30M-Prompt Analysis Proves Expert Quotes and Statistics Show Under 5% Correlation With Citations in LLMs

Conventional generative engine optimization (GEO) dictates that injecting expert quotes and statistics forces LLM citations.

However, analyzing 30 million prompt-and-response pairs proves these elements provide negligible algorithmic lift, reveals Michal Suski.

Testing relevant statistics against citation frequency shows it is one of the weakest correlating factors.

Across major LLMs, inserting data points or quote blocks yields less than a 5% correlation with retrieval inclusion.

For specific engines like Perplexity, that correlation drops to nearly zero.

The algorithm does not assign higher retrieval weight simply because a page contains quantified statements or dedicated quote formatting.