GEO is still a young discipline, but something interesting is already happening: researchers are trying to detect it.

A new paper called GEO-Flag: Detecting and Measuring GEO-Optimized Web Content introduces GEOFlagBench, a benchmark containing 3,200 webpages across 400 queries and eight GEO optimizer families. The researchers then developed a detection system and applied their pipeline to pages retrieved from Google Search and Gemini-grounded results.

Across 10,095 available pages, the system classified an estimated 8.90% as GEO-optimized. Among the smaller subset of pages carrying usable dateModified information, the reported rate reached 16.36% for pages marked as modified in 2026.

Those numbers need some caution. This does not mean that 8.9% of the entire web is GEO optimized, and the 16.36% figure comes from a much smaller subset of the dataset. Even the researchers found that existing detection approaches could confuse GEO intervention with signals related to how the content was authored.

The more interesting part may therefore be the existence of the detector itself.

SEO spent years moving through the same cycle: optimization techniques emerge, people exploit them, search engines learn to identify patterns, and the techniques become less effective or require more sophistication.

GEO could eventually develop its own version of that arms race.

We touched the more aggressive end of this problem in Reddit Spam for AEO Is Becoming a Real Experiment. If people deliberately manufacture signals because AI systems retrieve particular sources, detection becomes almost inevitable.

The long-term GEO advantage may therefore not come from making a page look optimized for an LLM. It may come from publishing information that remains useful even after optimization patterns become detectable.

That distinction could become important much sooner than expected.

Source: Chu et al., GEO-Flag: Detecting and Measuring GEO-Optimized Web Content, arXiv:2608.16824.
Read the GEO-Flag paper