What actually moves AI visibility — and what doesn't.

The table summarizes the measured impact of content optimization methods on visibility in generative engine answers, from the GEO-bench evaluation of 10,000 queries.[1] Relative improvements are versus unoptimized source content; exact gains vary by query domain and position metric.

GEO-bench: measured effect of nine optimization methods (Aggarwal et al., KDD 2024)
MethodWhat it meansMeasured effect
Cite sourcesAdd credible citations to claimsup to +40%
Quotation additionInclude quotes from relevant sources~+30–40%
Statistics additionReplace vague claims with numbers~+30–40%
Fluency optimizationRewrite for direct, readable prose+15–30%
Easy-to-understandSimplify language and structure+moderate
Authoritative toneConfident, expert phrasing+moderate, domain-dependent
Technical termsAdd domain terminology+small, domain-dependent
Unique wordsIncrease lexical diversitynegligible
Keyword stuffingClassic SEO keyword injectionineffective / negative

Three findings that shape strategy

1. Evidence beats keywords. The tactics that gamed traditional search are the ones that fail with generative engines. Language models reward content that reads like a source worth citing — attribution, numbers, quotes — not content engineered for crawlers.

2. GEO favors the underdog. The study found the largest visibility gains accrued to lower-ranked websites. GEO disproportionately helps smaller businesses compete with incumbents, because a well-evidenced page from rank 5 can be cited over a vague page from rank 1.[1]

3. Effects are domain-dependent. The best-performing method differed by query category — statistics helped most for factual domains, quotations for people-centric ones. This is why auditing before optimizing is not optional.

References

  1. Aggarwal, P., Murahari, V., Rajpurohit, T., Kalyan, A., Narasimhan, K., Deshpande, A. “GEO: Generative Engine Optimization.” Proceedings of the 30th ACM SIGKDD Conference on Knowledge Discovery and Data Mining (KDD 2024). Preprint: arxiv.org/abs/2311.09735
  2. GEO-bench: benchmark of 10,000 queries for evaluating content visibility in generative engines, released with the paper above (geo-optim.github.io).