GEO study tests five optimization strategies on live pages in ChatGPT, Gemini and Grok
What this answers
- Do generative engine optimization (GEO) gains work on the live web?
- In a new arXiv benchmark called LIVE-GEO, the CORE rewriting strategy won the top spot 66.3%-76.3% of the time on three live AI engines, but the engines picked the same top product for only 48.8%-66.3% of the 80 fictional brands.
- Does the order of links change what ChatGPT recommends?
- In the LIVE-GEO test, ChatGPT picked the CORE-rewritten product first 73.8% of the time under one fixed link order and 93.8% under the opposite order; a randomized order gave 82.5%.
An arXiv preprint introduces LIVE-GEO, a benchmark that publishes generative-engine-optimization (GEO) rewrites of product pages as live webpages, then checks how AI search engines rank them. The authors rewrote product descriptions and reviews for 80 fictional brands across 18 product families with five existing strategies (CORE, C-SEO, E-GEO, AgenticGEO and Statistics Addition) and ran them through ChatGPT (GPT-6), Gemini 3.6 Thinking and Grok in Expert mode.
With prices frozen within ±0.2%, CORE won the top spot on the engines 66.3%-76.3% of the time, while the other strategies varied in relative rank. The three engines picked the same top product for only 48.8%-66.3% of brands. ChatGPT also reacted to the order of candidate URLs: CORE’s top-1 rate was 73.8% under one fixed order and 93.8% under the opposite, against 82.5% with the randomized order the benchmark uses. A numerical-faithfulness check flagged 86 of the 400 rewritten descriptions, including all 80 Statistics Addition descriptions, per the paper.
Why it matters: Most GEO tests hand the model text directly, so a benchmark that publishes rewritten pages on the live web and queries real engines is a step closer to what optimizers actually face, and it shows engines disagreeing on the winner.
Via arXiv GEO ↗
Posted to the wire November 3, 2026. Edited by Joe Balewski.