Research Note: What Gets Cited First in AI Answer Engines?

A May 2026 study tested 252,000 paired trials across six large language models to examine which of two competing retrieved sources was cited first. The result matters because AI visibility is increasingly shaped after retrieval, not only by whether a page was found.

The strongest drivers

The researchers found topical relevance and list position were the strongest drivers of first-citation selection in their controlled setting. Explicit price information and recent timestamps also helped consistently. Completeness and trust cues produced smaller gains, while formatting-only changes had relatively little effect.

That does not create a universal ranking recipe. Controlled retrieval experiments cannot prove durable organic discoverability across every production platform. But they do reinforce an important distinction: getting into the retrieval set and winning attention inside the retrieval set are different problems.

The Frick-E Energy™ interpretation

Machine-Readable Authority™ treats citation as one layer inside a larger authority system. A business can be retrieved, cited, and still be commercially misunderstood. That is why the measurement must include entity accuracy, attribution fidelity, evidence quality, and decision usefulness rather than stopping at citation count.

The study also supports something painfully unglamorous: substance matters. Clear topical alignment, concrete facts, and extractable evidence give machines something usable. Changing the font, sprinkling buzzwords, and hoping the model feels spiritually moved is not a research strategy.

Source: Vishwakarma, Kumar, and Jamidar, “What Gets Cited: Competitive GEO in AI Answer Engines,” May 25, 2026, arXiv:2605.25517. Analysis by Heather Fricke, Frick-E Energy™. First published August 7, 2026.