How to Get Your Brand Cited by AI Search

A meaningful share of the questions your customers used to type into a search box are now typed into a chat window. The answer arrives as a paragraph with two or three sources under it. Either you are one of them or you do not exist in that conversation.

This is a different game to ranking, and most of the advice circulating about it is warmed-over SEO. Here is what is actually different.

Answer engines quote passages, not pages

A search engine ranks documents. An answer engine retrieves chunks — a few hundred words at a time — and assembles a response from several of them. That single mechanical fact drives everything else.

It means the unit you are optimising is the paragraph. A brilliant page whose key claim only makes sense after reading the previous four sections will never be quoted, because the retrieved chunk arrives without that context. Every section has to survive being read alone.

Practically: state the conclusion in the section, not just in the introduction. Repeat the subject instead of leaning on "it" and "this". A paragraph that begins "This is why it matters" is unquotable. A paragraph that begins "Brand strategy matters because it eliminates options" can be lifted whole.

Answer the question in the first hundred words

Models favour sources that resolve the query quickly and unambiguously. The journalistic build-up — context, then nuance, then finally the answer — is exactly wrong here.

Put a direct, complete answer near the top, phrased as a standalone sentence. Then earn the rest of the page. This is also better for humans, who have been skimming for the answer since long before the machines showed up.

Be specific enough to be checkable

Vague copy is not retrieved because it is not useful to quote. "We deliver outstanding results for our clients" contains no information. "Most brand strategy projects take three to eight weeks" does, and a model can attribute it to you.

Numbers, ranges, dates, named trade-offs and explicit conditions are what get pulled. So is admitting the limits of a claim — "this works for B2B services and does not apply to marketplaces" reads as reliable rather than promotional, and models are visibly tuned toward sources that hedge accurately.

The structure that gets retrieved

• One question per page. A page trying to cover five topics gets beaten by five pages that each cover one.
• Descriptive headings. "How much does it cost" retrieves; "The investment" does not.
• Question-shaped headings. They match the way people actually ask, which is how the query is embedded.
• A summary box near the top containing the liftable answer.
• A genuine FAQ, marked up with FAQPage schema, with answers that are complete sentences rather than teasers.
• Visible dates. Published and updated. Freshness is a real retrieval signal, and an undated page is a risk a model would rather not take.
• Named authorship and a real organisation behind it.

Let the crawlers in

The retrieval-time crawlers are separate from the training crawlers, and separate from Googlebot. In your robots.txt, the ones that matter today are GPTBot and OAI-SearchBot (OpenAI), ClaudeBot (Anthropic), PerplexityBot, Google-Extended and CCBot (Common Crawl, which feeds many others).

Blocking them is a legitimate choice if your content is your product. But be clear about the trade: you are not protecting yourself from being trained on — that happened to whatever was already public — you are removing yourself from citations going forward.

Two more mechanical points. Serve your content in the HTML, not only after JavaScript runs; several retrieval crawlers do not execute scripts, and a page that renders client-side is an empty page to them. And publish an llms.txt at your root — a plain-text map of your best pages, which a growing number of agents read first.

What still comes from classic SEO

Do not throw out the old work. Answer engines lean heavily on conventional signals to decide who is credible: being linked to by sites in your field, having a coherent set of pages on one subject rather than one page on twenty subjects, loading quickly, and not lying. The fundamentals still apply — they are simply no longer sufficient.

The largest remaining advantage is topical depth. A single page about brand strategy competes with the internet. Twelve connected pages covering strategy, identity, pricing, briefs and rebrands make a model treat the whole domain as a source on the subject.

How to measure it

There is no rank tracker for this that I would trust yet. So do it by hand, and do it consistently.

Write down twenty questions a real prospect would ask before hiring you. Once a month, run all twenty through ChatGPT, Perplexity, Claude and a Google search that triggers an AI Overview. Record three things: whether you were cited, what was said about you, and who was cited instead.

That third column is the useful one. It tells you exactly which pages you are being measured against — and reading them will tell you, in about an hour, what you would have to publish to replace one.