Skip to main content
Original research

How AI assistants decide what to cite

A four-study research line testing which content signals actually predict whether AI platforms cite a website — and which signals don't, even when conventional wisdom says they should. Methodology, raw data, and what each finding means in practice.

Start here

Cross-study synthesis

What actually predicts AI citations

Six findings that held across every one of our studies, with practical implications for what to optimize. The fastest way to understand our full research line.

8 min read
Read the synthesis

The studies

Each study tests one specific question. Read in order if you want the full picture; jump to whichever matches your situation if you don't.

  1. Foundational study
    June 1, 2026

    The GEO citation study

    The question: Which content signals predict whether AI cites your site?

    61 sites · 17 industries · 3 phases

    Read the study
  2. Controlled follow-up
    June 2, 2026

    E-E-A-T & content type

    The question: Does the negative E-E-A-T effect hold when you control for what kind of page it is?

    40 URLs · 2×2 controlled design · methodology fix mid-study

    Read the study
  3. Commercial-outcome study
    June 2, 2026

    Ecommerce recommendation survival

    The question: Which brands survive the AI shopping journey all the way to a purchase recommendation?

    40 brands · 12 categories · 4 shopping-journey stages

    Read the study

Looking for "how do I do GEO?"

The studies on this page tell you what AI assistants actually reward and why. If you're looking for foundational explainers, pillar deep-dives, frameworks, and how-to guides instead, head to Learn GEO — that's where the practical guidance lives.