What counts as evidence

We specify the change and the expected difference before measuring. We also define what would count against the hypothesis, reducing the temptation to choose a convenient explanation afterward.

An accessible page is not necessarily discovered. An indexed page is not necessarily cited in an AI answer.

• Access Can the page's content be read?
• Discovery and indexing Has the engine found the page and included it in its searchable index?
• Citation Is the content used as a source in an answer?

When a citation is absent, we first investigate where this sequence stopped.

Where possible, comparison pages are published at the same time, holding other conditions constant. We record differences in length, structure, and discovery routes that could affect the result.

We repeat questions to distinguish individual answers from recurring patterns. Product responses and API responses are recorded separately.

Observations retain original responses, measurement dates, services, and conditions.

Interpretations describe what those observations support and which alternative explanations remain. An absence of citations alone does not establish that a content change was ineffective.

Conclusions may support or reject a hypothesis, show mixed results, or remain inconclusive. The direction of a conclusion is separate from the strength of its evidence.

Results from one topic or a new domain do not automatically generalize to every market or brand. We identify conditions that need replication and questions that remain open.

This document describes research principles. It is not a report establishing the effectiveness of a particular intervention.