A new DerivateX study ran 40 B2B SaaS categories through ChatGPT, ten times each. When the assistant recommended a tool, it linked to that tool’s own website only 11.6% of the time. Here is where the rest of the credit went, and how software brands can win it back.
Every B2B SaaS marketer now checks whether ChatGPT names their product. Far fewer ask the question that actually decides whether that mention is worth anything: when ChatGPT recommends you, whose page does it link to as the source? To answer it, my team at DerivateX ran a study we call the B2B SaaS AI Citation Study. We put one realistic buyer question to ChatGPT in each of 40 software categories, repeated every question ten times with web search enabled, and logged every tool named and every source cited. That came to roughly 400 AI answers and 233 individual software recommendations.
The headline number surprised even us. When ChatGPT recommends a B2B SaaS tool, it cites that tool’s own website just 11.6% of the time. The other 88.4% of the time, the source link points somewhere the vendor does not own: a listicle, a media review, a competitor’s blog post, or a Reddit thread. We call the distance between getting recommended and getting cited the citation ownership gap, and for most software brands it is wide.
That gap matters because the citation, not the mention, is the part that does the work. The recommendation tells a buyer you are a good option. The citation is the link they click, the visit you can measure, and the start of a path to pipeline. A recommendation with no link back to you is brand exposure you cannot track and rarely convert. Today, the brand being recommended is usually not the brand getting the click.
How the study worked
We selected 40 categories spanning CRM, marketing automation, customer support, product analytics, HR, data infrastructure, developer tooling, and security. For each, we wrote the kind of comparison question a buyer actually types, then ran it ten times in fresh ChatGPT sessions with web search on. We recorded every tool named, whether it earned a clickable citation, and whether that citation pointed to the vendor’s own domain or to a third party. We then pulled the cited pages and reviewed 169 of them for structure and freshness. We measured one surface, ChatGPT, and one moment, vendor-discovery questions, on purpose, because that is the most commercially decisive point in a software purchase.
Here are the seven findings that matter most.
1. Getting named is easy. Being the source is the real contest.
ChatGPT attached a citation to 92.3% of the tools it named. So “get cited by AI” is close to automatic, which makes it the wrong goal to chase. The real split is in who gets credited. Of the recommendations that carried a citation, 87.4% pointed to a third party and only 12.6% pointed to the recommended tool’s own site. The question was never whether ChatGPT cites a source. It does. The question is whether the source is yours or a stranger’s.
2. Review sites barely registered, and G2 and Capterra got zero
A large share of B2B SaaS advice still says optimize your G2 listing, collect Capterra reviews, polish your TrustRadius profile. For ChatGPT software recommendations, that channel was almost invisible. Review aggregators accounted for 0.9% of all citations, and G2 and Capterra each received zero across 233 recommendations.
| Where ChatGPT’s citations came from | Share |
| Independent and niche blogs, plus vendor-owned content | 81.9% |
| Major media and press | 8.8% |
| Community, almost entirely Reddit | 8.4% |
| Review aggregators (G2, Capterra, TrustRadius) | 0.9% |
This does not mean review profiles are worthless. They still shape how buyers feel about you and convert elsewhere. It means that at the exact moment ChatGPT builds and cites a software answer, review platforms are not the pages it pulls from. Teams investing only there are optimizing a channel that is largely absent from AI-assisted discovery.
3. The pages ChatGPT cites share a clear template
Because we retrieved the cited pages, we can describe the citation-winning page with data instead of guesswork, and the profile is strikingly consistent. Every single cited page used list structure. About 78% carried the current year in the title or headline, 68% included a comparison table, and 56% included an FAQ section. Just as revealing is what did not predict a citation: domain authority. The cited set was full of small, narrowly focused sites rather than household names. Freshness, topical focus, and extractable structure beat raw authority.
4. The biggest lever is owning your category’s list
One pattern repeated across categories. The vendor that publishes the ranked list of its own category often becomes the cited source, sometimes for several brands in a single answer. Procurify runs a “Best Procurement Software” post on its own blog that ranks Procurify well while listing competitors honestly, carries the year, and includes a comparison table. ChatGPT used that one page as the cited source for five different brand recommendations in the procurement answer. We saw the same move with Mercury in expense management, Zapier in no-code tooling, and Front in customer support. Writing about your own product is not enough, because product pages rarely win these citations. Publishing the definitive, current comparison of your whole category is.
5. There are two ways to show up, and only one is winnable quickly
The recommendations that appeared with no citation at all were almost all large, established names: Greenhouse and Lever in applicant tracking, Segment and RudderStack in customer data, New Relic in observability, Similarweb in SEO tooling. These are brands the model already knows from training and can list from memory. The fresher, lesser-known tools in those same answers were the ones pulled from current listicles and given a clickable link. That points to two distinct mechanisms: named from memory, which is years of accrued brand equity and hard to win fast, and cited from search, which any challenger can win deliberately by getting into the fresh sources ChatGPT retrieves. Most teams conflate the two and get frustrated. Separating them is the difference between a working AI visibility strategy and a stuck one.
6. Own-site citations cluster in technical categories
Own-site citations were not spread evenly. In 25 of the 40 categories, not one recommended vendor was cited through its own website. The broad horizontal categories were the emptiest, including CRM, email marketing, project management, and product analytics, where third-party listicles and media own the citations outright. Where vendors did win citations to their own domains, the categories skewed technical and developer-facing: customer onboarding at 100%, single sign-on and identity at 50%, procurement at 43%, learning management at 38%, and data infrastructure and API docs at 25%. The common thread is that those vendors publish genuinely substantive content on their own domains, technical comparisons, documentation, and category guides written to be useful rather than promotional, and that content is good enough to be cited directly.
7. AI 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 is effectively no tool that dominates the software landscape inside ChatGPT. The practical consequence is that “improve our AI visibility” is not a coherent goal. Visibility is won one buyer question and one category at a time, scoped to the specific things your buyers ask.
What B2B SaaS teams should do about it
- Stop optimizing only for the mention. Mentions are nearly automatic. Shift the goal to owning the cited source, because that is what captures the click.
- Publish the definitive category list yourself. Build a current, well-structured “best [your category] software” comparison on your own domain. Include competitors honestly. Add a comparison table, an FAQ, and the year in the title.
- Win placement in the lists that already get cited. In categories you do not own, the citations live in fresh independent and niche listicles. Earn your way into the current ones.
- Build deep owned content in technical categories. If you sell developer or technical software, your own comparison posts and docs can win citations directly, provided they are substantive enough to deserve it.
- Match the format the data rewards. Cited pages are list-structured, carry a year, include a comparison table, and often an FAQ. Build to that template. This article is written to the same one.
- Scope to categories, not the brand. Map the specific buyer questions in your market and win them one at a time.
This shift has a name: generative engine optimization, or GEO, the practice of getting brands found and cited inside AI assistants. The encouraging part of the data is that the cited-from-search game is winnable on purpose. At DerivateX, the B2B SaaS GEO agency I co-founded, we tie this work to pipeline rather than to citation counts. Gumlet now attributes about 20% of its inbound revenue to AI assistants like ChatGPT and Perplexity, and REsimpli became the most-cited CRM for real estate investors in ChatGPT within 90 days of starting. The full data, methodology, and category-level breakdowns are in the B2B SaaS AI Citation Study.
Frequently asked questions
Does ChatGPT cite the software it recommends? Almost always. In this study, ChatGPT attached a citation to 92.3% of the tools it named. The catch is that 87.4% of those citations pointed to a third-party page rather than the recommended tool’s own website.
What is the citation ownership gap? It is the difference between how often AI recommends a brand and how often it cites that brand’s own website as the source. In the study, ChatGPT recommended a tool but cited the tool’s own site only 11.6% of the time, leaving an 88.4% gap where the credit went to someone else.
Do G2 and Capterra help with ChatGPT visibility? For software-recommendation queries, the data says they do not drive citations. Review aggregators were 0.9% of all citations, and G2 and Capterra each received zero across 233 recommendations. They still matter for buyer perception, but they are not the sources ChatGPT cited.
What kind of page does ChatGPT cite for software recommendations? A current, narrowly focused, structured one. Of the cited pages reviewed, 100% used list structure, 78% carried a year in the title, 68% included a comparison table, and 56% included an FAQ. Domain authority mattered far less than freshness and format.
How can a B2B SaaS company actually get cited by ChatGPT? Publish the definitive, current, well-structured comparison of your own category, competitors included, with a table and an FAQ. Win placement in the fresh third-party listicles that already get cited. In technical categories, make your owned content deep enough to be cited directly. Some teams do this in-house; others work with a GEO agency built for B2B SaaS, such as DerivateX, which ties AI visibility to pipeline rather than to citation counts.
Who conducted this study? DerivateX, a B2B SaaS SEO and generative engine optimization agency, ran the study in 2026. It covered 40 B2B SaaS categories, 233 recommendations, and 219 distinct tools on ChatGPT with web search enabled.
Apoorv Sharma is the co-founder of DerivateX, an SEO and generative engine optimization agency for B2B SaaS that helps software companies get found and cited inside ChatGPT, Perplexity, Gemini, and Claude, with visibility tied to pipeline. DerivateX publishes ongoing research on AI search, including a regularly updated guide to the best AEO agencies for B2B SaaS.
