Is UGC really AI's largest third-party source? 17.1% of ChatGPT SaaS citations — but 3–4% in our small-business corpus

What the G2 analysis found

The Growth Memo piece, published 2026-08-10 and syndicated by Search Engine Land on 2026-08-12, draws on an analysis Indig ran for G2 in February 2026: roughly 35,000 citation URLs captured in Profound — US only, ChatGPT only, month of December 2025 — across 3,177 SaaS-vendor prompts classified into four buyer-journey stages (discovery 376, exploration 1,179, evaluation 1,367, focused evaluation 255). Each citation URL was reduced to a root domain, classed as review platform, UGC platform, publisher, or vendor/other, and de-duplicated to one record per run, intent and domain. The findings: UGC platforms hold 17.1% of cited domains overall against 4.0% for publishers; the UGC share barely moves across journey stages (17.8%, 18.2%, 15.1%, 17.2%); and "Wikipedia, Reddit, and LinkedIn account for 99% of UGC citations in this sample," with Wikipedia alone running 10.1–14.0 points of the 17-point floor.

The same cut on our corpus reads 3–4%, not 17%

Our instrument differs in ways that matter — 12 pinned small-business buyer questions × 4 engines (ChatGPT, Perplexity, Gemini, Claude) × 3 samples = 144 production answers per run, citations read from each engine's returned citation payload, counted here as domain-answer pairs (a domain cited by one answer counts once, closely mirroring the G2 dedup rule) — so this is a consistency observation across verticals, not a replication. On the three dated runs:

Widening the UGC bucket to also count YouTube, Medium, Quora and Facebook only lifts the share to 5.1–6.5%. And the composition inverts the G2 picture: in SaaS prompts Wikipedia is the largest UGC component; in our small-business buyer questions Wikipedia was cited by 4–5 answers out of ~430 citing pairs per run — near zero — while Reddit carries roughly three quarters of the UGC slice, cited by three of the four engines with Perplexity doing most of the citing. LinkedIn, in our corpus, is a Perplexity-only phenomenon.

Why both numbers are true — and what to do with that

• UGC share is vertical-dependent. SaaS categories have dense Wikipedia pages, active subreddit debates, and LinkedIn thought-leadership; a plumber, dentist or local agency category mostly has none of that. The encyclopedia layer that gives SaaS its 10–14-point Wikipedia floor simply does not exist for most local service queries.
• Engine mix changes the answer. The G2 sample is ChatGPT-only; our corpus spans four engines whose citation behavior differs sharply. Treating any one engine's source mix as "AI's" source mix overstates how far the finding travels.
• The actionable overlap is Reddit. It is the one UGC platform that is material in both corpora — which is consistent with community threads being the place where AI engines find candid vendor comparisons in nearly every vertical.

The practical takeaway for a small business is to distrust category-level averages in either direction: the source mix behind YOUR customers' questions is an empirical fact about your vertical, not a constant of AI search. Check which sources the engines actually cite when asked about a business like yours before spending an authority budget on platforms that may hold four percent — or seventeen — of your answer field.