Why an intervention can work after retrieval yet fail across the whole system—and what credible testing requires.
Conditional effects assume an upstream success
A conditional effect asks what happens given that an earlier stage has already occurred. In the foundational GEO experiment, the source was already included among documents supplied to the generator. The test therefore asked whether modifying an in-context source increased its answer visibility.
This is a valid causal question, but it is conditional on retrieval and context inclusion. It does not estimate the probability that the source would be organically retrieved.
End-to-end effects allow the whole chain to change
An end-to-end test lets activation, retrieval, reranking, context, citation, and user outcomes respond to the intervention. A rewrite might improve citation after retrieval while reducing retrieval rank. If the upstream loss is larger, the total effect can be negative.
A positive conditional effect does not guarantee a positive total effect.
Before and after is not a causal design
If brand mention rises from 20% on Monday to 35% on Friday, the model version, index, competitor set, activation rate, query results, or generation randomness may have changed. The difference cannot automatically be credited to the treatment.
Meaningful estimates need repeated runs, a comparison condition, controlled prompts, timing, engines, corpus, context order, and competing documents where possible.
Competition creates interference
Answer space is limited. When one source gains attributed share, another may lose it. When competitors adopt the same tactic, the advantage can shrink. GEO is partly a competitive allocation problem rather than an isolated page-quality problem.
Questions about this topic
What is a conditional GEO effect?+
It measures an intervention after an earlier stage—such as retrieval or context inclusion—has already been satisfied.
What is an end-to-end GEO effect?+
It measures the final outcome while all upstream and downstream stages are allowed to change in response to the intervention.
Why is a simple before-and-after comparison weak?+
Platform drift, search-index changes, competitor activity, activation differences, and random generation can explain the change without the intervention.
What is interference in GEO?+
It is the competitive effect in which one source’s treatment changes other sources’ outcomes because answer space and attention are limited.
References
Sources are listed in APA 7 style. Preprints are identified as such and should not be treated as peer-reviewed findings unless separately published.
- Aggarwal, P., Murahari, V., Rajpurohit, T., Kalyan, A., & Narasimhan, K. (2024). GEO: Generative engine optimization. In Proceedings of the 30th ACM SIGKDD Conference on Knowledge Discovery and Data Mining (pp. 5–16). Association for Computing Machinery. https://doi.org/10.1145/3637528.3671900
- Martinez, O. (2026). Optimizing visibility in generative engines: A critical survey of generative engine optimization (2023–2026) [Preprint]. arXiv. https://doi.org/10.48550/arXiv.2607.14035
- Zhang, K., He, X., & Yao, J. (2026). From citation selection to citation absorption: A measurement framework for generative engine optimization across AI search platforms [Preprint]. arXiv. https://doi.org/10.48550/arXiv.2604.25707