Which sites does ChatGPT cite most?

google.com is the site ChatGPT cites most, by a wide margin. It took 8.6% of citations, 5,871 in total, and appeared in 1,757 responses. The next named domain, PubMed, is far behind at 1.3%.

The top of this table is thin. Everything outside the top 15 domains, the Other bucket, accounts for 83.5% of citations (57,194). Even the leader holds less than a tenth, which means ChatGPT's sourcing is spread across a long tail of smaller sites rather than concentrated in a handful of publishers.

After google.com there is a steep drop. PubMed gets 1.3%, OpenAI's own help center (help.openai.com) gets 1.1%, and everything else sits at 0.7% or below. help.openai.com appears in 634 responses, second only to google.com's 1,757, so OpenAI leans on its own documentation more than almost any external source.

Official and academic domains dominate the list. Three Indian income tax domains all appear: incometaxindia.gov.in at 0.7%, incometax.gov.in at 0.6%, and wmstatic-prd.incometaxindia.gov.in at 0.3%. On the research side, PubMed (1.3%), PMC (0.6%), ScienceDirect (0.4%), doi.org (0.3%), and arxiv.org (0.3%) round out the top 15. ChatGPT reaches for primary sources, official references, and product documentation rather than news or editorial sites.

Reddit is the outlier, and it is not a good story for forums. Its share of citations fell 87.1 percentage points month over month to 0.5% (323 citations). It now trails help.openai.com and two Indian tax domains.

Lower in the table, developers.google.com shows the opposite pattern. It logged 469 citations across only 64 responses, the fewest of any named domain in the top 15. Where help.openai.com spreads across many answers, developers.google.com gets cited repeatedly inside a small set of technical ones.

For anyone chasing AI visibility, the message is sobering. The leader holds 8.6% and 83.5% of citations sit in a long tail no single publisher controls. There is no rank-for-ChatGPT playbook here; the sites that get cited are the ones ChatGPT already treats as authoritative, mostly official documentation, primary research, and government reference pages.

The practical move is to stop hunting for one citation slot and instead publish the kinds of pages ChatGPT already reaches for: docs, original research, and reference material on stable domains. Reddit's collapse is the cautionary note. A channel that was prominent a month ago can vanish from AI answers overnight, so no single source is a durable backdoor.