshop cost in the Netherlands" is a heading. "Our pricing philosophy" is not.

Include concrete facts. Numbers, dates, thresholds, ranges. "Fast" is not citable. "Under 2.5 seconds" is. "Recently" is not citable. "Since 28 June 2025" is.

Say when you last checked. Assistants weight recency, and readers do too. A dated fact with a stated source is far more likely to be quoted than an undated assertion.

Be willing to be definite. Hedged, committee-written text has nothing extractable in it. If you believe a template is the right answer under 1,500 euros, say that.

The technical side

Less than people expect, but not nothing.

Server-rendered HTML. If your content only exists after JavaScript runs, some crawlers will see an empty page. This is the same argument as ordinary SEO and it has become more important, not less.

Structured data. JSON-LD describing your organisation, your services, your prices, your articles, your FAQs. This is machine-readable fact, and it is the cheapest way to state things unambiguously.

A clean heading hierarchy. One h1, then h2 and h3 in order. Headings are the map an extractor uses.

Let the AI crawlers in. Your robots.txt decides this. Blocking them means guaranteed absence from AI answers, which for most businesses is a bigger cost than whatever the blocking was meant to protect. Check what yours currently says — a lot of default configurations are more restrictive than their owners realise.

An llms.txt file is worth adding. It is a simple text summary of what your site contains and where the important pages are. The convention is not universally supported yet, it costs nothing, and it makes your site easier to summarise correctly.

What does not work

Writing for the machine. Text stuffed with phrases you imagine an assistant wants reads badly to humans and is not more likely to be quoted.

Claiming things you cannot support. Assistants increasingly cross-check, and being contradicted by a more authoritative source is worse than silence.

Walls of text with no structure. If a reader cannot skim it, an extractor cannot either.

Trying to game citation. This is roughly where SEO was in 2005, and the same lesson applies: the shortcuts get closed and the sites that invested in being genuinely useful are the ones still standing afterwards.

The uncomfortable implication

The strategy that works here is to publish the information your competitors keep behind a contact form.

Publish your prices. Publish your process. Answer the questions people ask before they hire someone in your trade, including the ones where the honest answer is "you probably do not need us for this".

That feels counterintuitive and it is the whole point. A page that says "a landing page is 1,495 euros and here is exactly what that includes" can be cited. A page that says "contact us for a bespoke quote tailored to your needs" cannot be cited by anything, ever, because it contains no information.

How to start

Take the five questions your customers ask most often. Write a page for each that answers the question properly — in the first paragraph, with real numbers, with a heading that matches how the question gets asked.

Then check the technical basics: does the page render without JavaScript, does it have structured data, is the heading order sane, are the AI crawlers allowed in.

That is the whole method. It is not a trick and it will not work by next Tuesday. But it is the same work that makes your site useful to the people who arrive on it, which means it pays off even if the assistants never cite you at all.