In ChatGPT's SaaS answers, user-generated domains outcite publishers four to one

An analysis of roughly 35,000 ChatGPT citation URLs for US-only B2B SaaS prompts found that user-generated content platforms accounted for 17.1% of cited domains, against 4.0% for publishers ^[1][3]. The work was published by Kevin Indig on G2, using citations captured through Profound during December 2025 ^[2]. If that mix is representative, the source layer sitting in front of a software buyer is mostly made of properties no marketing team owns or can edit.

The arithmetic on the gap is straightforward: UGC domains carry about 4.3 times the cited-domain share of publisher domains in this sample ^[19]. The gap also holds across buyer stages. UGC stays in the high teens through the stages studied while publishers sit in the 4% to 5% band ^[6], and the evaluation stage, which is UGC's low point in the dataset, still runs well clear of the publisher range ^[7].

None of that displaces owned content. Vendor and other domains are the largest cited category at 70.4% overall ^[4] and remain the largest throughout the dataset ^[8], which is an argument for keeping documentation, pricing, and comparison pages accurate rather than for spending the budget on community seeding. Worth noting: the three reported buckets sum to 91.5%, leaving 8.5 points unaccounted for in the published breakdown ^[20]. That is a question to put to the methodology, not a finding.

Read the metric carefully. The classification is de-duplicated at the domain level ^[5], so this is a diversity count, not a volume count. As the second analysis in the supplied material points out, unique domains answer how many distinct sites were selected while citation volume answers how often they appeared, and neither on its own shows that a domain is consistently influential ^[15]. The same analysis reports that Search Engine Land's coverage of Indig's dataset found roughly 30 domains accounting for a large share of citations, with the first-party/third-party balance shifting by platform and prompt type ^[13]. A separate BeVisibleIQ study of 75 B2B SaaS buyer prompts across four engines likewise found that vendor and third-party sources do not appear at the same rate across engines ^[14]. Findings from US ChatGPT prompts do not transfer automatically to other engines or geographies ^[16], and the second source is explicit that the supplied evidence does not tie the full methodology cleanly to one published dataset ^[12].

What the number supports is narrow. The UGC platforms named include Reddit, YouTube, and Quora, which surface implementation discussion and product comparisons ^[9]; the analysis does not assess whether any individual citation is accurate ^[10], and it does not show that posting more often earns citations ^[11].

The operational consequence is consistency, not activity. Enterprise evaluators increasingly use AI assistants during software research, so citation behaviour shapes what reaches them before they ever load a vendor site ^[21], and mismatched pricing or product claims across documentation, review threads, and generated answers leave a buyer resolving the contradiction themselves ^[18]. The program the source recommends is unglamorous: track which domains appear for priority prompts and how the mix shifts by stage, keep owned product and pricing data correct, monitor recurring community questions for information gaps, and define who may respond and when an inaccuracy gets escalated ^[17].