Last updated: 7 August 2026
Every guide to AI citations assumes you are trying to get your content cited. This one is about something harder: getting your agency named when a buyer asks ChatGPT “who is the best lead generation agency for gyms?” or asks Perplexity to shortlist marketing agencies in their niche. Those agency-category prompts are answered differently to informational prompts, and most agencies are optimising the wrong surface entirely. We run an agency, we track our own citations daily, and this is the playbook we actually use.
The short answer
How does a marketing agency get cited by ChatGPT? Agency-category prompts (“best X agency for Y”) are overwhelmingly answered from listicles, comparison roundups, directories and community threads — not from agency homepages — so the fastest path is getting your agency named on third-party surfaces the engines already trust. On your own site, build one dedicated page per buyer prompt you want to win, with a direct extractable answer at the top, rather than one generic services page. Back it with verifiable proof (reviews, case studies, named numbers), because engines lift claims they can attribute. Then track your share of answer across ChatGPT, Gemini and Perplexity on a fixed prompt list, since each engine cites almost entirely different sources. Expect the first citations on third-party and long-tail prompts within weeks, and the big head-term prompts to take months of compounding presence.
How “best agency” prompts actually get answered
Run the prompt yourself: ask ChatGPT with browsing for the best marketing agency in any niche. Look at the citations. You will almost never see an agency’s own homepage cited as the authority for why that agency is good. You will see roundup articles (“Top 10 SaaS marketing agencies in 2026”), comparison pages, directory profiles, and occasionally a Reddit thread where someone asked the same question.
This is the structural fact the whole agency play hangs on: for category-level commercial prompts, the engines want a source that compares options, not a source that is one of the options. Your homepage saying “we are the best” is a claim. A listicle that ranks you third out of ten is evidence. The engine’s job is to synthesise a recommendation it can defend, and a comparison page lets it do that in one retrieval.
That doesn’t mean your own site is useless — it wins a different class of prompt, which we’ll get to. But if your entire AI-visibility effort is on-domain, you are invisible for the exact prompts where buyers pick agencies. We covered the general mechanics of what makes a retrieved page actually get cited in our breakdown of the 15% of pages ChatGPT actually cites; this post is about which pages to point that machinery at when the product is your agency itself.
The five surfaces that win agency-category citations
Here is how the surfaces stack up for “best [niche] agency” style prompts, based on what we see cited across our own daily polls and the category prompts we monitor:
| Surface | Citation likelihood (category prompts) | Your control | Effort to land | Notes |
|---|---|---|---|---|
| Third-party listicles / roundups | Very high | Low–medium | Medium | The dominant cited format for “best agency” prompts. Pitch inclusion, or publish the comparison yourself and be honest in it. |
| Directories (Clutch, DesignRush, local equivalents) | High | Medium | Low | Structured, review-backed, constantly refreshed — everything engines like. A complete profile with real reviews is cheap insurance. |
| Reddit and community threads | High (engine-dependent) | Low | Medium–high | Heavily weighted by Perplexity in particular. Cannot be faked at scale; earned mentions only. |
| Press and industry publications | Medium–high | Low | High | Slow to land, but citations from editorial domains persist and compound. |
| Your own site | Low for category prompts; high for specific buyer prompts | Total | Medium | Wins long-tail “how much does X cost” / “X agency for Y niche” prompts, not head-term roundup prompts. |
Notice the inversion: the surfaces with the highest citation likelihood are the ones you control least. That is not a bug. Engines lean on third-party surfaces for exactly the reason buyers do — they are harder to game. The agencies winning AI recommendations right now are the ones treating listicle inclusion, directory hygiene and community reputation as a channel with an owner and a monthly cadence, not as PR leftovers.
Why you must exist on surfaces you don’t own
There’s a second reason the third-party layer matters: the engines barely agree with each other. In our 5,051 first-party citation polls, 74% of the prompts our domain was ever cited on were cited by only one of the three engines we track — just five prompts were ever cited by all three. If ChatGPT, Gemini and Perplexity each assemble answers from largely different source pools, one great page on one surface cannot cover you. You need your agency’s name present in several independent pools so that whichever pool an engine draws from, you’re in it.
The per-engine differences are stark enough that we wrote a separate deep dive on how ChatGPT and Perplexity cite almost completely different sources — short version, ChatGPT leans editorial and structured, Perplexity leans community. For an agency that means: the listicle and press work is largely a ChatGPT/Gemini play, while Reddit presence is disproportionately a Perplexity play. If your buyers skew technical or sceptical, the Reddit side matters more than it looks; we’ve written up how B2B companies earn Reddit citations without getting nuked by moderators separately.
Practical sequencing for an agency starting from zero third-party presence:
- Directories first. One afternoon of work. Complete profiles, real client reviews, niche and location fields filled in — the structured data engines extract from these is exactly the “agency X serves niche Y in region Z” fact they need to include you in an answer.
- Listicles second. Find every “best [your niche] agency” roundup that currently gets cited, and pitch the publishers with something that makes their list better: a genuine differentiator, a verifiable stat, a named client result. Where no good roundup exists, publish one on your own domain and include competitors honestly — a comparison page that only flatters its author doesn’t get cited.
- Community third. Slowest and least controllable, but the citations are the stickiest. Answer real questions in your niche’s subreddits and communities as a named practitioner, months before you need the mentions.
- Press last. Highest effort per placement. Prioritise industry publications your buyers already name-drop, not generic PR volume.
On your own site: one page per buyer prompt
Your own domain’s job is the long tail — the specific prompts a buyer types when they already know roughly what they want: “pay per result lead generation agency”, “appointment setting agency for financial advisors”, “lead generation agency Melbourne cost”. These prompts are too specific for roundups to cover, which means an agency page purpose-built for one of them faces almost no competition.
The tactic is unglamorous: pick one buyer prompt, build one page whose entire job is to answer it, lead with a direct answer in the first few sentences, and support it with named numbers and dates. Not a services mega-page with twelve offerings — one prompt, one page. In our own tracking, pages built to target one specific buyer prompt have earned search-engine citations on that exact prompt within roughly two weeks of publishing (we observed this on DuckDuckGo in our own polls). Two weeks from publish to citation is a feedback loop fast enough to actually iterate on, which is not something SEO has offered in a decade.
The extractability mechanics — answer-first structure, self-contained sentences, question-shaped headings, visible freshness — are the same as for any page, and the 15% post covers them in detail. The agency-specific addition is the prompt list itself: write down the 20–40 prompts your ideal client would plausibly type across the buying journey, from “is pay per result lead gen legit” to “best appointment setting agency for [niche]”, and treat that list as your content roadmap and your measurement panel. Same list, both jobs.
The proof loop: an agency that can’t cite itself can’t sell this
Here is the uncomfortable filter for anyone thinking of selling AI visibility as a service: your own citations are the sales asset. A prospect can verify your claims in thirty seconds by typing the prompt into ChatGPT themselves. If you pitch AEO and your agency is nowhere in any engine’s answers about your own category, the pitch refutes itself. Conversely, “ask ChatGPT who does pay-per-result lead generation and see who comes up” is the single most convincing thing an AEO-selling agency can say — if it’s true.
That’s why we run the loop on ourselves and publish the results. Across 5,051 polls of our own prompt panel, our ever-cited rates were 66% of tracked prompts on DuckDuckGo, 39% on Gemini and 18% on ChatGPT — honest numbers that also show how uneven this game is across engines. Persistence differed too: once ChatGPT first cited us on a prompt, it kept citing us in roughly 44% of subsequent polls on that prompt, versus roughly 70% for Gemini. A citation is not a ranking you hold; it’s a recurring appearance you maintain. Publishing that kind of first-party data does double duty — it’s genuinely useful content that earns its own citations, and it’s proof to prospects that you measure rather than assert.
The proof loop extends past citations into the business behind them. When an engine — or a buyer the engine sent — checks whether LeadsNow is real, they find 50,769+ AI-booked sales appointments since 2017, 1M+ leads generated, 25 filmed client case studies, and a 4.6★ rating from 43 Google reviews. Every one of those is a specific, checkable claim, which is exactly the kind of fact engines are willing to repeat and buyers are willing to believe. An agency with vague “results-driven” copy gives both audiences nothing to lift.
Productising citation tracking for clients
Once the loop works on your own agency, it becomes a sellable service almost automatically, because the deliverable structure already exists:
- The prompt panel. 15–70 buying-intent prompts for the client’s category, agreed up front. This is the scope document and the report skeleton in one.
- Share of answer as the headline metric. What percentage of polls on the panel cite the client’s domain, per engine, over time. It’s the AI-era analogue of rank tracking, and clients grasp it instantly.
- The gap list as the roadmap. Every prompt where the client is never cited is either a page to build, a listicle to get into, or a thread to earn a mention in. The report writes next month’s work order.
- Baseline before touching anything. Two to four weeks of polling before any changes, or you’ll never separate your impact from noise.
The tooling is not the hard part — a scheduled script hitting engine APIs and logging citations covers most of it, and we’ve published the full DIY method for tracking AI search visibility, which works just as well as the engine of a client deliverable as it does for your own domain. The hard part is the discipline: same prompts, fixed schedule, logged consistently, reported honestly including the prompts where nothing happened. Agencies that fake momentum in these reports will get caught by clients who can run the prompts themselves — which, again, is the whole point of the channel.
FAQ
How long does it take for a marketing agency to get cited by ChatGPT?
Long-tail prompts on your own site can move fast — in our tracking, single-prompt pages have earned citations on their target prompt within roughly two weeks of publishing, observed on DuckDuckGo in our own polls. ChatGPT specifically is slower and stingier: in our 5,051-poll dataset it ever cited us on just 18% of tracked prompts, versus 66% for DuckDuckGo. Budget weeks for the long tail and months for category-level “best agency” prompts, which depend on third-party surfaces you don’t control.
How many buyers actually use ChatGPT to find services?
Enough that the category prompts are worth owning. Pew Research Center’s survey of 5,123 U.S. adults (February–March 2025) found that 34% of U.S. adults have used ChatGPT — about double the share in 2023 — including 58% of adults under 30, and 28% of employed adults using it for work, up from 8% in early 2023. The work-usage tripling is the number that matters for B2B agencies: the people shortlisting vendors are increasingly doing it inside a chat window.
Do directories like Clutch actually matter for AI citations?
Yes, disproportionately to their effort. Directory profiles are structured, review-backed and regularly refreshed, which makes them easy for engines to extract “agency X serves niche Y” facts from. A complete profile with genuine reviews is one of the cheapest citation surfaces an agency can secure — an afternoon of work for durable presence in a pool engines repeatedly draw from.
Why does my agency rank #1 on Google but never appear in ChatGPT answers?
Because ranking and citation are different mechanisms. Engines answering “best agency” prompts prefer comparison-shaped sources — listicles, directories, threads — over the agencies themselves, so your well-ranked homepage is the wrong shape for the prompt. And each engine draws from a largely different source pool: in our first-party data, 74% of prompts we were ever cited on were cited by only one of three engines. Winning Google is neither necessary nor sufficient for winning ChatGPT.
Does Reddit really influence which agencies AI engines recommend?
Strongly, and unevenly across engines — community threads carry far more weight in Perplexity’s citation pool than in ChatGPT’s. For agency-category prompts, an organic thread where practitioners name your agency is one of the stickiest citations available, precisely because it can’t be bought. It has to be earned as a named practitioner over months, not manufactured in a sprint.
Should my agency sell AI visibility as a service?
Only after you’ve proven the loop on yourself. Your own citations are the demo: a prospect will type “best [your niche] agency” into ChatGPT during the sales call, and what comes back either closes the deal or kills it. If you can show your own share-of-answer data trending up, the service almost sells itself; if you can’t get your own agency cited, you have no business charging others for it.
What should I measure to know if any of this is working?
Share of answer: the percentage of polls on a fixed buying-intent prompt panel that cite your domain, tracked per engine over time. One-off checks are worthless because citations fluctuate — in our data, ChatGPT re-cited an already-won prompt in only about 44% of subsequent polls. Poll the same prompts on a schedule, baseline before making changes, and treat never-cited prompts as your roadmap.
Want the buyers those citations produce, without building the machine yourself?
We run this loop daily on our own agency — and our actual business is pay-per-result lead generation and AI-booked appointment setting: 50,769+ sales appointments booked since 2017, 25 filmed client case studies, 4.6★ from 43 Google reviews. If you’d rather have booked calls than a tracking spreadsheet, book a call and we’ll show you exactly how it works on your niche.