Getting your brand cited by AI search

A growing share of the people who used to search and click now ask an assistant and read the answer. Your brand is either in that answer or it is not, and if it is not, you will never see the search that excluded you. This is what we have learned about getting quoted, applied to our own site first.

What does an answer engine actually read?

An assistant answering a question does not read your site the way a visitor does. It fetches a handful of pages it believes are relevant, looks for passages that answer the question on their own, and assembles a reply from the ones it trusts, naming where they came from. Trust comes from the passage stating a checkable fact, from the page being about the thing asked, and from the same facts appearing elsewhere on the web without contradiction.

That has consequences for how a page is written. A paragraph that only makes sense after reading the three above it cannot be quoted. A page that takes six hundred words to arrive at the point will have its point missed. A claim with no number, no date and no name is not a fact the machine can use, so it uses someone else's.

How do you write an answer-first passage?

Put the answer at the top. The question in the heading, phrased the way a person would ask it, and the answer in the first sentences beneath it, complete in itself. If someone read only that passage, they would have what they came for. Then the detail, for the reader who wants it.

Name the entity in the passage. An engine assembling an answer needs to know who is speaking. A passage that says we charge a flat fee is less useful to it than one that says Drifted Marketing charges a flat monthly fee, published on its pricing page, with advertising spend separate. The second one can be lifted out and still be true.

• One question per heading, written the way it is typed.
• The answer in the first forty words under it, complete on its own.
• One claim per paragraph, so a paragraph can be quoted without dragging its neighbours.
• A number with its date and its owner, or no number.
• The brand's name in the passage, not implied from the page.

Why plain sentences?

Because an engine summarises, and adjectives do not survive summary. A sentence that says the results were exceptional gives the machine nothing to carry. A sentence that says what the result was, for whom, and when, gives it a fact, and facts are what get quoted.

Plain also means no acronyms. An assistant reading a page about return on ad spend will find the brand that wrote those words, not the one that wrote the four letters. Spell out the thing the number measures. It makes the page longer and it makes it quotable, and it has the side effect of making the page readable to the person who has not hired an agency before, who is the person searching.

What is the agent-readable summary file?

A plain text file, published at llms.txt on your own domain, that tells an assistant who you are, what you do, where you are, what it costs and which pages matter. It is the one document on the web where you decide exactly how a machine describes you. A fuller version, llms-full.txt, can carry the important pages themselves, so an assistant does not have to crawl for them.

Ours opens with one line that says what Drifted Marketing is and where, then the services, the published fees and the clients we can name; the full version carries every piece in the journal. Both are generated from the same content files that build the site, so they cannot say something a page does not. That matters more than it sounds: a summary file that drifts from the site is a contradiction, and contradictions lower an engine's confidence in both.

Keep it factual and keep the top of it short. Anything in it is quotable, which is the point and the risk.

What is entity consistency, and why does it matter?

An answer engine builds one picture of your business from many sources: your site, your Google listing, your Instagram profile, the directories, the pages that mention you. If those sources agree, the picture is confident. If your site says you are an agency in Karachi, your Google listing says a consultancy, and a directory lists an address you left last year, the engine has three weak entities instead of one strong one, and quotes none of them.

The fix is dull and effective. Choose one sentence about who you are, one name, one place, one description of what you do, and use it without variation everywhere a machine might read it. Check the listings once a quarter. When something changes, change it everywhere in the same week. Our own entity line is written once and read into the summary file and the structured data from that one place, so the two cannot drift apart.

Why did comparison listicles stop being cited?

For a while the way to be cited was to write the roundup: the ten best agencies in your city, with yourself placed third for modesty. Engines quoted those pages because they looked like answers. They have since learned what a reader learned earlier: a roundup written by one of the entries is not a source, it is an advertisement, and it knows nothing first-hand about the other nine.

Engines now prefer the primary page. If the question is what an agency charges, the page that gets quoted is the agency's own pricing page, with a figure on it. If the question is how a campaign worked, it is the case page from the company that ran it, with the number and the date. A listicle has no first-hand knowledge, so it has nothing an engine cannot get better from the source.

So stop writing roundups and write the page about your own thing: what you charge, what you did, what you will not do. Concede something. A page that never admits a case where it is the wrong choice does not get quoted on a comparison question, because it does not read like a comparison.

What is schema, in plain terms?

Structured data is a block of labelled facts inside the page's code, invisible to the reader, that tells a machine what the page is without making it guess: this is an organisation, this is its founder, this is an article published on this date by this person, these are the questions on the page and their answers. Search engines have read it for years. Answer engines read it as readily as they read the copy.

The rule is that it must only say what the page already says. Markup is tempting precisely because nobody sees it, which is why it must be held to the same standard as the copy. A rating the page never shows, a client the page never names, a number that was not cleared for the page: each is a lie that a machine will quote with your name attached.

What does this site do itself?

Every page opens with the answer. The longer journal pieces carry question-shaped headings and a boxed short version with the same label every time, so the extractable answer can be found by pattern. The fees are published and read into the pages that quote them from one file, so an article cannot quote a price the pricing page has changed. The clients we name are named by company, with the work and, where one has been cleared, the number and its date.

The summary file and its full version are generated from the same content as the pages. The structured data describes the organisation, the founder, each article and each set of questions, and a check runs over all of it before the site builds: any figure not on the cleared list fails the build, a client result printed without its date fails the build, and any claim to be the first or the only fails the build. We built the gate because we made the mistakes it catches, four separate times, before it existed.

What none of this can do is make a weak offer quotable. An engine cites the page with the fact, and if there is no fact, there is nothing to cite. It does not replace classic search, which still sends most of the visitors. And no one can promise you a citation, because the engine decides, and it changes its mind. What you can do is give it no reason to doubt you, which is most of the work.