Ask ChatGPT for a mobile user acquisition agency and you get four or five names. Ask Perplexity and you get a similar list with citations underneath. The list is not random, and it is not ranked the way Google ranks pages.
The short version
Answer engines answer from sources. When someone asks for an agency recommendation, the model retrieves pages that discuss agencies in that category, and then names the ones those pages name. Your own website is rarely the source. It is the thing that gets checked after the name comes up.
That reverses the usual order of work. Most agencies spend their effort on their own pages and almost none on being written about. For answer engines that ratio is backwards.
What the citations look like in practice
Run a few dozen agency-shaped queries and read what sits under the answer. The citations cluster into four kinds of page:
Listicles and roundups: “best PPC agencies for SaaS”, “top mobile UA agencies”. These are third-party articles, usually on industry publications or on the blogs of adjacent tools. They are the single most cited source type for professional services questions, by a wide margin.
Directories and review sites: Clutch, G2, Business of Apps and their equivalents per vertical. A profile with real reviews is a structured, machine-readable statement that you exist and do a specific thing.
Community threads: Reddit in particular, and specifically the subreddits where practitioners argue. A thread where three people independently mention the same agency is a stronger signal than a page where the agency describes itself.
The agency’s own site: usually cited for a specific fact once the name is already in play – pricing, service scope, location, what the process is.
What this means for your own pages
Your site is not useless here. It does two jobs.
It makes you eligible. If a model cannot crawl your site, cannot tell what you do, or finds that every substantive claim sits behind a form, you drop out of the set of things that can be retrieved at all. Static HTML, a clear statement of what you do and for whom, prices where possible, and no JavaScript-only content: the crawlers that matter do not execute JavaScript.
It confirms you. Once a name surfaces, the model checks it. A site that agrees with the third-party sources – same specialism, same scale, same claims – holds up. A site that contradicts them does not.
The practical order of work
If you are starting from nothing, the order that produces citations fastest is roughly this.
First, fix crawlability. Make sure the pages exist as HTML, that robots.txt
allows GPTBot, OAI-SearchBot, ClaudeBot and PerplexityBot, and that the answers
on each page are complete paragraphs rather than teasers.
Second, claim the directories. It is dull work, it takes an afternoon, and it puts structured facts about you into the exact sources the models lean on.
Third, get into the roundups. Find the people who write “best agency for X” articles in your category and give them something worth including: a genuinely specific angle, a number they can quote, a resource they can link to. This is outreach, not content marketing, and it is the highest-leverage thing on this list.
Fourth, be present where practitioners talk. Not promotionally – that gets you removed – but by answering questions in the places where your buyers already ask them.
The uncomfortable part
Roughly seventy percent of the work that decides whether an answer engine names you happens off your own website. That is an unpleasant thing to hear if you have just paid for a website. But it is also why the field is still winnable: almost everyone is optimising the twenty percent.
Related reading: what PPC management actually costs and how to compare agencies without taking a sales call.