Original research
What we measured ourselves, and the data behind it
Each report states its finding in one sentence, publishes the dataset that produced it, and lists what the measurement cannot tell you.
These are measurements rather than opinions. We put a fixed set of buying questions to public AI answer engines on a schedule and record what they cite, so every figure below has a download behind it and a method that says how it was counted. The headline sentence on each report is generated from that dataset rather than transcribed, which means it cannot quietly come to disagree with the file underneath it.
Which sources AI answer engines cite when buyers ask about business software
For commercial software buying questions, AI citations are extraordinarily fragmented, not concentrated. Across 4,515 citations we recorded 2,347 distinct domains, 72% of which were cited exactly once, and the top 25 domains together accounted for 12.9% of all citations. Reddit, the domain most often described as the single biggest source in AI answers, was 0.86%. Wikipedia was 0.22%. Capterra was zero.
Read the reportUpdated 2026-08-24. 2 downloads, CC BY 4.0.
How often AI assistants actually search the web before answering a buying question
34.1% of answers to commercial buying questions involved no web retrieval at all, and it depends enormously on how the question is phrased: 92.8% of category questions ("what are the best AI sales agents") triggered a search, against 38.7% of problem questions ("how do I stop losing deals to slow follow-up"). The answers that skip retrieval are drawn entirely from what the model already believes about the market, and no amount of work on your own website reaches them.
Read the reportUpdated 2026-08-24. 1 download, CC BY 4.0.