Methodology

This page is the standing standard for everything published here. Each study also carries its own methodology section with the specifics for that piece of work. If a published item does not meet this page, that is a defect, and the correction policy below applies.

What every study must ship with

A study missing any of these six does not publish:

1. A reproducible methodology section. Query sets, sampling dates, engines and models tested, versions, and exclusions, in enough detail to rerun the collection.
2. A downloadable dataset. The actual series behind the study, published as CSV and linked from the study page. Corrections bump a version suffix; prior versions stay live.
3. Charts rendered from the exported series. The chart and the CSV are the same data by construction, so a reader can verify any figure against the file.
4. A named author.
5. A citation block with title, author, publisher, date, and URL.
6. A press-ready summary: one paragraph a journalist can lift verbatim.

Notes are shorter pieces between studies. They require items 1 through 4; the citation block and press summary are optional there. Sample sizes are smaller and turnaround is faster, under the same integrity rules.

Collection

Data comes from scheduled, automated sampling of search engines and AI assistants against fixed query sets. Each study's methodology section states which engines and models were sampled, on what schedule, over what window, and with what exclusions. Raw series land on disk before any chart is rendered; charts are generated from the exported CSV at build time, as static SVG.

Integrity rules

• Zero fabricated data. Every number traces to a collected series. No illustrative or estimated figures presented as findings.
• Methodology transparency beats favorable results. If a result is ambiguous or a sample is thin, the study says so.
• Findings that cut against the commercial interests of the publisher or its sibling projects publish anyway.
• Real people, real quotes, primary sources only.

Corrections

Corrections get a dated changelog entry on the study page. Datasets are versioned, never silently replaced: a corrected file publishes under a bumped version suffix, the prior version stays live, and the changelog links both. If a finding does not survive its correction, the study says that too.

Who does the work

Studies are authored under a named byline. Collection is automated; collection QA, analysis, and drafting are done in-house at The Discoverability Company, whose relationship to this publication is described on the about page.