The post
I've been running buyer questions for small businesses through ChatGPT, Gemini, Claude, Perplexity and Meta's Muse, 3 runs each, web search on, and logging every cited URL. 60 businesses, 4,364 answers. Mostly small SaaS, a few local and service businesses, so treat it as a floor.
The part that changed how I advise people: the AIs don't read the same places.
- Reddit was a source in 49 of 60 checks, but 143 of its 205 citations came from Gemini. Claude cited Reddit in 0 of 262 answers that cited anything. Muse 0 of 562.
- G2 or Capterra: about 1 in 5 of Claude's cited answers (57 of 262). ChatGPT about 1 in 100 (9 of 894).
- When Claude linked anything, 68% of the time at least one link was a "best / top / alternatives / vs / review" page. ChatGPT 6%. ChatGPT mostly linked to product pages.
- The same tool-list sites showed up across unrelated categories: toolradar (28 checks), alternativeto (22), sourceforge (21).
- Of the 132 answers that named the business, 99 also cited its own site.
- Run-to-run: when a business was named at all, it was named in all 3 runs only 21 of 72 times.
So "get mentioned on Reddit" is mostly Gemini advice, review sites are mostly Claude advice, and your own pages carry ChatGPT.
Caveats: API answers, not the consumer apps; 5 questions per business; I count citations, I can't see why a model picked them.
Disclosure: I run Heard of You, a small service that runs these checks for businesses, and the study is on its site. The numbers are free to quote.
Method and the full tables: https://heardofyou.io/research/what-5-ais-recommend-60-checks
Curious whether people tracking this at scale see the same Gemini/Reddit skew.
View the original thread on Reddit →