We ran 100 real B2B buyer prompts through a search-enabled AI model.
Not abstract prompts. Buyer prompts.
Things like CRM comparisons, API tools, marketing platforms, cybersecurity software, e-commerce platforms, HR tools, and AI products.
Then we extracted every citation.
The final dataset had:
- 100 prompts
- 328 citations
- 239 unique domains
- 3.3 citations per response on average
The findings were blunt.
AI does not mostly cite vendor pages.
It does not seem especially obsessed with documentation.
It heavily favours editorial content, comparison pages, “best tools” roundups, year-stamped titles, and structured articles that help it answer buyer questions quickly.
A few numbers from the report:
- TechRadar appeared in 23% of all responses
- YouTube was the second most-cited domain
- 133 citations pointed to blog/article URLs
- Only 5 citations pointed to documentation
- 55% of cited titles included “best” or “top”
- 48% included a year
- About 25% of cited URLs used comparison-style language
This does not mean everyone should now publish 40 lazy “best software” listicles and call it an AI visibility strategy.
That would be very on-brand for the industry. Also mostly rubbish.
The better lesson is that AI systems appear to cite pages that help them make buyer-facing judgements: comparisons, rankings, tradeoffs, current options, category context, and clear structure.
In other words, the content that gets cited is often not the content that explains your product.
It is the content that helps a buyer choose.
We published the full breakdown here:
It includes the methodology, most-cited domains, category breakdowns, title patterns, URL patterns, and the full prompt/citation appendix.
Useful if you care about how B2B buyers are going to discover vendors when search becomes less like a results page and more like a recommendation engine.
