Where Do AIs Really Get Their Answers?

TL;DR – Wikipedia still rules ChatGPT, Reddit runs Google’s AI Overviews, and Perplexity lives on forums. But 86% of the sources that each model cites don’t overlap at all, so a one-channel content strategy no longer works.

Large-language-model search (ChatGPT, Google AI Overviews, Perplexity, Gemini, Claude, etc.) now sits at the very top of the prospect’s journey. Visibility here isn’t about blue-link rankings, it’s about whether the AI even knows your content exists. That hinges on the domains it likes to cite.

Take-home: the three biggest answer-engines lean on Wikipedia, Reddit and YouTube, but in very different proportions.

Ahrefs’ June 2025 analysis of 76.7 M AI answers found that only 14% of the top-50 cited sites appear in all three engines – meaning 86% are unique to one assistant (source).

Traditional publishers are fighting back by licensing data. News Corp’s pact with OpenAI alone is reportedly worth >$250 M over five years (source).

A Writesonic study of 1 M Google AIOs shows 40.6% of citations still come from Google’s top-10 results, so classic SEO helps – but it isn’t enough (source).

Meanwhile, Pew Research finds that when an AI summary appears, users click a traditional link only 8% of the time (vs 15% when no summary shows). Worse, just 1% click the links inside the summary (source).

If you’re a cosmetic or plastic-surgery practice, your authoritative “home turf” pages (before/after galleries, procedure explainers) still matter – but patient stories on Reddit, YouTube (or AnzoloMed.com) are what AI surfaces first. Being part of those conversations (ethically!) is no longer optional.