ChatGPT, Google and Perplexity Rarely Cite the Same Sources, Study Finds

If you run a website or a business, you have probably heard about GEO — "generative engine optimization," the new art of getting AI chatbots to mention your brand instead of a competitor's. Marketing tools now sell single scores that claim to measure how visible you are across AI search. A new paper says that promise is shaky at best.

The researcher asked ChatGPT, Microsoft Copilot, Google and Perplexity the same 15 commercial questions on one day, then tracked which web pages each engine cited. That produced 589 citations pointing to 528 different URLs across 356 domains. The overlap was startlingly thin. On average, any two engines shared only about 0.8% of their cited links, and most pairs shared none at all. Looking only at the top five results per engine, there was exactly zero overlap across all 60 engine pairings. Nearly 97% of the cited pages showed up on just one engine. Separately, when the same engine was checked a day later, roughly two-thirds of the cited pages had changed.

The study is careful about what this does and does not mean. It does not prove that engine-free page scoring is worthless — a score calculated without querying a live AI can still measure genuine page quality or how well a page matches a topic. What it cannot do is predict your actual visibility, because that depends on whether a specific engine happens to surface you at all, and then whether it chooses to quote you. The author suggests reporting four separate numbers instead of one: how well your page fits the topic, how often you appear, how often you get quoted, and your final visible share.

For a small business owner, the practical takeaway is simple. There is no single "AI ranking" to optimize for, and no single dashboard that captures it. Checking one chatbot tells you almost nothing about the others, and yesterday's results may not hold today.