The B2B SaaS AI Citation Study

The citation ownership gap

The citation ownership gap is the difference between how often AI recommends a brand and how often it cites that brand's own website as the source. In this study, that gap is wide.

Most software marketers assume that if AI is recommending them, AI must be reading and crediting their content. The data says otherwise. The recommendation and the citation are two separate events. A buyer can read that your product is a strong choice, click the source link beside it, and land on a review site, a competitor's comparison post, or a Reddit thread, never on your own pages.

This matters because the citation is the part that drives the referral visit, the discovery, and the eventual pipeline. A recommendation with no link back to you is brand exposure you cannot measure and cannot convert. The vendor that owns the cited source captures the click, and today that vendor is usually not the one being recommended.

Getting named is easy. Being the source is the real contest.

ChatGPT attaches a citation to 92.3% of the tools it names. Only 7.7% of recommendations had no source at all. So the common goal teams set, "get cited by AI," is close to automatic and therefore the wrong target.

The strategic implication is a reframe. The question is not whether AI will cite a source when it recommends you. It will. The question is whether the cited source is yours or a stranger's. Right now it is almost always a stranger's, which means the entire game is about controlling the page that earns the citation rather than chasing the mention itself.

The G2 and Capterra assumption does not hold

A large share of B2B SaaS GEO advice points teams toward review platforms: get listed on G2, collect Capterra reviews, optimize your TrustRadius profile. In this dataset, that channel barely registers as a cited source.

Here is the full breakdown of where citations actually came from:

Source type	Share of citations
Independent and niche blogs, plus vendor-owned content	81.9%
Major media and press (TechRadar, Forbes Advisor, Reuters)	8.8%
Community and social (almost entirely Reddit)	8.4%
Review aggregators (G2, Capterra, TrustRadius)	0.9%

This does not mean review profiles have no value. They shape buyer perception and conversion in other channels. It means that for ChatGPT software recommendations specifically, review platforms are not the pages ChatGPT pulls from when it builds and cites an answer. Teams optimizing only there are investing in a channel that is largely invisible at the exact moment of AI-assisted discovery.

What cited pages actually look like

Because we retrieved the cited pages, we can describe the citation-winning page with data rather than guesswork. The profile is remarkably consistent.

Every single page ChatGPT cited used list structure. Roughly three in four signaled freshness with a year in the title, with examples such as "Best A/B Testing Tools for Growth Teams in 2026" and "15 Best Email Marketing Platforms 2026." Just as telling is what did not predict citation: domain authority. The cited set is dominated by small, narrowly focused sites rather than household names.

The highest-leverage move is to own the category list

A specific pattern recurred across categories: the vendor that publishes the ranked list of its own category gets cited as the source, often for several brands in the same answer.

Procurify, a procurement software vendor, runs a "Best Procurement Software" article on its own blog. The post ranks Procurify favorably while also listing competitors honestly, and it carries the current year and a comparison table. ChatGPT used that single page as the cited source for five different brand recommendations inside the procurement answer. The same move appears with Mercury in expense management, Zapier in no-code tooling, and Front in customer support.

The takeaway is unusually concrete. Writing about your own product is not enough, because product pages rarely win these citations. The leverage is in publishing the definitive, current, well-structured comparison of your whole category, including competitors with intellectual honesty, so that you become the page AI pulls from to populate the entire answer. Done well, this turns one asset into citations for multiple recommendations at once.

There are two different ways to be visible in AI answers

The recommendations that were named without any citation are revealing. They cluster among large, established category leaders: Greenhouse, Lever, Workable, and iCIMS in applicant tracking; Segment, RudderStack, and Treasure Data in customer data platforms; New Relic in observability; Similarweb and Screaming Frog in SEO tooling. These are precisely the names a large language model already knows from its training data and can list from memory, without needing to fetch a web source.

The fresher, lesser-known tools in those same answers are the ones pulled from current listicles and given a clickable citation. That points to two distinct mechanisms of AI visibility:

• Named from memory. Established incumbents earn unlinked mentions from the model's internal knowledge. This is brand equity accrued over years and is hard to win quickly.
• Cited from search. Everyone else earns a linked citation by appearing in fresh, well-structured third-party content the model retrieves at answer time. This is winnable deliberately and quickly.

A challenger brand cannot shortcut its way into the model's memory. It can, however, win the second game on purpose by getting into and publishing the current sources ChatGPT retrieves. Most teams conflate these two paths, and separating them is the difference between an effective AI visibility strategy and a frustrated one.

Own-site citations cluster in technical and developer categories

Own-site citations were not spread evenly. In 25 of the 40 categories, not a single recommended vendor was cited via its own website. The broad horizontal categories were the emptiest: CRM, email marketing, project management, product analytics, marketing automation, and live chat all returned zero own-site citations. In those markets, third-party listicles and media own the citations entirely.

Where vendors did earn citations to their own domains, the categories skewed technical and developer-facing:

Category	Own-site citation share
Customer onboarding	100%
Single sign-on and identity (WorkOS, Stytch, FusionAuth)	50%
Procurement and spend management	43%
Learning management (Docebo, Absorb, Thinkific)	38%
Data infrastructure (Airbyte) and API docs (Mintlify)	25%

The common thread is that these vendors publish strong, genuinely substantive content on their own domains: technical comparison posts, documentation, and category guides written to be useful rather than promotional. That content is good enough to be cited directly. In crowded horizontal categories, your near-term job is to win placement in the third-party pages that own the citations, or publish your own category list. In technical and developer categories, your own content can win the citation directly, if it is deep enough to deserve it.

Citations concentrate, so winning one page can win the answer

At the domain level the landscape has a long tail, with 124 unique source domains across the citations. The top, however, is concentrated. Reddit is the single most-cited domain, followed by TechRadar. The top 10 domains together account for roughly 30% of all citations.

Most-cited source domains	Citations
reddit.com	16
techradar.com	11
procurify.com (one page, five recommendations)	5
forbes.com	4
zapier.com	4

Within a single answer, ChatGPT frequently leans on one page to fill multiple recommendation slots, as the Procurify example shows. A single well-placed, well-structured list can therefore carry an entire category answer. That is why the work compounds: one strong asset can earn several citations rather than one.

AI software visibility is fragmented and category-specific

Across the 40 categories, 219 distinct tools were named, and 94% of them appeared in only one category. The most-named single tool showed up in just three. There are effectively no tools that dominate across the software landscape inside ChatGPT.

The practical consequence is that "improve our AI visibility" is not a coherent goal. Visibility is won one buyer query and one category at a time. Strategy has to be scoped to the specific questions your buyers ask, not pursued as a single brand-wide initiative.