← Knowledge LibraryChapter 14 · Causal Foundations

Experimental Design for GEOMeasurement Is Not Causality

How causal estimands, counterfactuals, and stage boundaries turn an observed GEO change into a testable treatment-effect question.

How causal estimands, counterfactuals, and stage boundaries turn an observed GEO change into a testable treatment-effect question.

An observed change is not a treatment effect

If unbranded visibility rises from 18% to 29% after a rewrite, the intervention is only one possible explanation. Engine updates, index refreshes, prompt variation, stochastic generation, competitor activity, earned media, seasonality, locale, and search activation can all move the same metric.

Observed change ≠ causal effect.

Define the causal estimand before running the test

The estimand specifies the treatment, unit, outcome, population, engine conditions, and time window. In plain language, it asks how a metric would differ under the intervention versus the counterfactual condition without it.

Causal target

Treatment effect = Outcome treated − Outcome counterfactual

Experimental design constructs a credible counterfactual

The same page cannot be simultaneously changed and unchanged under the exact same engine state. Randomized controls, matched comparisons, and quasi-experimental designs approximate this missing outcome by creating comparable untreated observations.

Conditional and end-to-end effects answer different questions

A fixed-context experiment estimates what happens after a page is already retrieved. An open-web experiment includes crawling, indexing, retrieval, reranking, context selection, and citation. Evidence that a rewrite improves post-retrieval prominence does not automatically prove organic discoverability, traffic, or revenue effects.

Frequently asked questions

Questions about this topic

Why does a before-and-after increase not establish causality?+

Time, platform drift, competitors, prompt variation, stochasticity, and other changes may explain the difference.

What is a causal estimand?+

The precisely defined treatment effect an experiment is designed to estimate for a unit, outcome, population, and condition.

What is the counterfactual?+

The outcome that would have occurred for the treated unit during the same conditions if treatment had not been applied.

What is a conditional GEO effect?+

An effect measured after a source has already entered retrieval or a fixed context.

Does a post-retrieval lift prove higher organic visibility?+

No. It does not test whether the content is crawled, retrieved, or selected from the open web.

Source notes

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.

  1. 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
  2. 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
  3. 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