How a small brand shows up in AI search

When someone asks an assistant for a good agency in Karachi, or a personalised newborn gift, or who can catalogue an auction from photographs, the assistant names somebody. The names it gives are not the biggest brands. They are the ones it is most certain about. A small brand can be that, and most are not, because certainty is built from dull work nobody did.

Why does an assistant name one brand and not another?

An engine builds a picture of your business from every place it can read: your site, your Google listing, your Instagram profile, a directory you forgot, a supplier's page that mentions you. From those it decides two things: what you are, and how sure it is. The second decides whether you get named. A brand described five different ways in five places is five weak guesses. A brand described one way everywhere is one confident fact, and confident facts are what an assistant repeats.

Size helps only because big brands are mentioned more. A small brand cannot buy mentions, but it controls every one of its own, and that is enough for most of the questions it wants to be the answer to. The question is rarely who the best agency in the world is. It is who does this, here, for a business like mine, and that question has a short list.

What is the entity, and why fix it first?

The entity is the machine's idea of you: one name, one place, one line about what you do. Most small brands have never written that line down, so it has been written for them, differently, by whoever filled in each profile. The company is a studio on the site, an agency on Google, a consultancy on a directory, and the address on one of them is the old one.

Write the line once. Who you are, where you are, what you do, for whom. Ours is one sentence: Drifted Marketing is a growth agency in Karachi, founded 2026, doing ads, content and AI apps for brands in the United States, the Netherlands, Australia and Pakistan. Then use that sentence, unaltered, everywhere a machine might read it. Spell the name the same way, with the same punctuation. Use one office address, written one way.

• The footer and the about page of your own site.
• The summary file at llms.txt.
• The structured data inside the page code, which a machine reads before it reads the copy.
• Your Google Business Profile, with the category matching what the site says you are.
• Every social profile, including the ones you stopped posting to.
• Every directory that lists you, including the ones you did not ask to be on.

Check it once a quarter. When something changes (an address, a name, a founder), change it everywhere in the same week, because a machine that finds the old and the new side by side does not pick the newer one. It trusts both less.

What does an answer-first page look like?

An engine does not read a page. It lifts a passage, and the passage it lifts is the one that answers the question on its own, without the paragraphs around it. Most pages build to the point over six paragraphs because that is how people were taught to write, and the engine, which is not a person, has stopped reading by paragraph six.

So put the answer under the heading, complete, and then the detail. Make the heading the question the reader typed. If the question is what you charge, the first sentence under it is the price. If the question is whether you work with a category, the first sentence is yes or no, with the client named by company. Then explain.

Every piece in this journal opens with a boxed short version, labelled the same way every time, so the extractable answer can be found by pattern rather than by reading. The headings are questions. The fees are read from one file into every page that quotes them, so no page can quote a price the pricing page has changed.

What is the summary file at llms.txt?

A plain text file on your own domain, at the address llms.txt, 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.

Write it the way you would brief a new colleague who will be asked about you by strangers: the one line first, then the services, then the proof you can name, then the price, then the links. Keep the top short, because the top is what gets quoted. Say nothing in it that a page on the site does not also say. A summary file that drifts from the site is a contradiction, and contradictions lower the engine's confidence in both.

Ours is generated from the same content files that build the pages, so it cannot disagree with them. A small brand without a build pipeline can do the same by hand: write the file from the site, not from memory, and update it the same day the site changes.

Why do profiles matter when nobody reads them?

Because machines do. The Instagram bio you wrote in a hurry, the Google listing a former intern set up, the directory page with the old phone number: each is a source, and each one that disagrees with the site is a vote against you. The fix is not to post more. It is to make every profile carry the line, the same name and the same address, and link to the site.

Google Business Profile is the one that matters most for a local question, because it is the source an engine trusts for where you are and whether you exist. Keep the hours true, answer the reviews, post to it now and then, and make the category match what the site says you are. A listing that says one thing while the site says another is the most common reason a real business is invisible to an assistant that would otherwise have named it.

Why do “best agencies” lists earn citations for your competitors?

The instinct is to write the roundup: the ten best agencies in your city, with yourself placed third for modesty. For a while engines quoted those pages, because they were shaped like answers. They have since learned what a reader learned earlier. A roundup written by one of the entries is an advertisement, and it knows nothing first-hand about the other nine.

Worse, it works against you in a specific way. A list page names your competitors beside you, in your own words, on your own domain. An engine reading it learns that those companies exist, what they do, and that you consider them peers. It now has nine more entities to be confident about, and the page it cites for the question is still not yours, because a list is not a primary source for anything.

Engines now prefer the page with first-hand knowledge. 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 its date. Write those pages. Concede something on them, because a page that never admits a case where it is the wrong choice does not read like an answer to a comparison question.

What does this site do, as the worked example?

The last row is the one that matters. We built the gate because we made the mistakes it catches, more than once, before it existed: a figure that reached a page without being cleared, a client's result attached to the wrong case, a number that was true once and quietly stopped being. An engine that finds one of those does not forgive the rest of the page.

What none of this can do is make a weak offer quotable. If there is no fact, there is nothing to cite. It does not replace ordinary search, which still sends most of the visitors. And nobody can promise you a citation, because the engine decides and changes its mind. What a small brand can do is give it no reason to doubt you, which is most of the work and almost none of the cost. The longer piece on the mechanics is at /blog/how-to-get-your-brand-cited-by-ai-search, and the service page is at /services/ai-search.