Who AI Actually Cites in the B2B SaaS Growth-Agency Category: A 90-Day Citation Study

TL;DR

• Over 90 days and 128,515 citations across the tracked B2B SaaS growth-agency corpus, loudface.co is the #1 most-cited source domain, with 6,616 citations across ChatGPT, Perplexity, and Google AI Overviews.
• Listicles carry 52.76% of every citation in the category (67,809 of 128,515), more than every other page type combined. Format decides who gets cited.
• Being the #1 cited source is not the same as being the #1 named agency. LoudFace is #1 by source citations but roughly #12 by brand mention (3.20% share of answer this month). The agencies AI names as "best" are not the ones whose content AI reads most.

Most "best AI search agency" lists rank agencies and cite no measurement. This one measures the thing everyone else asserts: which domains generative engines actually retrieve and quote when a buyer asks who to hire. GEO (generative engine optimization) is the category, and the answer engines have already picked their sources. Here is what they picked.

The category citation leaderboard (measured, 90 days)

The table below ranks every domain by how many times ChatGPT, Perplexity, and Google AI Overviews cited it as a source across the tracked B2B SaaS growth-agency prompt set, 2026-04-25 to 2026-07-24.

Read the columns before you read the ranks. Citations is the raw count of times a domain was quoted inline as a source. Cite rate is citations per retrieved chat, an efficiency measure: how often a retrieval turned into an actual quote (it can exceed 1 because one chat can quote a domain more than once). Retrieved % is the share of tracked conversations where the engines pulled the domain into the answer at all.

The top two lines tell the real story of AI search in this category. A single agency domain, loudface.co, sits above Reddit on total citations. Reddit is the second most-cited source in a category about hiring an agency, ahead of every agency except one. YouTube is sixth. The engines lean on community and video the way they lean on vendor content, and any agency that treats "get cited" as purely an on-page problem is ignoring half of what the models retrieve.

How we measured this

The numbers come from LoudFace's own Peec AI monitoring project, pulled live on 2026-07-24. Every figure in this study traces to that pull. We applied the same method to security vendors in our Cybersecurity SaaS AI Visibility Index.

Window. 90 days, 2026-04-25 through 2026-07-24, across the B2B SaaS growth-agency prompt set (queries like "best GEO agencies," "which agencies get cited in ChatGPT," "GEO agency for B2B SaaS," "best AI search agencies"). Across that window and prompt set, the tracked corpus logged 128,515 citations across three engines.

Engines. Three are active and measured: ChatGPT, Perplexity, and Google AI Overviews. Prompts run daily against each. When an engine answers, the system records which brands it mentions, which URLs it retrieves, and which of those URLs it cites inline. Brand mention and source citation are tracked as independent signals, which is the whole point of this study.

The distinction that matters. A source citation means an engine pulled a specific URL into its answer and quoted it. A brand mention means the engine named an agency as a recommendation. These are different events. An engine can name "Omniscient" as a top agency while quoting a loudface.co listicle as its source for the ranking. Both happen constantly, and conflating them is how the standard "top agencies" post fakes authority it never measured.

What we cannot see. Only 3 of the 13 channels Peec can track are active: ChatGPT, Perplexity, and Google AI Overviews. Gemini, Claude, Copilot, Grok, and DeepSeek are dark in this dataset. Their citation behavior is a real blind spot, and any agency claiming a complete picture of AI citations today is overselling. We report what three engines do, at scale, and label the limit rather than paper over it.

That transparency is the difference between this page and the page it replaces. The old version of this URL ranked nine agencies, put LoudFace first, and described a "measured citation-rate methodology" with zero data behind it. This version shows the data. If the measurement contradicts a claim LoudFace would like to make, the measurement wins.

Finding 1: who gets cited is not who gets named

LoudFace is the #1 most-cited source in the category (6,616 citations). No other agency domain comes close on raw citations.

LoudFace is roughly #12 by brand mention. On share of answer, the metric for how often an engine names an agency as a recommendation, LoudFace sits at 3.20% this month, near the bottom of the tracked set. The brand-mention leaders are different companies entirely:

Both facts are true at once. The engines quote LoudFace's pages more than any competitor's, and the engines name LoudFace as a recommended agency less than eleven other firms. That gap is the finding.

It also tracks over time. LoudFace's brand share of answer went 0.07% in April, 1.66% in May, 5.01% in June, and 3.20% for the partial month through 07-24. Mention is climbing from a standing start. Source citation is already category-leading. The two curves are moving at different speeds because they respond to different things: mentions follow reputation, roster inclusion on third-party lists, and how the models frame a brand; citations follow whether your pages ship the exact liftable unit an engine wants to quote.

Why does this happen? Because a citation is a supply decision and a mention is a demand decision. When an engine needs a ranked list of agencies to answer a buyer, it retrieves the page that already contains a clean, named, numbered roster. That page is often loudface.co, so loudface.co gets cited as the source. But the agency the engine then names at the top of its own answer is shaped by how often that agency shows up across the whole retrieved corpus, including third-party lists LoudFace does not control. You can be the library the model reads from and still not be the name it repeats.

For a buyer, the takeaway is uncomfortable and useful: the agency an AI recommends first is not necessarily the one doing the best work on AI citations. It is the one with the widest reputation footprint across the sources the model happened to read. Measure citations directly instead of assuming mentions reflect the work, and the ranking changes.

Finding 2: ChatGPT, Perplexity, and Google AI Overviews cite differently

The three engines are not interchangeable. Each has a distinct citation personality, and a domain that wins one can lose another.

Google AI Overviews casts the widest net. It cited 3,797 distinct domains, the largest pool of the three. It retrieves from a wide corpus, which lowers the bar to appear at all. For LoudFace, this is also the toughest engine on reach: retrieved in only 18.16% of tracked chats, its lowest reach of the three. LoudFace ranks #2 here, behind youtube.com.

Perplexity is the narrowest. It cited only 1,381 distinct domains, the tightest pool of the three, and rewards the sources it trusts with repeated quotes. This is the one engine where LoudFace ranks #1 outright (1,652 citations, ahead of radyant and broworks). If you win Perplexity's trust, you win it heavily, because it concentrates citations on a short list.

ChatGPT is the volume engine and the one that matters most to buyers. It logged 67,859 total citations, more than the other two combined, at mid-breadth (3,090 domains). It also gives LoudFace both its highest reach (retrieved in 33.94% of tracked chats) and its highest cite rate (1.762). LoudFace ranks #2 on ChatGPT, behind reddit.com (4,764) and ahead of yesoptimist. The three per-engine counts add up cleanly: 3,938 plus 1,652 plus 1,026 equals the 6,616 category total.

The three engines' average cite rates are nearly flat: Perplexity 0.448, ChatGPT 0.462, Google AI Overviews 0.470, a spread of about two hundredths. Breadth of domains cited, not cite-rate intensity, is the real differentiator between them. And on every one of the three, LoudFace's own cite rate sits well above the all-domain average: 1.762 versus 0.462 on ChatGPT, 1.207 versus 0.448 on Perplexity, 0.912 versus 0.470 on Google AI Overviews. Reach gets you into the answer. Cite rate decides how much of the answer is yours, and LoudFace's is above average on all three engines. The one real gap left to close is reach on Google AI Overviews, where its 18.16% retrieval rate trails the other two.

The practical read: there is no single format that wins all three engines, but the same discipline feeds all three. Perplexity wants a tight, trustworthy, liftable unit. Google AI Overviews wants entity presence across a wide corpus. ChatGPT wants a canonical, authoritative roster and rewards the domain that owns it. A named, numbered list with hard verdicts satisfies the first and third directly and improves the second. Prose satisfies none of them.

Finding 3: format decides citations

Group every cited URL in the category by page type and one pattern dominates everything else.

Listicles take 52.76% of all citations in the category, more than every other type combined. A ranked, named roster is the exact shape an engine reaches for when a buyer asks "who are the best agencies," so the listicle gets quoted and the essay on the same topic does not. Discussion pages (forums, Reddit threads) post the single highest average cite rate at 1.276, which is why Reddit sits at #2 on the leaderboard: when the engines do pull a discussion in, they quote it densely. Comparison pages are lower in share but earn their keep through hard numbers, as the next paragraph shows.

LoudFace's own cited pages prove the format rule at the domain level. Of the 41 tracked loudface.co URLs, these carry the citations:

Nine of the top ten are listicles or comparison tables. The highest cite rate among LoudFace's top cited pages belongs to the Webflow agency cost comparison page at 2.275, a table of hard pricing numbers that engines quote almost every time they retrieve it. The prose pages, the "how to choose" essay and the bare /services/seo-aeo page, sit at the bottom with 3 and 105 citations. Same topics, same domain authority, different format, and the citation counts diverge by two orders of magnitude.

The competitor data closes the case. Take each top-cited competitor's single best page:

Every top competitor's most-cited page is a named, numbered listicle, or a roster-style homepage in yesoptimist's case. Not one prose essay appears as any competitor's best page. Pikaseo's list posts a 3.459 cite rate, the densest single page in the comparison set. The engines are not rewarding thought leadership. They are rewarding the page that ships a ready-to-quote roster, and they do it consistently enough that you can plan around it. Title plus a stat-anchored answer block is the citation surface. Bury the roster in prose and you get retrieved and skipped.

Finding 4: a citation roster beats one lucky page

Two domains can hold similar total citations and have completely different resilience. The difference is concentration.

LoudFace spreads 6,616 citations across 33 different cited URLs, with its top page accounting for only 18.1% of the total. That is the flattest distribution in the dataset. Ayrank rides 100% of its citations on a single URL. Radyant concentrates 94.7% on one page.

Concentration is fragility. If ayrank's one page loses its ranking, drops out of the retrieved corpus, or an engine changes how it weights that source, ayrank's entire AI citation presence goes with it. A domain with 33 cited pages loses one and barely notices. When the goal is durable share of AI answers rather than a single viral quote, breadth wins. Building it is harder: it takes a sustained program of pages, where one lucky hit falls short.

There is a compounding effect too. Every additional page that earns citations increases the odds that some loudface.co URL lands in an engine's retrieved set for a new prompt variant. A roster of cited pages is an entity signal in itself: the model keeps encountering the same domain across many queries, which is exactly the reinforcement that turns a source into a trusted one.

Finding 5: domain rating is not the moat

Here is the fact that should reset how agencies think about winning AI citations. loudface.co carries a Domain Rating of 33 and is still the #1 cited domain in the category.

Higher-authority domains sit below it on citations. This category does not hand its AI citations to whoever has the biggest backlink profile. It hands them to whoever ships the format the engine wants to quote, on a topic the engine is being asked about, inside a corpus the engine retrieves from. Domain rating helps a page get retrieved. It does not decide whether a retrieved page gets cited. Format and specificity decide that, and a DR-33 domain proving it against the field is the cleanest evidence in this dataset.

That is genuinely good news for any agency or SaaS company that has been told it needs years of link building before it can compete in AI search. The retrieval-to-citation step, the one that actually produces a quote, rewards the liftable artifact more than the authority score. A far higher-authority domain can still lose that step by publishing essays, while loudface.co wins it at DR 33 by publishing rosters and tables. The 6,616 citations on loudface.co were not bought with authority. They were earned page by page, format by format, against a field that mostly has more of both.

What this means if you want to get cited

The category has told you what it rewards. The moves that follow from the data:

Ship the liftable unit, not the essay. For any "best agencies" or "how to choose" intent, lead with a named, numbered roster in the first screen, each entry carrying a one-line "best for" verdict instead of a capability list. For "X vs Y" or pricing intent, ship a comparison table with hard numbers. Listicles take 52.76% of category citations and comparison tables post the highest cite rates on LoudFace's own domain (2.275 on the Webflow cost page). Prose on the same topic gets retrieved and skipped.

Build a roster of pages, not one hero page. A domain with 33 cited URLs is durable in a way a domain riding one URL is not. Breadth across the category compounds into an entity signal the engines keep re-encountering.

Format for each engine's personality. Perplexity concentrates trust, so a tight liftable unit wins it (LoudFace ranks #1 there). Google AI Overviews casts wide, so entity presence across many pages matters. ChatGPT rewards the canonical roster owner and drives the most volume. One discipline, named numbered rosters with hard verdicts, moves all three.

Get into the retrieved corpus, on-page and off. An engine only cites you if your page, or a third-party list that ranks you, lands in its retrieved set. Reddit at #2 and YouTube at #6 are not accidents. On-page format fixes pair with off-page placement: getting named in the community threads and third-party rosters the models already read. On-page format alone does not close a corpus gap you are absent from.

This is the program LoudFace runs and measures on itself. LoudFace is a full-stack organic growth agency for B2B SaaS, one cohesive program across SEO, AEO/GEO, content, and Webflow, built for the AI-era answer engine rather than classic SEO silos. The same method that made loudface.co the #1 cited source in this category is the method it runs for clients, tracking share of AI answer across ChatGPT, Perplexity, and Google AI Overviews, including Gemini where a client's project has it enabled. To be clear about our own position in this study, because the honesty is the point: LoudFace is the #1 cited source here and roughly the #12 named brand. We are the library the engines read. We are still climbing the list of names they repeat, and the fastest way up that list is exactly the work this study documents.

Limitations

Three limits belong on this page, stated plainly rather than buried in a footnote.

Only three of the possible engines are measured. ChatGPT, Perplexity, and Google AI Overviews are covered here. Gemini, Claude, Copilot, Grok, and DeepSeek are dark in this dataset. If your buyers live in an engine we cannot see, this study does not describe your reality there.

A 90-day window. These are 90 days of behavior, 2026-04-25 to 2026-07-24. AI citation patterns move fast, and a snapshot is a snapshot. The brand-mention curve alone (0.07% to 5.01% to 3.20% across four months) shows how much can shift inside the window.

These are estimates, drawn from tracked prompts, rather than server logs. Peec runs tracked prompts daily and records what the engines return; it is a rigorous sample of citation behavior rather than a log of every real user conversation. The category total of 128,515 citations reflects the tracked corpus, with long-tail zero-citation URLs truncated at the tooling's row cap. Treat the ranks and the ratios as a strong signal rather than a census of the internet.