Methodology
Every report ships its own methodology block, and every figure traces to a named source. The standards below govern all of them.
Start with the research question
Each report defines the exact question, the unit of analysis, and what would count as evidence before drawing a conclusion.
Use a source hierarchy
Primary experiments, peer-reviewed papers, preprints, official documentation, and first-party datasets are preferred. Secondary reporting is used for discovery or context and labeled as such.
Preserve experimental boundaries
Model, version, language, prompt, retrieval setup, sample, and date are part of the finding. Results are not generalized beyond the protocol without evidence.
Separate mechanisms
Recall, retrieval, ranking, citation, recommendation, and downstream choice are measured as different stages rather than collapsed into one visibility score.
Disclose uncertainty
Preprint status, inaccessible full text, commercial conflicts, small samples, observational designs, and unresolved confounds remain visible in the article.
Review before publication
Each report passes source, factual, methods, adversarial-evidence, line-edit, and cold-read checks. Blocking findings return the draft to revision.
Follow the Institute’s research
For publication updates or editorial questions, contact the Search Institute directly.
Behind the Research
View all →Our mission and standards
Primary evidence, bounded claims, transparent methods, named authorship, and visible corrections.
Methodology
How the Institute handles sources, protocols, uncertainty, and review before publication.
About the Institute
Independent research on how search and AI systems retrieve, rank, cite, and recommend information.