In a benchmark of B2B SaaS buyer queries, AI search citations did not concentrate the way Google’s top 10 blue links historically have. Across 1,055 citations, the top 10 domains captured only about 17%, the top 50 about 37%, and roughly 62% of cited domains appeared exactly once. The pattern looks fragmented across a long tail of more than 500 domains, at least in this sample. That has a practical implication for generative engine optimization, which I will get to. First, the honest framing: these are numbers from one dataset, so read them as directional.
I’m Andrii Byzov, a fractional CMO for B2B tech, and I run AI search optimization for B2B SaaS. The data here comes from my own AI-Search Visibility Benchmark, and you can read the full methodology in the flagship benchmark report. I am reporting what I found, not claiming it is the last word.
Key takeaways
- In this benchmark, 148 B2B SaaS buyer queries across 28 categories in Perplexity and ChatGPT produced 1,055 citations across 508 unique domains.
- The top 10 domains captured only about 17% of citations, and the top 50 about 37%, which is a flatter distribution than Google’s historical click concentration.
- Roughly 62% of cited domains appeared exactly once, suggesting a very wide long tail rather than a handful of dominant sources.
- The most-cited domains were TechRadar (48), G2 (28 including learn.g2), Reddit (25), HubSpot (21), Zapier (17), Zoho (15), and PCMag (9) in this sample.
- The likely GEO implication: many entry points, but no single placement appears to win AI search on its own. This is one dataset, so treat it as a hypothesis.
What the benchmark measured
The benchmark ran 148 B2B SaaS buyer queries across 28 software categories through Perplexity and ChatGPT, then collected every domain those engines cited in their answers. The queries were the kind a buyer actually asks: best tools in a category, comparisons, and product research questions. The result was 1,055 total citations pointing to 508 unique domains.
That ratio alone is worth sitting with. Just over two citations per domain, on average, across the whole set. If a small number of sources dominated AI answers, you would expect a much higher concentration. Instead the citations spread out. Again, this is one sample with one methodology, so it describes this dataset rather than all of AI search.
The distribution is flatter than Google’s
Here is the part that surprised me most. The top 10 domains together accounted for only about 17% of all citations. The top 50 reached about 37%. And roughly 62% of the 508 domains were cited exactly once.
| Domain tier | Share of citations (this sample) |
|---|---|
| Top 10 domains | ~17% |
| Top 50 domains | ~37% |
| Cited exactly once | ~62% of domains |
Compare that to the mental model most B2B marketers carry from search engine optimization. Google has historically concentrated attention and clicks into the top handful of results, which is why ranking position mattered so intensely. In this benchmark the citation pattern looked different in shape: more sources, each appearing fewer times, with a long tail that runs into the hundreds.
I want to be careful here. This is not evidence that Google is going away, and it is not a claim that AI search is somehow fairer or that one source is more trusted than another. It is a description of how citations were distributed in this particular set of queries. The shape is flatter. That is the finding.
Who actually got cited
The most-cited domains in the sample give a sense of the kinds of sources these engines reached for. TechRadar led with 48 citations. G2 had 28, counting its learn.g2 subdomain. Reddit appeared 25 times, HubSpot 21, Zapier 17, Zoho 15, and PCMag 9.
A few observations, offered as patterns rather than rules:
- Editorial review media shows up strongly. TechRadar and PCMag are publisher review sites, and they appeared often. That is consistent with what I see in the role of review sites in AI search visibility, though this is appearance count, not a measure of influence.
- Review platforms and communities both appear. G2 and Reddit both ranked near the top, which suggests structured review data and unstructured community discussion can both surface. I would not over-read it.
- Vendor and tool sites appear too. HubSpot, Zapier, and Zoho are operators in or adjacent to the categories queried, so their own properties got cited as well.
Even the most-cited domain, TechRadar at 48, represents under 5% of the 1,055 total citations. No single source dominated. The leaders led by a little, not by a lot.
Why fragmentation widens the GEO opportunity
If you take the distribution at face value, and I would only take it as far as one dataset allows, it cuts two ways for generative engine optimization.
The encouraging read: a fragmented citation landscape implies many entry points rather than a few gates you must own. When 62% of cited domains appear once, the field is not locked up by incumbents the way a Google top 10 can feel locked. A well-placed mention on a review site, a substantive community thread, a comparison page, or your own answer-shaped content could each become a citation. The surface area is large.
The sobering read: if no single placement captures a meaningful share of citations, then no single placement wins AI search on its own. There is probably no equivalent of the number one blue link to chase. That points toward coverage. Being present across review platforms, communities, editorial media where it is earned, and your own content seems likelier to compound than betting everything on one spot. This is a hypothesis from one benchmark, not a guarantee, and your category may behave differently.
That is also why measurement matters more than ever. If citations are spread thin, you cannot eyeball your presence. You have to track it across engines and queries, which is the work I describe in how to track AI visibility and in my AI search optimization approach. The full numbers and method live in the AI-Search Visibility Benchmark report if you want to check my work.
The honest caveat
This is one benchmark: two engines, 148 queries, 28 categories, a single point in time. The distribution I found may not hold in your category, in other engines, or six months from now. Citation counts are not a measure of trust, influence, or buying impact, only of how often a domain appeared in these answers. Read every number here as directional.
What I am fairly confident about is the shape of the thing. In this sample, AI search did not reward concentration the way classic search did. It spread citations across a long tail, which suggests the GEO opportunity is wide and the worst strategy is probably to bet on one placement. If you want to pressure-test what this means for your category, I am on LinkedIn, and I help B2B SaaS teams with GEO directly.
FAQ
How concentrated are AI search citations in B2B SaaS? In my benchmark of 148 B2B SaaS buyer queries across 28 categories in Perplexity and ChatGPT, the top 10 domains captured only about 17% of 1,055 citations, the top 50 about 37%, and roughly 62% of the 508 cited domains appeared once. Citations looked fragmented, not concentrated, in this sample. Figures are directional.
Which domains get cited most in AI search for B2B SaaS? The most-cited domains were TechRadar (48), G2 (28 including learn.g2), Reddit (25), HubSpot (21), Zapier (17), Zoho (15), and PCMag (9). These are appearance counts in one sample, not a ranking of trust, and the long tail beyond them is very wide.
How is this different from Google’s top 10 blue links? Google has historically concentrated clicks in a small set of top results. In this benchmark the pattern looked flatter, with the top 10 domains holding only about 17% of citations and most domains appearing once. That is a different shape, not evidence that any channel is dead or any source is more trusted.
What does fragmented citation mean for GEO strategy? It suggests many entry points rather than a few you must own, which can widen the opportunity, but it also means no single placement appears to win AI search alone. Coverage across review sites, communities, and your own answer-shaped content may matter more than one spot. Test it in your own category.